Wearable device cross-platform compatibility test method based on dynamic adaptation

Abstract

The invention discloses a wearable device cross-platform compatibility test method based on dynamic adaptation, and relates to the technical field of intelligent wearable devices, and the method comprises the following steps: collecting interaction feature data generated in the interaction process of a wearable device and a platform through a plurality of different platforms, comprising an equipment response time delay characteristic, a signal transmission characteristic and a user interface response characteristic; and respectively constructing a platform feature space and an equipment behavior space based on the interaction feature data, generating a mapping network by using a graph neural network to form a mapping relationship between equipment and platform interaction features, and predicting a possible compatibility problem between the wearable equipment and a new platform. According to the invention, the problem of low efficiency of the traditional cross-platform compatibility test is solved.

Description

Technical Field

[0001] The present invention relates to the technical field of smart wearable devices, and more specifically, to a cross-platform compatibility testing method for wearable devices based on dynamic adaptation. Background Art

[0002] As an emerging type of smart device, wearable devices have gradually become an important part of daily life, such as smart bracelets, smart watches, smart glasses, etc. These devices usually need to interact with various different smart terminal platforms (such as smartphones, tablets, personal computers, etc.). However, due to significant differences in hardware parameters, communication protocols, operating system versions, etc. among different platforms, there are many problems in the cross-platform compatibility of wearable devices, such as poor connection stability, high response latency, abnormal interactions, etc. Traditional compatibility testing methods usually require a large amount of tedious manual testing and are difficult to actively predict and discover potential compatibility problems. Therefore, there is an urgent need for an efficient and intelligent cross-platform compatibility testing method to discover and solve the above problems in advance. Summary of the Invention

[0003] The technical problem to be solved by the present invention is to provide a cross-platform compatibility testing method for wearable devices based on dynamic adaptation to solve the problems mentioned in the background art.

[0004] To achieve the above object, the present invention adopts the following technical solutions:

[0005] A cross-platform compatibility testing method for wearable devices based on dynamic adaptation, comprising:

[0006] Collecting interaction feature data generated during the interaction between the wearable device and the platform through multiple different platforms, where the interaction feature data includes any one or more of the following: device response delay feature, signal transmission feature, user interface response feature;

[0007] Based on the interaction feature data, respectively constructing a platform feature space and a device behavior space, and using a graph neural network to generate a mapping network to form a mapping relationship between the interaction features of the device and the platform;

[0008] When a new platform to be tested appears, first collect the interaction feature data of the new platform and map it in the mapping network, and predict possible compatibility problems between the wearable device and the new platform through a feature propagation algorithm.

[0009] Optionally, when constructing the mapping network, an autoencoder is used to establish a normal mode of interaction features of the wearable device on each platform;

[0010] When the deviation between the actual interaction feature data and the normal mode of interaction features exceeds a predetermined threshold, a reverse reinforcement feedback mechanism is triggered. Using the magnitude of the deviation as a reverse reward signal, it automatically guides the wearable device to actively generate and deeply detect more potentially abnormal interaction scenarios.

[0011] Optionally, the reverse reward signal r generated by the reverse reinforcement feedback mechanism satisfies:

[0012] r = -λ||x real -x norm ||2;

[0013] In the formula, λ is a proportionality coefficient, x real is the actual interaction feature data vector, and x norm is the interaction feature data vector of the normal mode.

[0014] Optionally, a virtual platform simulation module is set inside the wearable device. Using the Monte Carlo random sampling algorithm combined with the key feature parameters in the platform feature space, it automatically generates virtual platform scenario combinations and actively simulates the interaction process between the device and the platform based on the virtual platform scenario combinations, so as to identify potential compatibility anomalies in advance.

[0015] Optionally, when the virtual platform simulation module generates virtual platform scenario combinations, the sampling distribution P of the key feature parameters F of the platform feature space satisfies:

[0016]

[0017] In the formula, β is a coefficient controlling the degree of randomness, E(F) is the abnormal tendency energy function of the feature parameters, and Z is the normalization factor.

[0018] Optionally, an abnormal feature network is set inside the wearable device to automatically record the abnormal feature data that appears during each interaction process between the device and the platform, forming an abnormal feature memory bank;

[0019] The abnormal feature network is trained using a random forest classifier based on the abnormal feature memory bank to identify newly emerging abnormal feature data and automatically match the known abnormal data patterns stored in the abnormal feature memory bank.

[0020] Optionally, after the abnormal feature network identifies the abnormal feature data, it automatically generates corresponding repair instructions based on the abnormal feature memory bank. The repair instructions include any one or more of the following: Bluetooth protocol parameter optimization, data sending rate adjustment, and interface interaction mode switching.

[0021] Optionally, the acquisition frequency of the interaction feature data is adaptively and positively correlated with the frequency of anomalies detected during the device interaction process. When the frequency of anomalies during the device interaction process increases, the acquisition frequency increases; when the frequency of anomalies decreases, the acquisition frequency decreases.

[0022] Optionally, there is a real-time data feedback channel between the virtual platform simulation module and the reverse reinforcement feedback mechanism. Through the real-time data feedback channel, the abnormal data found during the virtual simulation process is real-time fed back to the reverse reinforcement feedback mechanism to update the interaction feature data vector of the normal mode.

[0023] Optionally, the abnormal data patterns in the abnormal feature memory bank are regularly updated online according to the newly added abnormal feature data, and the exponential weighted moving average algorithm is used to determine the weights of the abnormal feature patterns. The weights of the abnormal feature patterns that have appeared recently are greater than the weights of the abnormal feature patterns that have appeared historically.

[0024] The advantages of the present invention over the prior art are as follows: The present invention proposes a cross-platform compatibility testing method for wearable devices based on dynamic adaptation. By using a graph neural network to generate a mapping network, it can efficiently and accurately establish the interaction feature mapping relationship between the device and the platform. When a new platform appears, the feature propagation algorithm can be used to quickly predict the possible compatibility problems between the device and the platform, effectively solving the problems of long time-consuming, low efficiency, and difficulty in actively discovering anomalies in the traditional manual testing method. The present invention further introduces a reverse reinforcement feedback mechanism. When there is a large deviation between the actual interaction feature data and the normal mode, it automatically triggers in-depth mining of abnormal scenarios, effectively improving the initiative and coverage of anomaly detection. At the same time, through the built-in virtual platform simulation module, a large number of virtual test scenarios are actively generated using the Monte Carlo random sampling method to discover possible compatibility problems in advance, and the abnormal features are automatically memorized and actively repaired through the abnormal feature network, significantly improving the adaptive ability of the wearable device to different platforms and reducing the manual maintenance cost and anomaly repair time. Description of the Drawings

[0025] Figure 1 is the overall flowchart of the method of the present invention;

[0026] Figure 2 is the flowchart of the data acquisition stage of the present invention;

[0027] Figure 3 is the flowchart of the feature space construction and mapping network generation of the present invention;

[0028] Figure 4 is the flowchart of the new platform testing and anomaly detection of the present invention;

[0029] Figure 5 is the flowchart of the reverse reinforcement feedback and virtual simulation of the present invention;

[0030] Figure 6 This is the flowchart of exception handling and self - repair of the present invention. Detailed implementation manners

[0031] The following describes the detailed implementation manners of the present invention with reference to the accompanying drawings.

[0032] In today's rapidly developing wearable device market, ensuring that devices can achieve seamless compatibility on different platforms is an important issue. The present invention proposes a cross - platform compatibility testing method for wearable devices based on dynamic adaptation, which systematically solves compatibility problems and improves the adaptability of devices through a series of innovative technical means. The implementation process of this method will be described in detail below.

[0033] As Figure 1 shown in the overall flowchart of the present invention, it includes the following steps:

[0034] Collect interaction feature data generated during the interaction between the wearable device and the platform through multiple different platforms. The interaction feature data includes any one or more of the following: device response delay feature, signal transmission feature, user interface response feature;

[0035] Based on the interaction feature data, construct a platform feature space and a device behavior space respectively, and use a graph neural network to generate a mapping network to form a mapping relationship between the interaction features of the device and the platform;

[0036] When a new platform to be tested appears, first collect the interaction feature data of the new platform and map it in the mapping network, and predict possible compatibility problems between the wearable device and the new platform through a feature propagation algorithm.

[0037] As Figure 2 shown, the present invention first collects interaction feature data generated during the interaction between the wearable device and the platform through multiple different platforms. These data cover multiple aspects of the interaction between the device and the platform, such as device response delay feature, signal transmission feature, and user interface response feature.

[0038] Specifically:

[0039] The device response delay feature can be the time length for the device to give a response after receiving a platform instruction;

[0040] The signal transmission feature can include the data transmission rate and the packet loss rate or stability during the transmission process;

[0041] The user interface response feature can be the delay time or smoothness of the interface feedback after the user clicks on the interface.

[0042] By collecting this data on different platforms, the performance characteristics of wearable devices in various environments can be comprehensively captured.

[0043] Based on the collected interaction feature data, the present invention constructs a platform feature space and a device behavior space respectively. The platform feature space is a set of feature vectors composed of the interaction feature data of each platform, used to characterize the characteristics of different platforms; while the device behavior space is a set of behavior patterns of the wearable device on these platforms, reflecting the reaction modes of the device in different environments.

[0044] To make these spaces more representative, the data can be processed through feature extraction and dimensionality reduction techniques (such as principal component analysis or t-SNE) to extract key information and reduce the computational complexity. Then, a mapping network is generated using a graph neural network to form the mapping relationship between the device and the platform interaction features.

[0045] As Figure 3 shown, by treating the platform and the device as nodes in a graph and the interaction features as the information of the edges connecting these nodes, the graph neural network can effectively learn the complex dependence relationship between the two. For example, the signal transmission rate of a certain platform may affect the response delay of the device, and this relationship can be learned and represented through the nodes and edges of the graph neural network.

[0046] As Figure 4 shown, when a new platform needs to be tested, the present invention will first collect the interaction feature data of the new platform and input it into the constructed mapping network for mapping. Through the feature propagation algorithm, potential compatibility problems between the wearable device and the new platform can be predicted.

[0047] The working principle of the feature propagation algorithm is to propagate the feature information of the new platform in the graph and infer the behavior performance of the device on the new platform based on the existing mapping relationship. For example, if the signal transmission characteristics of the new platform are similar to those of a known platform, the algorithm can predict the response of the device under similar conditions through propagation calculation, so as to identify potential incompatibility points in advance. This method can discover problems before actual deployment and greatly improve the test efficiency.

[0048] In the process of constructing the mapping network, the present invention also uses an autoencoder to establish the normal mode of the interaction features of the wearable device on each platform. The autoencoder compresses and decodes the interaction feature data to learn the internal structure of the data, thereby generating a feature distribution reflecting the normal interaction state.

[0049] For example, the normal mode may include the response time range of the device in a stable network environment or the typical rate of signal transmission. When the deviation of the actually collected interaction feature data from this normal mode exceeds a predetermined threshold, a reverse reinforcement feedback mechanism is triggered. This deviation can be measured by calculating the Euclidean distance between the actual feature vector and the normal feature vector, and the threshold can be set according to the statistical distribution of historical data, such as taking the upper limit of the 95% confidence interval of the normal deviation.

[0050] As Figure 5 shown, the reverse reinforcement feedback mechanism uses the magnitude of the deviation as the reverse reward signal, automatically guiding the wearable device to actively generate and detect more potentially abnormal interaction scenarios. For example, when the deviation is large, the device may attempt to adjust transmission parameters or switch interaction modes to explore potential anomalies and verify their impacts.

[0051] The calculation formula for the reverse reward signal γ generated by the reverse reinforcement feedback mechanism is:

[0052] r = -λ||x real -x norm ||2;

[0053] where λ is a proportionality coefficient used to adjust the amplitude of the reward signal and can be adjusted according to the specific application scenario, such as set to 0.1 or 1; X real represents the vector of the actually collected interaction feature data, and x norm represents the vector of the interaction feature data of the normal mode, and ||·||2 represents the L2 norm used to quantify the distance between the two.

[0054] To further improve the coverage of compatibility testing, the present invention sets up a virtual platform simulation module inside the wearable device. This module uses the Monte Carlo random sampling algorithm, combined with the key feature parameters in the platform feature space, to automatically generate virtual platform scenario combinations. These key feature parameters may be the specific attributes of the platform, such as the operating system version, hardware performance metrics, or communication protocol type. By randomly sampling these parameters, a variety of possible platform environments can be simulated.

[0055] For example, the Monte Carlo algorithm may generate a scenario where the operating system version is older and the hardware performance is lower to test the performance of the device under such conditions. During the sampling process, the sampling distribution of the key feature parameter F satisfies:

[0056]

[0057] Among them, β is a coefficient that controls the degree of randomness of parameters. The larger the value, the lower the randomness. Usually, it can be set to 0.5 or 1 according to test requirements; E(F) is the abnormal tendency energy function of characteristic parameters, which is used to measure the possibility of a certain parameter combination causing an abnormality and can be defined through historical data or expert experience; Z is a normalization factor to ensure that the sum of the probability distribution is 1.

[0058] Based on these virtual platform scenarios, the module actively simulates the interaction process between the device and the platform, so as to identify potential compatibility anomalies in advance, such as connection interruptions or response delays that may occur under certain extreme conditions.

[0059] In addition, the present invention also sets up an abnormal feature network inside the wearable device, which is used to automatically record the abnormal feature data that appears during each interaction process between the device and the platform, forming an abnormal feature memory bank. These abnormal feature data may be the specific values that cause problems, such as too high response delay or unstable signal transmission rate.

[0060] As Figure 6 shown, the abnormal feature network is trained using a random forest classifier based on the data in the memory bank to identify newly emerging abnormal features and match known abnormal patterns. The random forest can efficiently classify abnormal types by constructing multiple decision trees and integrating their results. For example, the training data may include historical response delay anomalies and their corresponding scenarios, and the classifier determines the abnormal category of new data based on this. When an abnormality is identified, the network will automatically generate corresponding repair instructions based on the abnormal feature memory bank. These repair instructions may include optimizing Bluetooth protocol parameters to enhance connection stability, adjusting the data sending rate to match the processing capacity of the platform, or switching the interface interaction mode to adapt to the user interface requirements of the platform.

[0061] In order to capture abnormal situations more timely, the present invention sets the acquisition frequency of device interaction feature data to be adaptively adjusted, making it positively correlated with the detected abnormal occurrence frequency. When the device senses an increase in the abnormal frequency, such as multiple connection interruptions occurring within a short period of time, the acquisition frequency will automatically increase to record interaction data more densely; conversely, when the abnormal frequency decreases, the acquisition frequency will be correspondingly reduced to save computing and storage resources. This dynamic adjustment mechanism ensures obtaining more detailed information at critical moments while optimizing resource usage during normal operation.

[0062] Between the virtual platform simulation module and the reverse reinforcement feedback mechanism, the present invention also designs a real-time data feedback channel. Through this channel, the abnormal data found during the virtual simulation process can be transmitted to the reverse reinforcement feedback mechanism in real time for updating the interactive feature data vector of the normal mode. For example, if it is found during the simulation that the device response is abnormal in a certain platform scenario, this information will be immediately fed back and the definition of the normal mode will be adjusted to make it closer to the actual operating state. This real-time interaction enhances the adaptability of the system.

[0063] Finally, the abnormal data patterns in the abnormal feature memory bank will be updated online regularly according to the newly added abnormal feature data, and the exponential weighted moving average algorithm is used to determine the weights of the abnormal feature patterns. This algorithm assigns higher weights to the abnormal patterns that have appeared recently to reflect the latest abnormal trends. For example, the time decay factor can be set to 0.9 in the weight calculation, making the contribution of the most recent data greater. This update mechanism ensures that the abnormal feature network always makes decisions and repairs based on the latest abnormal information.

[0064] Through the above method, the present invention combines multiple technologies such as the mapping network, the reverse reinforcement feedback mechanism, the virtual platform simulation, and the abnormal feature network to achieve the compatibility testing and automatic optimization of wearable devices on different platforms. This method not only improves the comprehensiveness and accuracy of the testing but also significantly enhances the adaptive ability of the device and the user experience.

[0065] To more intuitively demonstrate the actual application effect of the present invention, a specific case can be used to elaborate in detail on the implementation process of this method in the cross-platform compatibility testing of wearable devices and the significant advantages it brings.

[0066] Imagine an intelligent watch as a wearable device that needs to be tested for compatibility on multiple smartphone platforms. These platforms include common Android and iOS systems, as well as some customized operating systems, such as a special system based on Linux. Different platforms have differences in hardware configurations, operating system versions, Bluetooth protocol implementations, etc., which may cause various problems when the intelligent watch connects to different platforms, such as unstable connections, data synchronization delays, or slower user interface response speeds. To solve these problems, the test team designed a systematic method using data collection, feature mapping, and intelligent algorithms to comprehensively evaluate and optimize the cross-platform performance of the intelligent watch.

[0067] The first step of the test is to collect detailed data generated when the intelligent watch interacts with these platforms from multiple different platforms. Assume that the platforms involved in the test include Android 10, Android 11, iOS14, iOS15, and a customized system based on Linux.

[0068] The main types of data that the team focuses on include device response latency, signal transmission characteristics, and the reaction speed of the user interface. For example, on the Android 10 platform, testers will record the specific time when the smartwatch gives a response after receiving an instruction from the mobile phone, measure the transmission rate and stability of the Bluetooth signal at the same time, and also evaluate the time required for the interface to give feedback after the user clicks on the watch interface. These data collection processes are also carried out on other platforms to ensure coverage of all key interaction scenarios.

[0069] In this way, the testing team obtained a complete picture of the performance of the smartwatch on different platforms and clearly saw the differences between different systems. For example, iOS 15 may be more stable in signal transmission, while Android 11 may have a slightly longer response latency in some cases.

[0070] With these interaction feature data, the testing team began to build two core spaces: one is the platform feature space, and the other is the device behavior space. The platform feature space is a vector composed of the interaction feature data of each platform. For example, the vector of Android 10 may contain specific metrics such as the average response latency and signal transmission rate; while the device behavior space reflects the behavior patterns of the smartwatch on these platforms, such as the distribution of response times.

[0071] Next, the team used graph neural network technology to generate a mapping network that connects platform features and device behavior features. This network is like a complex map, where each platform and device behavior are regarded as individual nodes, and the connecting lines (edges) between the nodes represent the interaction relationships between them. For example, the Android 10 platform node may be connected to the response latency feature node of the smartwatch on this platform, and the weight of this edge may be associated with the specific latency value. Through the learning of the graph neural network, the system gradually captures the deep dependence relationship between platform characteristics and device behavior, forming a reliable prediction model.

[0072] When faced with a new platform, such as the newly released Android 12 system, the testing team does not need to start a comprehensive test from scratch, but can use the existing mapping network for a quick evaluation. First, the interaction feature data of Android 12 can be collected, such as Bluetooth transmission rate, system response time, etc., and then these data are input into the mapping network. The system will use the feature propagation algorithm to transmit the feature information of the new platform in the network and predict the performance of the smartwatch on Android 12 based on the existing mapping relationship.

[0073] For example, the algorithm might notice that certain features of Android 12 are similar to those of Android 11, and thus speculate that the response latency of the smartwatch on these two platforms might be similar. However, considering that Android 12 introduced some new features, such as an updated Bluetooth protocol, the system can also predict potential compatibility issues. For example, the new protocol might lead to unstable connections. This predictive ability allows the testing team to identify problems in advance without spending a lot of time deploying and testing on real devices.

[0074] To further improve the accuracy of testing, the team also introduced autoencoders to establish normal interaction patterns of the smartwatch on various platforms. By analyzing historical data, the autoencoders learn the feature distribution under normal circumstances. For example, on Android 10, the normal response latency might be between 50 milliseconds and 100 milliseconds, and the signal transmission rate is stable at around 1 Mbps. If during an actual test, the response latency of a certain platform suddenly soars to 200 milliseconds, the system will calculate the deviation between the actual data and the normal pattern and determine whether it exceeds a preset threshold (such as 2 standard deviations of the normal deviation). Once the deviation exceeds the standard, the system will activate an inverse reinforcement feedback mechanism, using the magnitude of the deviation as a signal to guide the smartwatch to actively explore more scenarios where anomalies might occur. For example, the watch might automatically adjust Bluetooth connection parameters or simulate data transmission under high load to identify the root cause of the problem. In a test, assuming the actual latency is 200 milliseconds, the normal latency is 75 milliseconds, and the deviation is 125 milliseconds, the system calculates a negative reward signal (the specific formula is r = -0.1 × 125 = -12.5), and this signal drives the watch to optimize its internal algorithm or communication protocol to improve performance.

[0075] To enable testing to cover more scenarios, the smartwatch is built with a virtual platform emulation module. This module uses the Monte Carlo random sampling algorithm, combined with key parameters in the platform feature space (such as the operating system API level, Bluetooth version, processor performance, etc.), to automatically generate various virtual platform scenarios. For example, it might simulate a platform with a lower API level and an older Bluetooth version, or an environment with weaker processor performance. When sampling, the system adjusts the probability of parameter combinations based on historical data to avoid over-focusing on known problems. Through these virtual scenarios, the module can simulate the interaction process between the smartwatch and the platform, such as data synchronization on a low-performance platform, thereby identifying potential problems in advance, such as data processing delays that might occur when the processor is insufficient.

[0076] In addition, the smartwatch is equipped with an abnormal feature network, which is responsible for recording the abnormal data that appears in each interaction and forming an abnormal feature memory bank. For example, if a certain platform experiences connection interruptions multiple times, the system will record information such as the Bluetooth signal strength and transmission rate at that time. This network uses a random forest classifier to train a model based on the data in the memory bank to identify new abnormalities and match known patterns. For example, it may find that disconnections are likely to occur when the signal strength is lower than a certain value. When a new abnormality is detected, the system will automatically generate repair instructions, such as increasing the number of Bluetooth reconnections or reducing the data transmission frequency. These repair measures directly act on the watch to improve its adaptive ability.

[0077] To capture abnormalities more promptly, the data acquisition frequency is designed to be adaptively adjusted and linked to the frequency of abnormality occurrences. Under normal circumstances, the acquisition frequency is once per second; if there are 3 connection interruptions within 5 minutes, the frequency will increase to once every 0.5 seconds to ensure more detailed data is recorded. When the abnormalities decrease, the frequency will gradually return. This dynamic mechanism ensures sufficient information at critical moments.

[0078] There is also a real-time feedback channel between the virtual platform simulation module and the reverse reinforcement feedback mechanism. If it is found in the simulation that the response latency is abnormally high in a certain scenario, this information will be immediately transmitted to the feedback mechanism to update the definition of the normal mode, making the system more in line with the actual situation. The abnormal feature memory bank will also be updated regularly, using the exponentially weighted moving average algorithm to adjust the weights of the abnormal patterns (the formula is w new = 0.9w old + 0.1w current ) to ensure attention to the latest abnormal trends.

[0079] Through this case, we can see the powerful advantages of the present invention in practical applications. The mapping network and the feature propagation algorithm make the compatibility prediction of the new platform efficient, saving a large amount of test resources. The combination of the reverse reinforcement feedback and the virtual simulation module actively mines potential problems and improves the comprehensiveness of the test. The abnormal feature network enhances the adaptive ability of the device and automatically repairs problems. The adaptive acquisition frequency and the real-time feedback channel ensure the efficient operation of the system. Generally speaking, this set of methods provides an automated and intelligent solution for the cross-platform compatibility testing of wearable devices, which is not only highly practical but also shows great market potential.

[0080] The above description is only the preferred specific implementation manner of the present invention, but the protection scope of the present invention is not limited thereto. Any person skilled in the art within the technical scope disclosed by the present invention, according to the technical solution and inventive concept of the present invention, makes equivalent substitutions or changes, and should be covered by the protection scope of the present invention.

Claims

1. A cross-platform compatibility testing method for wearable devices based on dynamic adaptation, characterized in that, Including: Collecting interaction feature data generated during the interaction process between the wearable device and platforms through multiple different platforms. The interaction feature data includes any one or more of the following: device response latency feature, signal transmission feature, user interface response feature; Respectively constructing a platform feature space and a device behavior space based on the interaction feature data, and using a graph neural network to generate a mapping network to form a mapping relationship of the interaction features between the device and the platform; When a new platform to be tested appears, first collect the interaction feature data of the new platform and map it in the mapping network, and predict possible compatibility problems between the wearable device and the new platform through a feature propagation algorithm.

2. The cross-platform compatibility testing method for wearable devices based on dynamic adaptation according to claim 1, wherein: When constructing the mapping network, use an autoencoder to establish the normal mode of the interaction features of the wearable device on each platform; When the deviation between the actual interaction feature data and the normal mode of the interaction features exceeds a predetermined threshold, trigger a reverse reinforcement feedback mechanism, use the numerical value of the deviation as a reverse reward signal, and automatically guide the wearable device to actively generate and deeply detect more potentially abnormal interaction scenarios.

3. The cross-platform compatibility testing method for wearable devices based on dynamic adaptation according to claim 2, characterized in that: The reverse reward signal r generated by the reverse reinforcement feedback mechanism satisfies: r = -λ||x real - x norm ||2; where λ is the proportionality coefficient, and x real is the actual interaction feature data vector, and x norm is the interaction feature data vector in the normal mode.

4. The cross-platform compatibility testing method for wearable devices based on dynamic adaptation according to claim 1, characterized in that: A virtual platform simulation module is provided inside the wearable device. Using the Monte Carlo random sampling algorithm and combining key feature parameters in the platform feature space, automatically generate a virtual platform scenario combination, and actively simulate the interaction process between the device and the platform based on the virtual platform scenario combination, so as to identify potential compatibility anomalies in advance.

5. The cross-platform compatibility testing method for wearable devices based on dynamic adaptation according to claim 4, characterized in that: When the virtual platform simulation module generates a virtual platform scenario combination, the sampling distribution P of the key feature parameter F of the platform feature space satisfies: In the formula, β is a coefficient controlling the degree of randomness, E(F) is an abnormal tendency energy function of the feature parameter, and Z is a normalization factor.

6. The cross-platform compatibility testing method for wearable devices based on dynamic adaptation according to claim 1, characterized in that: An abnormal feature network is provided inside the wearable device, and automatically records the abnormal feature data that appears during each interaction process between the device and the platform to form an abnormal feature memory bank; The abnormal feature network is trained using a random forest classifier based on the abnormal feature memory bank to identify newly emerging abnormal feature data and automatically match the known abnormal data patterns stored in the abnormal feature memory bank.

7. The cross-platform compatibility testing method for wearable devices based on dynamic adaptation according to claim 6, characterized in that: After the abnormal feature network identifies the abnormal feature data, it automatically generates corresponding repair instructions based on the abnormal feature memory bank. The repair instructions include any one or more of the following: Bluetooth protocol parameter optimization, data transmission rate adjustment, and interface interaction mode switching.

8. The cross-platform compatibility testing method for wearable devices based on dynamic adaptation according to claim 1, characterized in that: The acquisition frequency of the interaction feature data is adaptively positively correlated with the frequency of anomalies detected during the device interaction process. When the frequency of anomalies during the device interaction process increases, the acquisition frequency increases; when the frequency of anomalies decreases, the acquisition frequency decreases.

9. The cross-platform compatibility testing method for wearable devices based on dynamic adaptation according to claim 4, wherein: A data real-time feedback channel is provided between the virtual platform simulation module and the reverse reinforcement feedback mechanism. Through the data real-time feedback channel, the abnormal data found during the virtual simulation process is real-time fed back to the reverse reinforcement feedback mechanism to update the interaction feature data vector of the normal mode.

10. The cross-platform compatibility testing method for wearable devices based on dynamic adaptation according to claim 6, characterized in that: The abnormal data patterns in the abnormal feature memory bank are regularly updated online according to newly added abnormal feature data, and the exponential weighted moving average algorithm is used to determine the weights of abnormal feature patterns. The weights of abnormal feature patterns that have occurred recently are greater than those of abnormal feature patterns that have occurred historically.

Images