Device health prediction method and computer-readable storage medium

Abstract

Disclosed in the present application are a device health prediction method and a computer-readable storage medium. The method comprises: determining first health data of a target device, wherein the first health data comprises battery health data of the target device within the current time period; and on the basis of the first health data, determining battery health prediction data of the target device within a future preset time period. The solution can efficiently and accurately predict the battery health condition of a target device within a future preset time period. Therefore, early warning can be further performed prior to a potential fault of a device, such that health management can be effectively implemented on the device, the occurrence of the device fault can be prevented, and economic losses caused by the device fault can be reduced, thereby reducing operation and maintenance costs and improving the operating efficiency.

Description

Equipment health prediction methods and computer-readable storage media

[0001] This application claims priority to Chinese Patent Application No. 202410918445.1, filed on July 9, 2024, entitled "Method for Predicting Equipment Health and Computer-Readable Storage Medium", the entire contents of which are incorporated herein by reference. Technical Field

[0002] This application belongs to the field of equipment health prediction technology, and in particular relates to an equipment health prediction method and a computer-readable storage medium. Background Technology

[0003] With the development of manufacturing and information technology, terminal devices are constantly evolving towards intelligence, efficiency, and diversified applications. Terminal devices can assist people in completing various processing tasks, making life more convenient. The proper functioning of terminal devices is essential for ensuring the successful completion of these tasks; if a terminal device is in poor condition, it will be difficult to complete the task, and may even lead to losses for the user.

[0004] Terminal devices typically use batteries with limited capacity, and the health of the batteries directly affects the health of the devices. Taking smart payment devices, including point-of-sale (POS) terminals, as the number of charge-discharge cycles increases, the battery performance gradually degrades, leading to a decline in the overall performance of the device. If the device continues to be used after its performance has degraded to a certain extent, it is very likely to malfunction during payments, resulting in a series of serious problems such as transaction interruptions, lost sales, customer dissatisfaction, and data loss, causing unavoidable economic losses to merchants.

[0005] In existing technologies, some solutions focus on the impact of battery performance on the health of terminal devices, but they usually only involve a qualitative evaluation of the current health status of the battery in the terminal device, rather than a real prediction of the health status of the terminal device. Therefore, these solutions still cannot effectively solve the problem of device failure caused by the degradation of terminal device performance. Summary of the Invention

[0006] This application provides a device health prediction method and a computer-readable storage medium, which can efficiently and accurately predict the battery health status of a target device within a preset period in the future. Therefore, it can provide early warnings before the device may fail, thus effectively realizing device health management, preventing the occurrence of failures, reducing economic losses caused by device failures, thereby saving operation and maintenance costs and improving work efficiency.

[0007] A first aspect of this application provides a device health prediction method, including:

[0008] Determine the first health data of the target device, wherein the first health data includes the battery health data of the target device in the current time period;

[0009] Based on the first health data, determine the battery health prediction data of the target device for a future preset period.

[0010] A second aspect of this application provides a device health prediction apparatus, comprising:

[0011] The determination module is used to determine the first health data of the target device, wherein the first health data includes the battery health data of the target device in the current time period;

[0012] The prediction module is used to determine the battery health data of the target device within a preset future period based on the first health data.

[0013] A third aspect of this application provides a terminal device, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the steps of the device health prediction method described above.

[0014] A fourth aspect of this application provides a computer-readable storage medium storing a computer program that, when executed by a processor, implements the steps of the above-described device health prediction method.

[0015] The device health prediction method provided in the first aspect of this application first determines first health data, including the battery health data of the target device in the current time period, and then determines the battery health prediction data of the target device in a future preset period based on the first health data. This scheme can efficiently and accurately predict the battery health status of the target device in the future preset period. Therefore, it can further provide early warning before the device may fail, thereby effectively realizing the health management of the device, preventing the occurrence of device failures, reducing economic losses caused by device failures, and thus saving operation and maintenance costs and improving work efficiency.

[0016] It is understood that the beneficial effects of the second to fourth aspects mentioned above can be found in the relevant descriptions in the first aspect mentioned above, and will not be repeated here. Attached Figure Description

[0017] To more clearly illustrate the technical solutions in the embodiments of this application, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0018] Figure 1 is a schematic flowchart of a device health prediction method provided in an embodiment of this application;

[0019] Figure 2a is a partial schematic diagram of the data list in the initial health dataset provided in one embodiment of this application;

[0020] Figure 2b is a schematic diagram of the data list obtained by performing a preliminary transformation on the data list in Figure 2a;

[0021] Figure 2c is a visualization of the battery health data of the first device after time series transformation;

[0022] Figure 2d is a partial schematic diagram of a data list of a device health dataset provided in an embodiment of this application;

[0023] Figure 3a is a schematic diagram of how battery health values ​​change with device age according to an embodiment of this application;

[0024] Figure 3b is a schematic diagram comparing the changes in battery health values ​​of devices with different usage frequencies as the device ages, according to an embodiment of this application.

[0025] Figure 3c is a schematic diagram comparing the changes in battery health values ​​of devices of different network types with device age according to an embodiment of this application;

[0026] Figure 3d is a schematic diagram comparing the changes in battery health values ​​of devices in different usage regions with device age, according to an embodiment of this application.

[0027] Figure 4a is a schematic diagram of the division of the training dataset and the validation dataset provided in an embodiment of this application;

[0028] Figure 4b is a schematic diagram of the prediction results of the equipment health prediction model provided in one embodiment of this application;

[0029] Figure 5 is a schematic diagram of a report settings interface provided in an embodiment of this application;

[0030] Figure 6 is a schematic diagram of the structure of a device health prediction device provided in an embodiment of this application;

[0031] Figure 7 is a schematic diagram of the structure of a terminal device provided in one embodiment of this application. Detailed Implementation

[0032] In the following description, specific details such as particular system architectures and techniques are set forth for illustrative purposes and not for limitation, in order to provide a thorough understanding of the embodiments of this application. However, those skilled in the art will understand that this application may also be implemented in other embodiments without these specific details. In other instances, detailed descriptions of well-known systems, apparatuses, circuits, and methods have been omitted so as not to obscure the description of this application with unnecessary detail.

[0033] It should be understood that, when used in this application specification and the appended claims, the term "comprising" indicates the presence of the described features, integrals, steps, operations, elements and / or components, but does not exclude the presence or addition of one or more other features, integrals, steps, operations, elements, components and / or a collection thereof.

[0034] It should also be understood that the term “and / or” as used in this application specification and the appended claims means any combination of one or more of the associated listed items and all possible combinations, and includes such combinations.

[0035] Furthermore, in the description of this application and the appended claims, the terms "first," "second," "third," etc., are used only to distinguish descriptions and should not be construed as indicating or implying relative importance.

[0036] References to "one embodiment" or "some embodiments" as described in this specification mean that one or more embodiments of this application include a specific feature, structure, or characteristic described in connection with that embodiment. Therefore, the phrases "in one embodiment," "in some embodiments," "in other embodiments," "in still other embodiments," etc., appearing in different parts of this specification do not necessarily refer to the same embodiment, but rather mean "one or more, but not all, embodiments," unless otherwise specifically emphasized. The terms "comprising," "including," "having," and variations thereof mean "including but not limited to," unless otherwise specifically emphasized.

[0037] As mentioned earlier, terminal devices typically utilize batteries with limited capacity. With each charge-discharge cycle, battery performance gradually degrades, significantly reducing the usability of the terminal device. Therefore, the health of the battery in a terminal device is closely related to the overall health of the device. For example, typically, for lithium batteries in terminal devices, when the maximum capacity decreases to 80% of the rated capacity, the battery is considered to have reached its failure threshold. Therefore, predicting battery health is crucial for predicting the health of the terminal device. However, existing battery health prediction schemes only involve a qualitative assessment of the current health status of the battery in the terminal device, rather than a true prediction of the terminal device's health status in advance. Therefore, they cannot effectively solve the problem of device failure caused by performance degradation in the terminal device.

[0038] This application provides a device health prediction method, which can be used to predict the health status of various terminal devices and the health level of terminal devices after a period of time. This facilitates the development of predictive maintenance strategies for terminal devices, effectively avoiding device failures caused by performance degradation. The terminal devices applicable to this device health prediction method include, but are not limited to: smartphones, POS machines, tablets, laptops, smartwatches, and smart home devices (such as smart door locks, smart monitoring devices, and smart speakers). For simplicity, the following description uses a POS machine as an example to illustrate the specific implementation of the device health prediction method.

[0039] As shown in Figure 1, the device health prediction method provided in this application embodiment includes the following steps S120 and S140:

[0040] Step S120: Determine the first health data of the target device. The first health data includes the battery health data of the target device in the current time period.

[0041] In this embodiment, the target device is a terminal device for which health prediction is desired. The number of target devices can be one or more. That is, health prediction management can be performed on a single terminal device, or batch prediction management can be performed on multiple terminal devices or even a group of terminal devices simultaneously. When there are multiple target devices, the first health data for each target device can be determined separately in this step.

[0042] In this embodiment, the first health data includes at least the battery health data of the target device in the current time period. Furthermore, in one example, the first health data may also include battery health data of the target device in at least one time period prior to the current time period. In another example, the first health data may also include associated feature data related to the battery health of the target device. The associated feature data may specifically be feature data of at least one feature dimension of the target device. These feature dimensions can be arbitrarily set according to actual needs, including but not limited to intrinsic device features and extrinsic device features. Intrinsic device features may include features related to the device's own performance, such as battery rated capacity, battery materials, and other battery attribute features. Extrinsic device features may include device usage characteristics such as device usage time, user usage frequency, device usage region, device network type, charging frequency, and charging method.

[0043] The time period can be a predefined unit of time close to the prediction time. This unit of time period can be 1 hour, 1 day, 1 week, etc. Taking a 1-day unit of time period as an example, the current time period can be the day of prediction, or it can be the previous day. If the current time period is the prediction day, the first health data can at least include the target device's battery health data for the prediction day. Alternatively, it can also include the target device's battery health data for the week or month preceding the prediction day. Of course, it can also include feature data from one or more of the aforementioned feature dimensions of the target device.

[0044] Battery health data can include one or more physical quantities that characterize the degree or condition of a battery's health. For example, battery health data can include the Battery Health Index (BHI), current capacity, original rated capacity, cycle count (the total number of charge-discharge cycles the battery has undergone), voltage levels of the battery under different conditions, battery internal resistance, charge-discharge rate, charge-discharge curve, and so on.

[0045] In one implementation, the battery health data of the target device includes the battery health value of the target device. Specifically, the battery health value can be the ratio of the battery's relative capacity to its original rated capacity, reflecting the battery's performance status and degree of aging. The battery health value can be calculated using any suitable method. Step S120 determines the first health data of the target device, including step S121.

[0046] Step S121: Determine the battery health value of the target device in the current time period based on the charging parameters of the target device in at least one charging cycle during the current time period. The charging parameters include charging current, charging time, and battery rated capacity.

[0047] Taking a day as an example, this step determines the battery health value of the target device in the current time period based on the charging parameters of the target device during at least one charging cycle of the day.

[0048] For example, the target device could be an Android phone. This step can use the native Android API to obtain low-level parameters such as real-time charging current, charging time, and battery rated capacity during multiple charging cycles within the same day, calculating the milliamp-hours (mAh) added to the phone during each charge (actual mAh change). Then, by dividing the actual mAh change during each charge by the percentage increase in battery capacity, the phone's current actual capacity (estimated capacity) can be obtained. Next, the ratio of the actual capacity to the battery's rated capacity is calculated and multiplied by 100% to obtain the battery health value for the current charge. For example, if the phone was charged once that day, the battery health value for that day can be obtained based on the calculation result of that charge. For cases where the phone is charged multiple times a day, the result of a single calculation may have some randomness. Therefore, multiple calculations can be performed, and the median of the most recent calculations can be used as the final battery health value to monitor and evaluate the device's battery life. For example, if the phone was charged five times that day, the above method can be used to calculate the battery health value corresponding to each charge, resulting in five battery health values. The median of these five battery health values ​​can then be used as the battery health value for the day.

[0049] Similarly, for methods that include battery health data of the target device in other time periods prior to the current time period in the first health data, the above calculation method can also be used to obtain the battery health values ​​for other time periods. For example, the battery health values ​​of the mobile phone to be predicted for each day of the cumulative month can be calculated and used as the first health data of the mobile phone to be predicted. Alternatively, the battery health values ​​of the mobile phone to be predicted for each day of the cumulative month of the cumulative month can be combined with the feature data of at least one of the above-mentioned feature dimensions of the mobile phone and used as the first health data of the mobile phone.

[0050] Step S140: Based on the first health data, determine the battery health prediction data of the target device within a future preset period.

[0051] The battery health prediction data for the target device determined in this step within a preset future period includes battery health prediction data for at least one point in time within that preset future period. The preset future period can be arbitrarily set according to actual needs, as long as it is a time period that has not yet arrived after the current period. For example, the preset future period could be the next week, next month, next two months, next three months, next six months, or next year, etc. Taking a preset future period of two months as an example, the battery health prediction data determined in this step can be battery health prediction data for any suitable period within the next two months. Taking a unit time period as one day and battery health data including battery health values ​​as an example, the battery health prediction data can include the battery health values ​​of the target device on any one or more days within the next two months. For example, this step can obtain the battery health values ​​of the target device for each day within the next two months.

[0052] In this step, various suitable methods can be used to determine the predicted battery health data of the target device within a predetermined future period based on the initial health data. Exemplarily, and not limitingly, a health prediction model can be pre-built. For example, the target device is a mobile phone running an Android system. A large amount of battery health data for this type of phone can be collected in advance, including the current battery health value of each phone, its battery health value over a past period, and related characteristic data (daily usage time, application usage, game and video playback, charging frequency, charging duration, whether fast charging is used, etc.). Then, based on historical data, the downward trend of the battery health value of each phone can be analyzed. The impact of these related characteristic data on the battery health value can also be analyzed. Afterwards, based on the data analysis results, a health prediction model that can predict the battery health data of the target device within a predetermined future period can be built, filtered, and trained. In one example, this health prediction model can be a traditional statistical model. For example, the health prediction model can be an autoregressive moving average model (SMA) or an autoregressive integrated moving average model (ARIMA) capable of modeling one-dimensional time series. The SMA model effectively reflects trend changes by using the average of observations from the past n time points as the predicted value for the next point, with relatively low computational cost. The ARIMA model combines autoregressive (AR), differencing (I, i.e., integrated), and moving average (MA) components, making it particularly suitable for non-stationary time series data. It transforms non-stationary time series into stationary ones for prediction through differencing. In another example, the health prediction model can also be a machine learning model, such as a deep learning model. In this step, the initial health data of the target device can be input into the built or trained health prediction model to obtain the model's predicted battery health data for the target device within a preset future period.

[0053] This is understandable, as this step yields battery health prediction data for the target device over a pre-defined future period. Therefore, it truly achieves accurate prediction of the target device's health. This allows for effective early warning before the target device's battery fails, giving users sufficient time for maintenance or upgrades. For example, it can help users or device administrators develop battery maintenance and replacement plans to avoid unexpected downtime; it can also predict battery replacement times, allowing for reasonable budget allocation and reducing emergency procurement costs; it can optimize device usage and charging strategies to extend battery life; it can take effective measures in advance to prevent data loss or device damage; planning based on predictions ensures the reliable operation of mission-critical equipment when needed, improving system reliability; it allows for measures to save or transfer critical data before the battery depletes, improving the security of data stored in the target device; and it can develop battery health management plans based on predictions, thereby extending the overall lifespan of the device.

[0054] The device health prediction method provided in this application first determines first health data, including the battery health data of the target device in the current time period, and then determines the battery health prediction data of the target device in a future preset period based on the first health data. This scheme can efficiently and accurately predict the battery health status of the target device in the future preset period. Therefore, it can further provide early warning before the device may fail, thereby effectively realizing the health management of the device, preventing the occurrence of device failures, reducing economic losses caused by device failures, and thus saving operation and maintenance costs and improving work efficiency.

[0055] In one implementation, step S140, which determines the battery health prediction data of the target device within a preset period based on the first health data, includes step S141.

[0056] Step S141: Input the first health data into the health prediction model to obtain battery health prediction data for the target device within a preset future period. The health prediction model is a deep learning model used for time series prediction.

[0057] Health prediction models can be various suitable deep learning models capable of time series forecasting. In one example, a health prediction model could be a Long Short-Term Memory Network (LSTM) model. This model, by introducing input gates, forget gates, and output gates, partially solves the gradient vanishing problem of recurrent neural networks, enabling fitting and prediction of time series data for each target device individually. In another example, a health prediction model could be a DeepAR model. This model belongs to the probabilistic prediction model based on an autoregressive recurrent neural network architecture, and can globally learn time series relationships and distribution characteristics from monitoring data. DeepAR models can easily incorporate other features of the predicted object into the calculation simultaneously. For devices with different feature data (e.g., devices from different regions, different models, different batches), the device health prediction results output by the DeepAR model can be different. That is, the DeepAR model can use feature data with different feature dimensions to predict the battery health of devices. The prediction target of the DeepAR model is the probability distribution of the sequence at each time step. Furthermore, the DeepAR model can also use cold-start prediction, meaning that the DeepAR model can perform new time series predictions using similar time series even with almost no historical data. For example, after training a DeepAR model, you can input only the target device's battery health data for the current time period (without needing to input other historical data), and the DeepAR model can predict the battery health data for a future period. Furthermore, DeepAR models can build unified prediction models for multiple time series, making them suitable for massive data scenarios. For instance, it can still achieve high computational efficiency and accurate prediction results for battery health prediction of a large number of devices.

[0058] In the above approach, utilizing a deep learning model designed for time series forecasting can improve the accuracy of device health prediction. Deep learning models can capture complex nonlinear relationships and long-term dependencies, providing high-precision prediction results. Furthermore, deep learning models can automatically learn and extract features from raw data, reducing the need for manual feature engineering. Properly trained deep learning models also possess good generalization ability, adapting to different data distributions and prediction tasks. In addition, deep learning models can handle large-scale datasets; their performance typically improves with increasing data volume. Moreover, deep learning models can be designed for real-time prediction, rapidly responding to newly collected data and updating predictions in real time. In summary, the above approach can improve the accuracy and efficiency of predicting the health of target devices.

[0059] In one implementation, the battery health data of the target device includes the battery health value of the target device. Step S141 inputs the first health data into the health prediction model to obtain the battery health prediction data of the target device within a preset future period, including steps S141a and S141b.

[0060] Step S141a: Input the first health data into the health prediction model and output the probability distribution of the predicted battery health values ​​of the target device at each time point within a preset future period. Step S141b: Determine the estimated battery health value of the target device at each time point based on the probability distribution of the predicted battery health values ​​at each time point.

[0061] Taking the DeepAR model as the health prediction model, a mobile phone as the target device, and the current time period as the prediction day as an example. In step S141a, at least the battery health value of the mobile phone on the prediction day can be input into the DeepAR model. The model can output the probability distribution of the predicted battery health value of the mobile phone for each day (each week or each month) in the next two months. In some preferred examples, for the case of predicting different target devices, at least one feature data of the mobile phone, such as the usage time, battery rated capacity, user device usage frequency, usage region, network type, charging frequency, and charging method, can be input together with the battery health value of the mobile phone on the prediction day into the DeepAR model, which can accurately obtain the probability distribution of the predicted battery health value of the mobile phone for each day in the next two months. In step S141b, the estimated value of the battery health value for that day can be determined based on the probability distribution (e.g., probability density function) of the predicted battery health value of the mobile phone for each day output by the model. For example, the predicted value corresponding to the maximum probability density value can be determined as the estimated value of the battery health value for that day. Alternatively, the predicted value corresponding to the median can also be determined as the estimated value of the battery health value for that day.

[0062] This approach provides more comprehensive predictive information by offering a probability distribution of battery health values ​​at each time point, rather than a single estimate. The probability distribution quantifies the uncertainty of the prediction, helping users understand its reliability. It can be used for risk assessment, for example, by identifying high-risk periods of declining battery health values ​​to take proactive measures; and it can provide data-driven decision support for equipment maintenance, replacement, and usage strategies. Furthermore, the model can be continuously updated based on new data to adapt to changes in battery health status.

[0063] It's understandable that in specific scenarios (such as predictive maintenance of a large number of target devices), probability distributions are more meaningful than single-point predictions. For example, in device health prediction, given the probability of a target device failing over a future period (the probability that the predicted health value is less than or equal to the predicted threshold for device failure), operations research optimization methods can be used to formulate optimal maintenance decisions. For instance, if the expected number of target devices to be managed is 1000, the prediction threshold could be 80%, and a suitable confidence threshold (e.g., 90%) could be pre-set, with the future period being, for example, the next two months. Exemplarily, based on the model output, the number of target devices with battery health values ​​below 80% and confidence levels above 90% within the next two months can be identified. Based on these counts, the device inventory or maintenance work for the following period can be planned. In other words, based on probability distributions, the focus can be on addressing target devices more likely to fail (with narrower and more concentrated confidence intervals), thus reducing maintenance costs and improving maintenance efficiency.

[0064] In one implementation, before inputting the first health data into the health prediction model in step S141, the method further includes steps S111 and S112.

[0065] Step S111: Determine the device health dataset. The device health dataset includes at least a time series of battery health data for each of the multiple first devices within a first historical period.

[0066] The first historical time can be arbitrarily set according to actual needs. For example, it can be a continuous time period before the current time period, such as 1 year, 2 years, etc. The device health dataset includes at least the battery health data of each of the multiple first devices within the first historical time period for multiple time periods, as well as the timestamps corresponding to each battery health data. These multiple time periods can be consecutive time periods within the first historical time period. Taking 900 first devices, the first historical time being 165 days before the prediction time, and each time period being 1 day as an example, the device health dataset determined in this step can include at least the battery health values ​​of 900 first devices (e.g., 900 mobile phones with Android system installed) for each day within these 165 days, as well as the timestamps (e.g., absolute dates or relative dates) corresponding to each battery health value. For example, these 900 mobile phones can be used as test devices in advance, and the real-time charging current, charging time, battery rated capacity, and other underlying parameters of each mobile phone in multiple charging cycles every day can be obtained through the native Android API. The battery health value of each mobile phone can be calculated daily using the method for calculating battery health values ​​in the aforementioned example. The multiple first devices can be different devices with the same internal and external characteristics. Alternatively, multiple first devices can be devices with different intrinsic and / or extrinsic characteristics. This allows for the differentiation of battery health data from different first devices with varying characteristics, broadening the applicability. In this example, the device health dataset can also include each first device's battery attributes, network type, usage time, usage frequency, network type, and usage region, etc.

[0067] In one implementation, step S111 determines the device health dataset, which may specifically include steps S111a and S111b.

[0068] Step S111a: Obtain the initial health dataset. The initial health dataset includes battery health data from at least multiple second devices at multiple points in time within a second historical period. The multiple second devices include multiple first devices, and the second historical period includes the first historical period.

[0069] The number of test devices included in the initial health dataset can be greater than or equal to the number of devices in the device health dataset. For example, the initial collected health dataset contains the battery health data of 3017 test devices (the second device), while the final device health dataset used for training only includes the health data of 886 test devices (the first device). Similarly, the time span corresponding to the battery health data in the initial health dataset can be greater than or equal to the time span corresponding to the battery health data in the device health dataset used for training. For example, the second historical time is the consecutive 200 days before the prediction time, and the first historical time is the consecutive 165 days before the prediction time; or, the second historical time and the first historical time can be the same. For example, both are consecutive 165 days from July 28, 2022 to January 6, 2023.

[0070] Taking the second historical period as an example, which is a continuous 165 days from July 28, 2022 to January 6, 2023, battery health data for 3017 test devices can be collected within this period and uploaded to the battery activity dataset in the device management platform. The battery activity dataset can be used to record daily battery consumption and battery health. For example, in the pre-operation and maintenance of smart payment devices, the battery activity dataset can be used to allow users to understand the device's activity level, monitor and evaluate usage frequency, the user's main device usage time, the main network type used, and other related characteristic data, and analyze the trend of battery health changes. The initial health dataset obtained after data collection can include at least the battery health data of 3017 test devices over 165 days, along with the timestamps corresponding to each data point. Alternatively, the initial health dataset can also include feature data for each test device across various characteristic dimensions. This feature data can be daily recorded feature data or feature data obtained by processing the 165-day recorded feature data to reflect the overall condition of the device over those 165 days.

[0071] Step S111b involves performing preprocessing operations on the initial health dataset to obtain a device health dataset consisting of multiple health data sequences. Each health data sequence corresponds to a first device, and each health data sequence includes a time series of battery health data for the corresponding first device within a first historical time period. The preprocessing operations include at least one of the following: outlier removal, missing data imputation, data format conversion, and data standardization.

[0072] As shown in Figure 2a, after collecting battery health data from 3017 test devices over 165 days, 120,860 rows of data were obtained in 8 columns. Each row includes not only the battery health value and time (time stamp) of a second device on a specific day, but also the device serial number, usage frequency, device age (i.e., the device's usage time, specifically in months), country (i.e., the region where the device is used), model, and network type. This data can be considered one-dimensional data. First, this one-dimensional data can be converted into two-dimensional data for easier preprocessing. The data for each device can be placed in one column, and the battery health values ​​of each device can be arranged in chronological order for convenient subsequent processing.

[0073] Then, outlier removal and missing data imputation can be performed. Specifically, abnormal battery health values ​​can be removed according to outlier removal rules. For example, for some inactive devices, they may not have been charged on some days within the 165 days, thus making it impossible to obtain battery health values. That is, the battery health values ​​of these devices at certain points in time are null. In this embodiment, all battery health data of devices with more than half of their values ​​being null can be removed to ensure the objectivity of the training process. Alternatively, for some unreasonable battery health values, such as a device showing a significant jump in its battery health value within the 165 days compared to the previous and subsequent days, these jump values ​​can be deleted. This avoids the adverse effects of such obviously abnormal battery health values ​​on training.

[0074] After the abnormal data removal operation is completed, missing data imputation can also be performed. For example, null values ​​in the battery health values ​​of the remaining devices after data removal can be filled. For instance, if a device has a null battery health value on a certain day, either the battery health value of the previous day or the average of the battery health values ​​of the following day can be used as the battery health value for that day. Finally, as shown in Figure 2b, there are 886 columns of data remaining, which represent the health data of 886 valid test devices (the first device). Each column also contains the associated characteristic data of the test devices.

[0075] To facilitate training, the processed health data after removal and imputation can be converted to a different data format. Specifically, the 886 columns of data in Figure 2b can be converted to a time series format. This allows the data to be visualized using third-party plotting libraries such as the matplotlib plugin, resulting in charts like those in Figure 2c. As shown in Figure 2c, for each first device, the battery health value fluctuates over time.

[0076] To more accurately predict device battery health, the health data of each device can be standardized. For example, when the device health dataset includes daily battery health values ​​and associated feature data of the device, various suitable standardization methods can be used to standardize the battery health values ​​and associated feature data of the device, which will then serve as feature parameters for model training. As shown in Figure 2d, the final device health dataset includes the standardized battery health value (ranging from 0 to 1) for each first device on each day, as well as the standardized values ​​of feature data for each device across five feature dimensions: country (region of use), model, network type, usage frequency, and device age (month). These standardized feature values ​​constitute the feature standardization array for each device. As shown in Figure 2d, each training data point in the device health dataset (for a health data sequence of each first device) includes the start time of the first historical period (July 25, 2022), the target value (i.e., the battery health value), and the feature standardization array.

[0077] The above approach can ensure the quality and consistency of model training data, enhance the model's generalization ability, improve model performance, reduce computational resource consumption, and guarantee the reliability and efficiency of model training.

[0078] Step S112: Train the health prediction model using the device health dataset until the preset termination condition is met.

[0079] This step involves using various suitable training methods to train the health prediction model. During training, model parameters are adjusted, and hyperparameter optimization is performed. Different evaluation metrics are considered, such as accuracy, recall, F1 score, and ROC-AUC. Preset termination conditions may include setting a certain performance threshold, training time limits, or training epoch limits. By iteratively training until the preset termination conditions are met, the model can be continuously optimized until optimal performance is achieved.

[0080] It's understandable that time-series data can capture the trend of device battery health status changing over time. Training models using historical data helps them learn the patterns of device health status changes over time, thereby improving the reliability of predictions. Setting preset termination conditions can effectively control the model training process, avoid resource waste, and ensure that training stops after the model reaches the expected performance. Model evaluation and adjustment during training help improve the model's generalization ability, ensuring that the model performs well on new and unseen data. Furthermore, this approach can be applied to different types of devices and different health indicators, has good scalability, and can be used in various scenarios. Therefore, it can provide scientific decision support for equipment operation and maintenance, reduce operational risks, and improve equipment utilization efficiency and economic benefits.

[0081] In one implementation, step S120 determines the first health data of the target device, including: determining the battery health data of the target device at least in the current time period and the feature data of at least one feature dimension of the target device as the first health data; wherein the feature dimension includes one or more of the following: battery attributes, network type, usage time, usage frequency, network type and usage region.

[0082] For example, before predicting the device health of a POS terminal, the battery health value of the POS terminal on the current day and in the preceding period can be calculated. Feature data such as the POS terminal's rated battery capacity, the main network type used by the user (data network 4G, 5G, or WiFi, etc.), usage time (e.g., cumulative months of use, the POS terminal has been used for 12 months), and usage frequency (e.g., number of times charged per week) can be obtained. These feature data and battery health values ​​can then be standardized using a method similar to step S111b to obtain the first health data.

[0083] Understandably, this approach considers not only the device's current battery health data but also feature data from various dimensions that influence battery health during device use. Comprehensive consideration of data from multiple dimensions provides a more holistic perspective, helping to predict battery health more accurately. Feature data such as device age and usage frequency are closely related to battery health and can significantly improve the accuracy of the prediction model. Different usage habits and environmental conditions have varying impacts on battery health; considering these factors allows for more personalized predictions. Furthermore, multi-dimensional feature analysis can reveal the interactions and influences between different factors, providing more information for a deeper understanding of battery health. In summary, comprehensively considering the device's current battery health data and other relevant features leads to more accurate predictions, providing strong support for device maintenance and user decision-making.

[0084] Analyzing battery health data curves from devices with different characteristics reveals that different devices exhibit varying battery health curve performance. For example, devices used more frequently experience faster degradation than those used less frequently; devices using mobile networks degrade faster than those using Wi-Fi. To identify the feature dimensions that significantly impact battery health, a data-driven approach can be employed to select at least one feature dimension (e.g., a key feature dimension) from a range of possibilities. The feature data of this key feature dimension from the test device can then be used as training data for the model. The device battery can be viewed as a black box, and statistical and machine learning methods can be used to establish a complex mapping relationship between the feature data of these key feature dimensions and battery health. This data-driven approach does not require in-depth knowledge of battery operating mechanisms and aging processes, nor does it require the establishment of specific physical models.

[0085] In one embodiment, the device health dataset also includes feature data for at least one feature dimension of each first device. The method further includes steps S101 to S103.

[0086] Step S101 involves acquiring battery health data from multiple third devices at multiple time points within a third historical period, as well as feature data from these third devices across various feature dimensions. The feature data for each feature dimension of the multiple third devices are not identical. The third device can be the aforementioned second device, or it can be a different device. The third historical period can be the second historical period, or it can be different from the second historical period. Feature dimension filtering can be performed using pre-collected data before acquiring the initial health dataset, or it can be performed after acquiring the initial health dataset, analyzing and filtering the data within the initial health dataset. The feature dimensions included in the statistics can include at least four: network type used, usage region, battery rated capacity, and usage time, and can also include other feature dimensions. For example, 20 feature dimensions can be selected for correlation statistical analysis.

[0087] Step S102: Analyze the feature data of each feature dimension of multiple third devices and the battery health data of multiple third devices at multiple time points in the third historical time period to obtain the correlation between each feature dimension and the battery health data.

[0088] For example, battery health data from 3000 test devices over a year can be pre-acquired. The characteristic data across 20 feature dimensions for these 3000 devices are not entirely identical. Analysis can then be performed on the characteristic data for each feature dimension, and correlation coefficients (such as Pearson correlation coefficient, Spearman rank correlation, or Kendall rank correlation) can be calculated between the characteristic data for each feature dimension and the battery health value. Hypothesis testing can be conducted to determine if the correlation is statistically significant. Based on the magnitude and significance of the correlation coefficients, the relationship between the characteristic data for each feature dimension and the battery health value can be interpreted. Visualization tools such as scatter plots and heatmaps can also be used to display the relationships between variables.

[0089] For usage time (e.g., device age), the trend of battery health values ​​changing with device age can be analyzed separately for one or more devices. As shown in Figure 3a, the battery health values ​​of multiple devices of the same model decreased with increasing device age. These devices experienced battery degradation at month 20, and by month 31, the median battery health had dropped to 91%. Furthermore, correlation coefficients between device health values ​​and device age can be obtained through data correlation statistical analysis.

[0090] For other feature dimensions, a controlled variable approach can be used to analyze the correlation between each feature dimension and battery health data. As shown in Figure 3b, the feature data of all tested devices except for usage frequency are the same. Significant differences exist in battery health values ​​for devices with different usage frequencies: devices charged 1-2 times per week only show battery degradation in the 22nd month, and the battery health is still 94% in the 31st month. However, devices charged more than 3 times per week show battery degradation as early as the 10th month, and the battery health is only 86% in the 31st month. As shown in Figure 3c, the feature data of all tested devices except for network type are the same. Significant differences also exist in battery health values ​​for devices with different network types: devices using 4G show battery degradation in the 11th month, and the battery health is 87% in the 31st month. Devices using WiFi show battery degradation in the 19th month, and the battery health is 86% in the 31st month. As shown in Figure 3d, the characteristic data of all the test devices involved in the analysis are the same except for the region of use. However, there are significant differences in battery health values ​​among devices from different regions: devices from China showed battery degradation in the 17th month, but still had 86% battery health in the 31st month. However, devices from Russia, due to the very cold climate, had a median battery health of only 80% in the 10th month, and only 71% in the 31st month.

[0091] Step S103: Based on the correlation, at least one feature dimension is selected from multiple feature dimensions. Each of the at least one feature dimension has a correlation with the battery health data that is greater than or equal to a correlation threshold.

[0092] The correlation threshold can be set arbitrarily according to actual needs. For example, the correlation between the feature data of each feature dimension and the battery health value can be represented by a correlation coefficient. In step S103, the correlation threshold can be a coefficient threshold of 0.7. After obtaining the correlation coefficient between the feature data of each feature dimension and the battery health value in step S102, in this step, feature dimensions with a correlation coefficient greater than or equal to 0.7 can be determined as the aforementioned key feature dimensions.

[0093] It is understandable that selecting features highly correlated with battery health data can improve the accuracy of predictive models. Furthermore, selecting the most relevant features reduces unnecessary features and simplifies the model structure. Highly correlated features are also more likely to maintain their predictive power on new data, thus improving the model's generalization ability. Eliminating features weakly correlated with battery health data reduces the risk of model overfitting and lowers the computational costs of model training and prediction. The selected features are easier to interpret, aiding in understanding the model's decision-making process. They are also likely to be more adaptable to different equipment and environmental conditions. Correlation analysis also helps identify key factors affecting battery health, providing direction for further research. By identifying features highly correlated with battery health, risks during equipment operation can be better managed and predicted. In summary, this approach can further improve the accuracy of equipment health prediction, providing a scientific basis for equipment maintenance and health management.

[0094] In one implementation, step S112 trains the health prediction model using the device health dataset, including the following steps:

[0095] Step S112a: Divide the device health dataset into a training dataset and a validation dataset. The training dataset includes battery health data of multiple first devices in a first historical period, and the validation dataset includes battery health data of multiple first devices in a second historical period, where the first period is earlier than the second period.

[0096] [Corrected according to Rule 91, 03.09.2025] As shown in Figure 4a, the preprocessed battery health data of 886 valid test devices over 165 days can be divided into a training dataset and a validation dataset. The first period is, for example, the first 105 days, and the second period can be the last 60 days. That is, the data of the first 105 days can be used as the training dataset (data represented by curve segment A in Figure 4a), and the data of the last 60 days can be used as the validation dataset (data represented by curve segment B in Figure 4a). This facilitates subsequent validation based on the data of these 60 days and uses the data of these 60 days to predict the battery health in the future.

[0097] Step S112b involves training a health prediction model using the training dataset, validation dataset, and a first loss function until a preset termination condition is met. The value of the first loss function is positively correlated with the difference between the predicted battery health data of each first device at each time point in the second period, output by the health prediction model, and the actual battery health data of that device at each time point in the validation dataset.

[0098] For example, a health prediction model is a DeepAR model. Preprocessed and segmented training and validation datasets (e.g., time series and key feature data containing battery health data from 886 valid test devices) can be input into a DeepAR model for training, ultimately resulting in a battery health prediction model applicable to devices in various scenarios (with different features).

[0099] In a specific training example, the training data batch size can be set to 20, the learning rate to start at 0.0005, and each training session to 200 epochs. Data from the first 60 days of the battery health data time series for each device can be used as the training context. Training will stop after 20 epochs if no progress is made, and the model with the lowest loss will be returned as the final model. To evaluate model accuracy, the root mean square error (RMSE) can be used as the evaluation metric, serving as the first loss function for training the health prediction model. The RMSE formula is:

[0100] In the formula, z represents the predicted battery health data for each first device at various time points in the second period (e.g., the last 60 days), as output by the health prediction model. i The first loss function represents the true value of the device's battery health data at various time points in the validation dataset, and n represents the number of samples, i.e., the number of devices in the first dataset. It's understandable that RMSE is suitable for evaluating situations where the estimation error is not significant. Training can end when the model reaches convergence by minimizing the first loss function mentioned above. For example, after training, the final RMSE result of the model is 0.06, indicating that the selected experimental data can accurately reflect the degree of battery degradation and can adapt to differences caused by inconsistencies in device characteristics.

[0101] [Corrected according to Rule 91, 03.09.2025] For example, after training, the battery health data of the target device (e.g., each first device as the target device) and the device's existing characteristic parameters, including country, model, network type, usage frequency, and device age (month), can be used to predict the battery health of the target device for the next 60 days. As shown in Figure 4b, curve C represents the true value in the validation dataset, curve D represents the predicted value (the predicted value corresponding to the median of the probability distribution), and the dashed box area X represents the 90% confidence interval. The figure shows two parts of the prediction results. The first part is from November 2022 to January 2023. It can be seen that, compared with the validation data, the 90% confidence interval of the predicted value covers the true value for most of the time, which solves the problem of the estimation model's lack of ability to fit extreme values. The second part is the predicted data of the battery health values ​​of each first device from February 2023 to March 2023, that is, the battery health degradation trend for the next two months was successfully predicted.

[0102] This approach can effectively improve the performance and reliability of health prediction models by making reasonable use of historical data, setting appropriate training and validation processes, and adopting a loss function that is positively correlated with the error.

[0103] In one implementation, the battery health prediction data includes predicted values ​​of the battery health of the target device at various time points within a preset future period. The method further includes the following steps:

[0104] Step S151: Determine whether the predicted value of the battery health value of the target device at at least one time point within a future preset period is less than the health value threshold.

[0105] Step S152: If yes, send a warning message and / or corresponding equipment maintenance suggestion message to the target device before a preset period in the future.

[0106] The health threshold can be arbitrarily set according to actual needs. For example, based on historical experience, devices with battery health below 80% significantly impact daily use, and users are generally advised to replace the battery or the device. Therefore, a health threshold of 80% can be set. Of course, multiple health thresholds corresponding to different risk levels can also be set. The preset future period is, for example, the next two months. In a specific example, the battery health value of each of the managed target devices can be predicted for the next month (e.g., the 30th day from the prediction date) and the next two months. It can also determine whether the battery health value is less than 80%. For each device, when the predicted battery health value for the next month or two months is identified as less than 80%, a warning can be issued one month in advance. For example, a pop-up notification can be sent to the app installed on the device, or the user can be reminded via email or in-app message to maintain or replace the device. Device maintenance suggestions can also be provided, such as suggesting closing unnecessary applications and turning off the device during idle periods.

[0107] In this approach, by predicting battery health values ​​and setting health thresholds, preventative maintenance can be implemented before battery performance degrades to unacceptable levels. Timely warnings and maintenance can reduce the rate of battery degradation, extend battery life, and decrease replacement frequency and costs. Furthermore, predicting battery health and taking appropriate measures ensures equipment continues to operate during critical tasks, improving equipment reliability and stability. The warning system can alert users before battery performance degrades to the point of affecting equipment operation, reducing unexpected downtime and production interruptions caused by battery issues. Moreover, by predicting battery health, companies can better plan maintenance resources, such as manpower and spare parts, achieving optimized resource allocation. It can also effectively reduce equipment failures caused by battery problems, improving user satisfaction and loyalty.

[0108] In one implementation, the method further includes the following steps:

[0109] Step S153: Provide a user interface and display health prediction information for the target device over a preset future period. The health prediction information includes battery health prediction data and / or the corresponding health risk level. Step S154: Update the health prediction information at a preset frequency.

[0110] In one example, the device management interface for operations and maintenance personnel can periodically push a list of battery health values ​​and / or health risk levels for each target device for the next month and / or two months. For example, a health value above 85% indicates low risk, a health value between 80% and 85% indicates medium risk, and a health value below 80% indicates high risk. This facilitates users in calculating the number of devices with future health risks and updating device inventory in a timely manner. Alternatively, the device management interface can display the number of medium-risk and high-risk devices in real time. It can also display other information affecting device availability, such as device age and remaining memory, along with their corresponding risk levels. For example, a device older than 24 months is high risk, a device older than 1-2 years is medium risk, and a device older than 1 year is low risk; remaining memory less than 3GB is high risk, remaining memory between 3GB and 8GB is medium risk, and remaining memory greater than 8GB is low risk. In another example, the above user interface can also correspond to the display interface of an app installed on the target device.

[0111] The preset frequency can be set arbitrarily according to actual needs. For example, it can be updated daily, weekly, or monthly. This allows users to be promptly informed of the target device's battery health prediction information, enabling them to monitor or maintain the device in advance.

[0112] By combining the user interface with regularly updated predictive information, users can easily obtain accurate device health information in a timely manner, making it easier to manage and maintain the health status of the device's battery.

[0113] As shown in Figure 5, a report settings interface can also be provided, through which the user can receive push settings information for various device health information. Push settings information may include: the notification method, the calculation frequency of health data (corresponding to the update frequency of health prediction information in step S154), and may also include different filtering conditions for different health information. For example, a device older than 24 months is considered a high-concern level, a device with less than 3GB of remaining memory is considered a high-concern level, a battery health value less than 85% in the next month is considered a medium-concern level, and a battery health value less than 85% in the next two months is considered a medium-concern level, etc.

[0114] This application also provides a device health prediction apparatus for executing the method steps described in the above method embodiments. This apparatus can be a virtual appliance within a terminal device, run by the terminal device's processor, or it can be the terminal device itself.

[0115] As shown in Figure 6, the device health prediction device 600 provided in this embodiment includes:

[0116] The determining module 610 is used to determine the first health data of the target device, wherein the first health data includes the battery health data of the target device in the current time period;

[0117] The prediction module 620 is used to determine the battery health data of the target device within a preset future period based on the first health data.

[0118] In one implementation, the prediction module 620 is specifically used for:

[0119] The first health data is input into the health prediction model to obtain the battery health prediction data of the target device in the future within a preset period; wherein, the health prediction model is a deep learning model used for time series prediction.

[0120] In one embodiment, the battery health data of the target device includes the battery health value of the target device, and the prediction module 620 includes:

[0121] The input / output unit is used to input the first health data into the health prediction model and output the probability distribution of the predicted battery health values ​​of the target device at various time points within a preset future period.

[0122] The first determining unit is used to determine the estimated value of the battery health value of the target device at a given time point based on the probability distribution of the predicted value of the battery health value at each time point.

[0123] In one embodiment, the device health prediction device 600 further includes:

[0124] The first data collection module is used to determine the device health dataset, wherein the device health dataset includes at least a time series of battery health data for each of a plurality of first devices within a first historical period.

[0125] The training module is used to train the health prediction model using the device health dataset until the preset termination conditions are met.

[0126] In one implementation, the training module includes:

[0127] A partitioning unit is used to divide the device health dataset into a training dataset and a validation dataset. The training dataset includes battery health data of multiple first devices in a first historical time period, and the validation dataset includes battery health data of multiple first devices in a second historical time period, where the first period is earlier than the second period.

[0128] The training unit is used to train the health prediction model using the training dataset, the validation dataset, and the first loss function until a preset termination condition is met. The value of the first loss function is positively correlated with the difference between the predicted value of the battery health data of each first device at each time point in the second period output by the health prediction model and the actual value of the battery health data of the device at each time point in the validation dataset.

[0129] In one implementation, the first data collection module includes:

[0130] A data collection unit is used to determine an initial health dataset, wherein the initial health dataset includes battery health data of at least a plurality of second devices at multiple time points within a second historical period, the plurality of second devices including a plurality of first devices, and the second historical period including a first historical period;

[0131] The preprocessing unit is used to perform preprocessing operations on the initial health dataset to obtain a device health dataset consisting of multiple health data sequences; wherein, each health data sequence corresponds to a first device, and each health data sequence includes a time series of battery health data of the corresponding first device within a first historical time period; the preprocessing operations include at least one of the following operations: outlier data removal, missing data imputation, data format conversion, and data standardization.

[0132] In one implementation, the determining module 610 includes:

[0133] The second determining unit is used to determine the battery health data of the target device at least in the current time period and the feature data of at least one feature dimension of the target device as the first health data; wherein the feature dimension includes one or more of the following: battery attributes, network type, usage time, usage frequency, network type and usage region.

[0134] In one embodiment, the device health dataset also includes feature data of at least one feature dimension for each first device, and the device health prediction device 600 further includes:

[0135] The second data collection module is used to determine the battery health data of multiple third devices at multiple time points within a third historical period and the feature data of multiple third devices in multiple feature dimensions, wherein the feature data of each feature dimension of the multiple third devices are not completely identical.

[0136] The analysis module is used to analyze the feature data of each feature dimension of multiple third devices and the battery health data of multiple third devices at multiple time points in the third historical time period, and to obtain the correlation between each feature dimension and the battery health data.

[0137] The filtering module is used to filter at least one feature dimension from multiple feature dimensions based on correlation, wherein the correlation between each of the at least one feature dimension and the battery health data is greater than or equal to a correlation threshold.

[0138] In one embodiment, the device health prediction device 600 further includes:

[0139] A module is provided to provide a user interface and display health prediction information of the target device in a future preset period on the user interface;

[0140] The update module is used to update health prediction information at a preset frequency; wherein, the health prediction information includes battery health prediction data and / or the health risk level corresponding to the battery health prediction data.

[0141] In one embodiment, the battery health prediction data includes predicted values ​​of the battery health of the target device at various time points within a preset future period, and the device health prediction device 600 further includes:

[0142] The judgment module is used to determine whether the predicted value of the battery health value of the target device at at least one time point within a preset future period is less than the health value threshold.

[0143] The sending module is used to send a warning message and / or corresponding equipment maintenance suggestion message to the target device before a preset period in the future, if the condition is met.

[0144] As shown in FIG7, this application embodiment also provides a terminal device 700, including: at least one processor 710 (only one processor is shown in FIG7), a memory 720, and a computer program 730 stored in the memory 720 and executable on at least one processor 710. When the processor 710 executes the computer program 730, it implements the steps of the above-mentioned device health prediction method.

[0145] Terminal devices may include, but are not limited to, processors and memory. Figure 5 is merely an example of a terminal device and does not constitute a limitation on the terminal device. It may include more or fewer components than shown, or combine certain components, or use different components. The processor may be a Central Processing Unit (CPU), or it may be other general-purpose processors, digital signal processors (DSPs), application-specific integrated circuits (ASICs), field-programmable gate arrays (FPGAs), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. A general-purpose processor may be a microprocessor or any conventional processor.

[0146] It should be noted that the information interaction and execution process between the above-mentioned devices / modules are based on the same concept as the method embodiments of this application. For details on their specific functions and technical effects, please refer to the method embodiments section, and they will not be repeated here.

[0147] Those skilled in the art will understand that, for the sake of convenience and brevity, the above-described division of functional modules is merely an example. In practical applications, the functions described above can be assigned to different functional modules as needed, that is, the internal structure of the device can be divided into different functional modules to complete all or part of the functions described above. The functional modules in the embodiments can be integrated into one processing module, or each module can exist physically separately, or two or more modules can be integrated into one module. The integrated modules can be implemented in hardware or as software functional modules. Furthermore, the specific names of the functional modules are only for easy differentiation and are not intended to limit the scope of protection of this application. The specific working process of the modules in the above system can be referred to the corresponding process in the foregoing method embodiments, and will not be repeated here.

[0148] This application also provides a computer-readable storage medium storing a computer program, which, when executed by a processor, can implement the steps in the above-described device health prediction method.

[0149] This application provides a computer program product that, when run on a terminal device, enables the terminal device to perform the steps in the above-described device health prediction method.

[0150] The above-described embodiments are only used to illustrate the technical solutions of this application, and are not intended to limit them. Although this application has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of this application, and should all be included within the protection scope of this application.

Claims

1. A device health prediction method, comprising: The method includes: Determine the first health data of the target device, wherein the first health data includes the battery health data of the target device in the current time period; Based on the first health data, predict the battery health of the target device within a predetermined future period.

2. The device health prediction method of claim 1, wherein, The step of determining the battery health prediction data of the target device within a preset future period based on the first health data includes: The first health data is input into the health prediction model to obtain the battery health prediction data of the target device in the future preset period; wherein, the health prediction model is a deep learning model for time series prediction.

3. The device health prediction method of claim 2, wherein, The battery health data of the target device includes the battery health value of the target device. The step of inputting the first health data into the health prediction model to obtain the battery health prediction data of the target device for a future preset period includes: The first health data is input into the health prediction model, and the probability distribution of the predicted battery health value of the target device at each time point in the future preset period is output. Based on the probability distribution of the predicted battery health value at each time point, the estimated value of the battery health value of the target device at each time point is determined.

4. The device health prediction method of claim 2, wherein, Before inputting the first health data into the health prediction model, the method further includes: Determine a device health dataset, wherein the device health dataset includes at least a time series of battery health data for each of a plurality of first devices within a first historical period; The health prediction model is trained using the device health dataset until a preset termination condition is met.

5. The device health prediction method of claim 4, wherein, The step of training the health prediction model using the device health dataset includes: The device health dataset is divided into a training dataset and a validation dataset. The training dataset includes battery health data of the plurality of first devices in a first period of the first historical time. The validation dataset includes battery health data of the plurality of first devices in a second period of the first historical time. The first period is earlier than the second period. The health prediction model is trained using the training dataset, the validation dataset, and the first loss function until the preset termination condition is met. The value of the first loss function is positively correlated with the difference between the predicted value of the battery health data of each first device at each time point in the second period output by the health prediction model and the actual value of the battery health data of the device at each time point in the validation dataset.

6. The device health prediction method of claim 4, wherein, The determination of the device health dataset includes: Determine an initial health dataset, wherein the initial health dataset includes at least battery health data of multiple second devices at multiple time points within a second historical period, the multiple second devices including the multiple first devices, and the second historical period including the first historical period; Preprocessing is performed on the initial health dataset to obtain the device health dataset consisting of multiple health data sequences; Each health data sequence corresponds to a first device, and each health data sequence includes a time series of battery health data of the corresponding first device within the first historical time period; the preprocessing operation includes at least one of the following operations: abnormal data removal operation, missing data imputation operation, data format conversion operation, and data standardization operation.

7. The device health prediction method of any one of claims 1 to 6, wherein, The first health data for determining the target device includes: The battery health data of the target device at least in the current time period and the feature data of at least one feature dimension of the target device are determined as the first health data; The feature dimensions include one or more of the following: battery attributes, network type, usage time, usage frequency, network type, and usage region.

8. The device health prediction method of claim 7, wherein, The device health dataset also includes feature data for each of the at least one feature dimension of the first device, and the method further includes: Determine battery health data of multiple third devices at multiple time points within a third historical period and feature data of multiple feature dimensions of the multiple third devices, wherein the feature data of each feature dimension of the multiple third devices are not completely identical; The feature data of each feature dimension of the plurality of third devices and the battery health data of the plurality of third devices at multiple time points in the third historical period are analyzed to obtain the correlation between each feature dimension and the battery health data; Based on the correlation, at least one feature dimension is selected from the multiple feature dimensions, wherein the correlation between each of the at least one feature dimension and the battery health data is greater than or equal to a correlation threshold.

9. The device health prediction method of any one of claims 1 to 6, wherein, The battery health prediction data includes predicted values ​​of the battery health of the target device at various time points within a preset future period, and the method further includes: Determine whether the predicted value of the battery health value of the target device at at least one time point within the future preset period is less than a health value threshold. If so, a warning message and / or corresponding equipment maintenance suggestion message will be sent to the target device before the preset future period.

10. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores a computer program that, when executed by a processor, implements the steps of the device health prediction method as described in any one of claims 1 to 9.

Images