Mobile terminal power supply fault detection method and system
By preprocessing and feature extraction of polarity signal data, a high-dimensional vector output is generated. Combined with Euclidean distance and support vector machine models, the problem of insufficient accuracy in mobile power fault detection is solved, and high-precision and stable fault detection is achieved.
Patent Information
- Application Number
- CN202511468710.1
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2025-10-15
- Publication Date
- 2026-01-02
AI Technical Summary
Existing technologies lack sufficient accuracy in detecting power failures in mobile devices under dynamic usage scenarios, making it difficult to adapt to baseline drift under different power consumption modes, leading to false alarms or missed detections, and failing to support efficient fault prediction and health management.
By acquiring polarity signal data and preprocessing it, the amplitude change trajectory features and duration features are extracted to generate a high-dimensional vector output. The Euclidean distance is calculated to mark potential anomalies. The adjustment strategy is verified by combining the support vector machine model, thus realizing closed-loop control from discovery to classification to adjustment to verification.
It improves the sensitivity and accuracy of anomaly detection, reduces the false alarm rate, ensures high-precision sensing and stability of power signals, and achieves accuracy and reliability in mobile terminal power fault detection.
Smart Images
Figure CN121256644A_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the field of mobile power management technology, and in particular to a method and system for detecting power failures in mobile devices. Background Technology
[0002] Currently, mobile power management has become a core area of operation for modern electronic devices, and its stability directly affects the device's functional performance and user experience. With the widespread adoption of smart terminals, the reliability requirements of power systems are constantly increasing. Any potential fault can lead to performance degradation or even failure, thereby impacting device availability and user safety. Therefore, how to conduct effective fault prediction and health management has become an important research direction for ensuring the long-term stable operation of mobile power systems.
[0003] In one existing technology, mobile terminal power fault detection first collects voltage, current and polarity signals at the power output terminal through an analog front end at a fixed sampling rate. Then, the collected signals are preprocessed by filtering and normalization to eliminate noise and power frequency interference. Statistical features such as peak values are extracted within a sliding window and combined with threshold values to determine anomalies. When an anomaly is determined, an alarm or protection mechanism is triggered, and the results are recorded for subsequent review.
[0004] However, fixed sampling rates and windowing strategies often fail to capture transient transitions, and filtering with fixed parameters often weakens or even masks short-term anomalies. Single-dimensional features struggle to characterize the dynamic trajectory of polar signals in both amplitude and duration dimensions, and cannot adapt to baseline drift in different power consumption modes. This leads to a high likelihood of false alarms or missed detections, resulting in insufficient accuracy for mobile device power fault detection in dynamic usage scenarios and hindering efficient fault prediction and health management. In summary, existing technologies suffer from insufficient accuracy in mobile device power fault detection under dynamic usage scenarios. Summary of the Invention
[0005] This invention provides a method and system for detecting power failures in mobile devices, in order to solve the problem of insufficient accuracy in detecting power failures in mobile devices under dynamic usage scenarios.
[0006] Firstly, in order to solve the above-mentioned technical problems, the present invention provides a mobile terminal power failure detection method, comprising: The polarity signal data is acquired and preprocessed to obtain the polarity signal sequence; The amplitude and duration values at each time point in the polar signal sequence are obtained and features are extracted to obtain amplitude change trajectory features and duration features, and a high-dimensional vector output is generated. Calculate the Euclidean distance of the high-dimensional vector output. If the Euclidean distance exceeds a preset distance threshold, it is marked as a potential anomaly, and an anomaly candidate list is obtained. Based on the anomaly candidate list, the amplitude change trajectory features and duration features of each potential anomaly are obtained. An anomaly score of the potential anomaly is calculated according to a preset anomaly judgment mechanism. The anomaly score is compared with a preset anomaly score threshold to confirm the anomaly type and obtain the anomaly classification result. Based on the anomaly classification results, preset power management parameters and real-time status information of the mobile device are obtained. Based on the real-time status information and preset status thresholds, the preset power management parameters are optimized to obtain a real-time adjustment strategy. The real-time adjustment strategy is validated using a preset support vector machine model, and the validation accuracy is calculated. If the validation accuracy is higher than a preset accuracy threshold, a fault detection report is generated.
[0007] In one optional implementation, the step of acquiring polarity signal data and preprocessing it to obtain a polarity signal sequence includes: Acquire polarity signal data; The polarity signal data is converted into a digital signal to obtain the original signal sequence; The features of the original signal sequence are extracted and noise is suppressed to obtain a denoised signal sequence. If the denoised signal sequence has fluctuations, then the denoised signal sequence is smoothed to obtain a polar signal sequence.
[0008] In one optional implementation, the step of acquiring the amplitude and duration values at each time point in the polar signal sequence and extracting features to obtain amplitude change trajectory features and duration features, and generating a high-dimensional vector output, includes: Obtain the amplitude and duration values at each time point in the polar signal sequence to obtain a feature set; The feature set is traversed point by point, and the amplitude change trajectory features and duration features are extracted according to the preset convolution kernel to obtain the initial feature map; The initial feature map is subjected to dimensionality reduction processing using a preset pooling layer to obtain a dimensionality-reduced feature map; The reduced-dimensional feature map is used to generate a high-dimensional vector output by calculating a preset activation function.
[0009] In one optional implementation, the Euclidean distance of the high-dimensional vector output is calculated. If the Euclidean distance exceeds a preset distance threshold, it is marked as a potential anomaly, resulting in an anomaly candidate list, including: Obtain the preset high-dimensional vector of the normal state; Calculate the Euclidean distance between the high-dimensional vector output and the high-dimensional vector in the normal state; If the Euclidean distance exceeds a preset distance threshold, it is marked as a potential anomaly, and the state characteristics of the potential anomaly are recorded. By integrating the aforementioned state features, an anomaly candidate list is obtained.
[0010] In one optional implementation, the step of obtaining the amplitude variation trajectory characteristics and duration characteristics of each potential anomaly based on the anomaly candidate list, calculating the anomaly score of the potential anomaly according to a preset anomaly judgment mechanism, comparing the anomaly score with a preset anomaly score threshold, confirming the anomaly type, and obtaining an anomaly classification result includes: Obtain the amplitude change trajectory features and duration features of each potential anomaly in the anomaly candidate list to obtain an anomaly feature set; The abnormal feature set is weighted to obtain an abnormal score, and the abnormal score is compared with a preset abnormal score threshold to obtain a preliminary abnormal classification set. The preliminary anomaly classification set is optimized to obtain the anomaly classification results.
[0011] In one optional implementation, the step of obtaining preset power management parameters and real-time status information of the mobile device based on the anomaly classification result, and optimizing the preset power management parameters based on the real-time status information and preset status thresholds to obtain a real-time adjustment strategy includes: Based on the anomaly classification results, preset power management parameters and anomaly status information are extracted from the preset anomaly classification template database. The real-time status information of the mobile device is obtained and compared with the abnormal status information. The data consistency between the real-time status information and the abnormal status information is analyzed to obtain a data consistency score. If the data consistency score exceeds a preset consistency threshold, the preset power management parameters are adjusted based on the real-time status information to obtain a real-time adjustment strategy.
[0012] In one optional implementation, the real-time adjustment strategy is validated using a preset support vector machine model, and the validation accuracy is calculated. If the validation accuracy is higher than a preset accuracy threshold, a fault detection report is generated, including: Obtain the mobile device's runtime data and runtime environment parameters; Using a preset support vector machine model, the verification accuracy of the real-time adjustment strategy is calculated based on the running data and the running environment parameters. The verification accuracy rate is compared with a preset accuracy threshold. If the accuracy rate exceeds the preset accuracy threshold, a fault detection report is generated.
[0013] Secondly, the present invention provides a mobile terminal power failure detection device, comprising: The data acquisition module is used to acquire polar signal data and preprocess it to obtain a polar signal sequence; The feature extraction module is used to obtain the amplitude value and duration value of each time point in the polar signal sequence and extract features to obtain amplitude change trajectory features and duration features, and generate a high-dimensional vector output. An anomaly labeling module is used to calculate the Euclidean distance of the high-dimensional vector output. If the Euclidean distance exceeds a preset distance threshold, it is labeled as a potential anomaly, and an anomaly candidate list is obtained. An anomaly classification module is used to obtain the amplitude change trajectory features and duration features of each potential anomaly based on the anomaly candidate list, calculate the anomaly score of the potential anomaly according to a preset anomaly judgment mechanism, compare the anomaly score with a preset anomaly score threshold, confirm the anomaly type, and obtain the anomaly classification result. The adjustment strategy generation module is used to obtain preset power management parameters and real-time status information of the mobile device based on the anomaly classification results, optimize the preset power management parameters based on the real-time status information and preset status thresholds, and obtain a real-time adjustment strategy. The fault report generation module is used to verify the real-time adjustment strategy using a preset support vector machine model, calculate the verification accuracy, and generate a fault detection report if the verification accuracy is higher than a preset accuracy threshold.
[0014] Thirdly, the present invention also provides an electronic device, including a processor, a memory, and a computer program stored in the memory and configured to be executed by the processor, wherein the processor executes the computer program to implement the mobile terminal power failure detection method described in any one of the above.
[0015] Fourthly, the present invention also provides a computer-readable storage medium comprising a stored computer program, wherein, when the computer program is executed, it controls the device where the computer-readable storage medium is located to perform the mobile terminal power failure detection method described in any one of the above.
[0016] Compared with the prior art, the present invention has the following beneficial effects: (1) This invention calculates the anomaly score of potential anomalies and classifies anomalies by combining them with a preset anomaly score threshold, thereby generating a fault classification result. Then, it optimizes the power management parameters by combining them with real-time device status information to form a real-time adjustment strategy. The adjustment strategy is verified by using a preset support vector machine model. This enables closed-loop control from discovery to classification to adjustment to verification in the fault detection and handling process, ensuring the safety and executability of the scheduling strategy and solving the problem of insufficient accuracy of mobile terminal power fault detection in the prior art.
[0017] (2) This invention acquires polar signal data and performs preprocessing, including denoising, normalization, baseline correction and event segmentation, to transform the original polar signal into a standardized polar signal sequence. In this process, interference factors are suppressed and different acquisition conditions are unified, ensuring the effectiveness of subsequent feature extraction. It can obtain continuous and reliable signal input in complex acquisition environments, which not only improves the sensitivity of anomaly detection, but also effectively reduces the false alarm rate, and realizes high-precision perception of power signals by mobile terminals.
[0018] (3) This invention extracts amplitude change trajectory features and duration features from polar signal sequences, combines the two types of features into a high-dimensional vector output, and then compares the Euclidean distance between the high-dimensional vector output and the normal behavior reference vector to obtain a potential anomaly candidate list. This avoids the limitations of traditional methods that rely on a single image threshold for judgment, and can characterize anomalies from the perspectives of overall shape and persistence. By jointly measuring trajectory and duration features, it can effectively filter out instantaneous noise interference, thereby improving the accuracy and stability of anomaly identification. Attached Figure Description
[0019] Figure 1 This is a schematic diagram of the mobile terminal power failure detection method provided in the first embodiment of the present invention; Figure 2 This is a schematic diagram of the mobile terminal power fault detection system provided in the second embodiment of the present invention. Detailed Implementation
[0020] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.
[0021] Reference Figure 1 The first embodiment of the present invention provides a method for detecting power failure in a mobile device, comprising the following steps: S11, acquire polarity signal data and preprocess it to obtain a polarity signal sequence; S12, obtain the amplitude value and duration value of each time point in the polar signal sequence and extract features to obtain amplitude change trajectory features and duration features, and generate a high-dimensional vector output; S13, calculate the Euclidean distance of the high-dimensional vector output. If the Euclidean distance exceeds a preset distance threshold, it is marked as a potential anomaly, and an anomaly candidate list is obtained; S14. Based on the anomaly candidate list, obtain the amplitude change trajectory features and duration features of each potential anomaly, calculate the anomaly score of the potential anomaly according to the preset anomaly judgment mechanism, compare the anomaly score with the preset anomaly score threshold, confirm the anomaly type, and obtain the anomaly classification result. S15, Based on the anomaly classification result, obtain the preset power management parameters and the real-time status information of the mobile device, and optimize the preset power management parameters based on the real-time status information and the preset status threshold to obtain a real-time adjustment strategy; S16, the real-time adjustment strategy is verified using a preset support vector machine model, and the verification accuracy is calculated. If the verification accuracy is higher than the preset accuracy threshold, a fault detection report is generated.
[0022] In step S11, acquiring polarity signal data and preprocessing it to obtain a polarity signal sequence includes: Acquire polarity signal data; The polarity signal data is converted into a digital signal to obtain the original signal sequence; The features of the original signal sequence are extracted and noise is suppressed to obtain a denoised signal sequence. If the denoised signal sequence has fluctuations, then the denoised signal sequence is smoothed to obtain a polar signal sequence.
[0023] It should be noted that the polarity signal is obtained from the power output of the mobile device by the sensor at a frequency of 1000 times per second. This signal contains polarity information that reflects the direction of current or the positive or negative of voltage, and characterizes the charging or discharging state of the device.
[0024] Then, an analog-to-digital converter (ADC) is used to convert the polarity signal into a digital signal with 12-bit resolution, obtaining the original signal sequence. Specifically, a sampling frequency of 1000Hz is selected, meaning the input signal is sampled 1000 times per second. The corresponding sampling period is 100μs, meaning the time interval between two adjacent samples is 100 microseconds. Within this sampling period, the sample-and-hold circuit latches the instantaneous value of the input voltage and holds that level until the analog-to-digital conversion is complete. Subsequently, the ADC converts the held voltage value into a 12-bit binary code within each sampling period to obtain the original signal sequence.
[0025] Feature extraction and noise suppression are performed on the original polarity signal sequence. Noise detection involves using a Fast Fourier Transform (FFT) to extract the spectral distribution, identifying and labeling high-frequency or low-frequency components outside the 10Hz-5kHz range as noise. The effective frequency band is set to 10Hz-5kHz because the main energy of the target polarity signal is concentrated in this range; DC drift below 10Hz and high-frequency components above 5kHz are mostly environmental interference or random noise and are therefore discarded. Specifically, the signal is first averaged within a short time window of 100μs to obtain signal feature values, capturing signal fluctuations on a short time scale, such as rapid voltage rises, falls, and instantaneous peaks. The 100μs window length is chosen based on a 100μs sampling period, ensuring that each window precisely covers the complete signal changes within an adjacent sampling period. This setting maintains the window length consistent with the sampling interval, capturing instantaneous signal changes while avoiding the smoothing or loss of detail that might occur with excessively long windows. Meanwhile, since the main rate of change of polar signals is relatively high, this window length can perform local averaging of random noise while ensuring time-domain resolution, achieving a balance between noise suppression and signal feature preservation. Then, components in the signal with a frequency exceeding 5kHz or an amplitude below 5mV are marked. These components are usually random noise. The signal feature values extracted within one time window before and after the noise nodes are averaged to obtain the denoised signal sequence.
[0026] Based on the denoised signal sequence, fluctuation detection is performed on each sampling point to determine if any abnormal changes exist. The specific operation is as follows: First, the amplitude difference between the current sampling point and the previous sampling point is calculated. When the amplitude difference exceeds 10 mV, the sampling point is considered to have significant fluctuation. The 10 mV threshold is set based on the difference between the amplitude deviation of each sampling data point and the overall mean, calculated through statistical analysis of historical sampling data. Then, the absolute value of all differences is taken, and the maximum value is calculated. The results show that the signal amplitude fluctuation will not exceed 10 mV, while changes exceeding 10 mV are considered to have fluctuation and require smoothing. If a sampling point is determined to have fluctuation, five sampling points before and after it (a total of 10 neighboring sampling points) are selected, and the arithmetic mean of their amplitudes is calculated. This arithmetic mean is used as the feature value of the window to replace the amplitude of the original sampling point, resulting in a polar signal sequence. This operation effectively reduces the impact of single-point anomalies on the overall sequence while maintaining the overall trend of the signal. Selecting 10 sampling points as the window length ensures smoothing while avoiding the loss of signal details due to an excessively long window.
[0027] In step S12, the step of acquiring the amplitude and duration values at each time point in the polar signal sequence and extracting features to obtain amplitude change trajectory features and duration features, and generating a high-dimensional vector output, includes: Obtain the amplitude and duration values at each time point in the polar signal sequence to obtain a feature set; The feature set is traversed point by point, and the amplitude change trajectory features and duration features are extracted according to the preset convolution kernel to obtain the initial feature map; The initial feature map is subjected to dimensionality reduction processing using a preset pooling layer to obtain a dimensionality-reduced feature map; The reduced-dimensional feature map is used to generate a high-dimensional vector output by calculating a preset activation function.
[0028] It should be noted that the polarity signal sequence is traversed point by point to extract the amplitude and duration values, thus obtaining the feature set. For example, if a signal sequence collects data 1000 times per second at a time interval of 0.001 seconds, and the amplitude values of a certain segment of the sequence are 4.2V, 4.3V, 4.1V, etc., after traversal, the amplitude value set [4.2, 4.3, 4.1, ...] and the duration value set [0.001, 0.001, 0.001, ...] can be obtained.
[0029] Based on this, a preset sliding weighted window is applied to perform feature calculations on the amplitude value set and the duration value set to obtain an initial feature map. According to actual R&D experience, selecting 5 sampling points as the window length can achieve a good balance between smoothing noise and preserving local signal details. Therefore, the preset window length is 5 sampling points, corresponding to convolution kernel weights of [0.25, 0.5, 1.0, 0.5, 0.25]. The weights are arranged in descending order of weight from the center point. Specifically, the center point usually corresponds to the most critical sampling position within the window, reflecting the amplitude peak or local changes, and needs to be given the largest weight to highlight the main signal features. Therefore, the center point is assigned a weight of 1. The sampling points on both sides of the window contribute to the overall trend, but may be affected by random noise. They are given smaller weights to smooth local fluctuations. Therefore, the weights are calculated according to the rule of decreasing by 1 / 2 from the center point. Finally, all node weights are divided by the sum of the weights for normalization to obtain the final weights. Subsequently, a sliding window is used to move point by point on the amplitude value set. Each time, the amplitude values of the 5 sampling points in the window are multiplied by the convolution kernel weights and summed to obtain the amplitude change feature of the center point of the window. The same method is applied to the duration value set to obtain the duration feature.
[0030] The calculation results for all windows are arranged sequentially to form an amplitude variation trajectory feature map and a duration feature map. Specifically, the minimum and maximum values are calculated for each feature map. Then, the minimum value is subtracted from each feature value, and the result is divided by the difference between the maximum and minimum values, mapping it to the range of 0 to 1. This ensures that features of different dimensions can be compared and merged at the same scale. The two feature maps are then concatenated as column vectors at corresponding time steps to form a two-dimensional matrix. Each row corresponds to the amplitude and duration features of the same time window, and each column corresponds to a different feature type. This matrix is the initial feature map, which can be viewed as a two-dimensional tensor.
[0031] After obtaining the initial feature map, dimensionality reduction is performed to extract more compact feature information. For this purpose, one-dimensional max pooling is used, with a window length of 2 sampling points and a stride of 2. Specifically, on the initial feature map, two sampling points are grouped together, and the maximum value within the group is selected as the output value of that window. Then, the sliding window moves sequentially with a stride of 2 to obtain a new dimensionality-reduced feature sequence. When the initial feature map length is odd, zeros are padded at the end to make the length even, satisfying the pooling window requirement. In this way, the length of the initial feature map is halved after each pooling operation, effectively reducing computational complexity while preserving the most salient local features and ensuring that the main variation patterns are not lost. The pooling window length and stride are chosen based on the following criteria: a length of 2 is sufficient to balance feature preservation and dimensionality reduction, and a stride of 2 ensures that the windows do not overlap, resulting in efficient and smooth computation. Subsequently, a preset activation function is applied to the dimensionality-reduced feature map to generate a high-dimensional vector output. The ReLU activation function is selected, taking the value between the maximum value and zero for each feature to obtain a high-dimensional vector output. After Min-Max normalization, the feature values are all in the [0,1] interval, and choosing ReLU does not cause information loss. The reasons for choosing ReLU include: nonlinear transformation can enhance the expressive power of features, highlight positive signal changes, and suppress negative or invalid information; it is simple to calculate and has a fast convergence speed, which is conducive to the stable generation of subsequent high-dimensional vector representations; in signal amplitude feature processing, ReLU can effectively retain the main positive fluctuation information and avoid interference introduced by small negative values. After pooling dimensionality reduction and ReLU activation processing, the initial feature map is transformed into a high-dimensional vector output, providing reliable input features for subsequent pattern analysis, anomaly detection, or classification.
[0032] In step S13, the Euclidean distance of the high-dimensional vector output is calculated. If the Euclidean distance exceeds a preset distance threshold, it is marked as a potential anomaly, resulting in an anomaly candidate list, including: Obtain the preset high-dimensional vector of the normal state; Calculate the Euclidean distance between the high-dimensional vector output and the high-dimensional vector in the normal state; If the Euclidean distance exceeds a preset distance threshold, it is marked as a potential anomaly, and the state characteristics of the potential anomaly are recorded. By integrating the aforementioned state features, an anomaly candidate list is obtained. It should be noted that, based on historical data from the mobile power supply under normal operating conditions, statistical analysis is performed on the collected high-dimensional vector, and the average value is calculated for each element. The historical samples are from monitoring data obtained after the device has been operating continuously for 24 hours or more. Specifically, based on the historical samples, the minimum and maximum values for each dimension are calculated, and then each element of the average value vector is normalized. The vector formed by the calculation results is used as the high-dimensional vector of the device's normal state. For example, suppose there are five sets of sample vectors in the historical data, namely [2.0, 1.5, 3.0, 4.2, 1.8], [2.4, 1.9, 3.4, 3.8, 2.1], [2.6, 1.7, 3.1, 4.0, 2.0], [2.5, 1.8, 3.3, 4.1, 2.2] and [2.3, 1.6, 3.2, 3.9, 1.9]. After element-wise averaging, the unnormalized normal state high-dimensional vector [2.36, 1.7, 3.2, 4.0, 2.0] is obtained. To eliminate the impact of dimensional differences on subsequent analysis, the system normalizes each dimension: first, it calculates the minimum and maximum values of that dimension across all samples; then, it subtracts the minimum value from the average and divides by the difference between the maximum and minimum values, mapping the result to a range of 0 to 1. The normalized vector calculated using this method is [0.6, 0.5, 0.5, 0.5, 0.5], which serves as the high-dimensional vector for the normal state.
[0033] Subsequently, the real-time generated high-dimensional vector output is compared with the preset normal-state high-dimensional vector, and Euclidean distance is used as the similarity metric for calculation. Specifically, the differences between corresponding elements of the two high-dimensional vectors are calculated one by one and squared. Then, all squared differences are summed, and the square root of the result is taken to obtain the Euclidean distance value. This distance value can quantitatively reflect the degree of deviation between the currently acquired signal characteristics and the normal-state characteristics, providing a basis for subsequent anomaly judgment.
[0034] When the calculated Euclidean distance exceeds 0.8, the high-dimensional vector output is marked as a potential anomaly, and the corresponding state characteristics are recorded. The threshold of 0.8 is selected based on the following: statistical analysis of high-dimensional vectors collected from a large number of mobile power supplies under normal operating conditions revealed that the Euclidean distance under normal conditions is usually concentrated between 0 and 0.6, while deviations exceeding 0.8 typically correspond to abnormal changes in signal amplitude or duration. Setting the threshold to 0.8 effectively ensures that while capturing true abnormal states, false alarms caused by normal fluctuations are reduced, thereby achieving stable and reliable anomaly detection.
[0035] The state features corresponding to high-dimensional vectors marked as potential anomalies are summarized and organized, then categorized and deduplicated according to time sequence or acquisition sequence to form a structured anomaly candidate list. Categorization and deduplication involve arranging potential anomaly vectors in chronological order based on timestamps. Then, feature vectors in adjacent time windows with an Euclidean distance less than a threshold (specifically set to 0.05; high-dimensional vector feature values are floating-point numbers, e.g., amplitude value 3.1415926; analysis of historical data, using ROC curves and comparative testing, revealed that the boundary between normal fluctuations and anomalies is in the 0.05-0.10 range; a value of 0.05 covers 92% of normal fluctuations) are considered duplicate anomalies, and only the earliest recorded instance is retained. Subsequently, with the acquisition sequence start time as minute 0, windows are divided every 5 minutes. Within the same time window, the record with the most significant or maximum amplitude feature is selected as the anomaly representative for that window, reflecting the temporal concentration of anomalies. Since the duration of a single anomaly event is concentrated between 2 and 4 minutes, and the same device may experience consecutive anomalies within a short period, this process is particularly effective. The system sets the time window length to 5 minutes, which can cover the complete process of both single and continuous abnormal events, avoiding the splitting of continuous abnormalities into different time periods, and ensuring that each time period contains a sufficient number of potential abnormal vectors for statistical analysis and subsequent processing.
[0036] For example, if three potential anomaly vectors appear within a time window of 15:00-15:05, after deduplication, two vectors with significant feature differences are retained, and the vector with the largest amplitude feature is selected as the anomaly representative of that window and recorded in the list.
[0037] In step S14, based on the anomaly candidate list, the amplitude change trajectory features and duration features of each potential anomaly are obtained; an anomaly score is calculated for each potential anomaly according to a preset anomaly judgment mechanism; the anomaly score is compared with a preset anomaly score threshold to confirm the anomaly type and obtain an anomaly classification result, including: Obtain the amplitude change trajectory features and duration features of each potential anomaly in the anomaly candidate list to obtain an anomaly feature set; The abnormal feature set is weighted to obtain an abnormal score, and the abnormal score is compared with a preset abnormal score threshold to obtain a preliminary abnormal classification set. The preliminary anomaly classification set is optimized to obtain the anomaly classification results.
[0038] It should be noted that, by iterating through each potential anomaly in the candidate anomaly list, the magnitude feature value of the potential anomaly is extracted (denoted as ). ) and duration feature value (denoted as The normal amplitude reference value is stored in the high-dimensional vector of the normal state. and normal duration baseline value Anomaly scores are calculated using a standardized weighted scoring system, with the specific formula as follows: in, =0.7, which is the amplitude feature weighting coefficient; =0.3 is the weighting coefficient for the duration feature; The abnormal amplitude characteristic value (unit: volts V); Normal amplitude reference value (unit: volts V); The duration of the anomaly is a characteristic value (in seconds). This is the baseline value for normal duration (unit: seconds). Weighting ratio. : The determination of the ratio 0.7:0.3 is based on historical data statistical analysis. Shapley value analysis of 1,000 sets of historical fault data showed that the average contribution rate of amplitude features was 68% (standard deviation ± 5%), and that of duration features was 32% (standard deviation ± 6%). Subsequently, the random forest algorithm was used to evaluate the feature importance, and the output feature importance was [0.72, 0.28], confirming the rationality of the 0.7:0.3 allocation.
[0039] Among them, the high-dimensional vector under normal conditions is obtained by statistically analyzing the collected high-dimensional vector based on historical data of the mobile terminal power supply under normal operating conditions, calculating the average value of each element, and forming the result into a vector.
[0040] A grading mechanism is established based on anomaly scores to determine the severity of potential anomalies. Specifically, anomaly scores below 30 are classified as mild; scores between 30 and 70 are classified as moderate; and scores above 70 are classified as severe. This grading mechanism allows for detailed classification of potential anomalies based on score ranges, enabling tiered responses and differentiated handling. The thresholds are set based on specific criteria. For example, historical anomaly scores approximate a normal distribution. First, the mean and standard deviation of all historical anomaly scores are calculated. Then, the anomaly scores are divided into multiple consecutive intervals. Analyzing the anomalies in different intervals, anomaly thresholds are determined. If the mean is 50 and the standard deviation is 20, then the mean minus the standard deviation equals 30. The sum of the mean and standard deviation is 70, yielding the threshold.
[0041] When the anomaly score is below 30, the corresponding anomaly characteristics are relatively similar to normal operating conditions and have no significant correlation with actual failure events. Therefore, this range can be reasonably considered as the range for judging mild anomalies. Secondly, when the anomaly score is between 30 and 70, the anomaly characteristics usually exhibit a certain degree of persistence or multi-dimensional cumulative effect, and show a significant non-linear correlation with performance degradation. If left unaddressed, there is a risk of gradually evolving into a serious failure. Therefore, this range is defined as moderate anomalies. Finally, when the anomaly score exceeds 70, the matching rate between its corresponding samples and historical major failure events exceeds 85%, and simulation environment testing verifies that it will cause irreversible functional failures to key functional modules. Therefore, this threshold is set as the criterion for judging severe anomalies.
[0042] Based on the initial classification of anomalies, amplitude feature values for each anomaly template are sequentially extracted from the corresponding anomaly template database, including peak value, local extrema, mean, and amplitude standard deviation. The absolute difference between the anomaly's amplitude feature value and the template's amplitude feature value is calculated and divided by the template's amplitude feature value to obtain the deviation value. The anomalies are further refined into the anomaly type with the smallest deviation value, resulting in the anomaly classification result. The anomaly template database is constructed by extracting amplitude feature values from various historical anomalies, calculating the average value as the feature value of the anomaly template, and storing them according to the severity of the anomalies. It also records the power management parameters used when the anomaly occurred and calculates their average value, resulting in the anomaly template database. For example, the anomaly template database contains {Anomaly type: Voltage descent; Peak value: 3.1V; Mean: 2.8V; Standard deviation: 0.25V; Data source: Statistics of 150 historical events}.
[0043] In step S15, obtaining preset power management parameters and real-time status information of the mobile device based on the anomaly classification result, and optimizing the preset power management parameters based on the real-time status information and preset status thresholds to obtain a real-time adjustment strategy, includes: Based on the anomaly classification results, preset power management parameters and anomaly status information are extracted from the preset anomaly classification template database. The real-time status information of the mobile device is obtained and compared with the abnormal status information. The data consistency between the real-time status information and the abnormal status information is analyzed to obtain a data consistency score. If the data consistency score exceeds a preset consistency threshold, the preset power management parameters are adjusted based on the real-time status information to obtain a real-time adjustment strategy.
[0044] It should be noted that, based on the anomaly classification results, anomaly status information and corresponding power management parameters are extracted from the preset anomaly classification template data. Specifically, the anomaly template database is created by extracting amplitude feature values from various historical anomalies, calculating their average values as feature values for the anomaly templates, and storing them according to the severity of the anomalies. Simultaneously, the power management parameters used when the anomaly occurred are recorded and their average values are calculated, resulting in the anomaly template database.
[0045] The real-time status information of the mobile device is acquired, and the absolute difference is calculated by matching it one-to-one with the status information of the abnormal template. This difference is then divided by the abnormal template status information to obtain the feature deviation. The average of all feature deviations is taken as the total deviation value. The reciprocal of the total deviation value is multiplied by 100 to convert it into a percentage score, which is the data consistency score.
[0046] For example, a mobile device collects real-time status information during operation, including four amplitude feature values: peak value 25V, mean value 18V, local extrema 22V, and amplitude standard deviation 3V. The corresponding abnormal template status information is: peak value 10V, mean value 8V, local extrema 10V, and amplitude standard deviation 1V. First, the difference between the real-time status and the template status for each dimension is calculated and divided by the template feature value to obtain the feature deviation. For example, the peak deviation is 1.5, the mean deviation is 1.25, the local extrema deviation is 1.2, and the amplitude standard deviation deviation is 2.0. Then, the feature deviations of all dimensions are averaged to obtain a total deviation value of approximately 1.4875.
[0047] The total deviation value is then converted into a percentage score to obtain the data consistency score. When the total deviation is 0, the consistency score is considered to be 100. A higher score indicates that the device status is closer to the template, while a lower score suggests that the device may have an anomaly. The formula for calculating the data consistency score is: in, The scaling factor, calibrated based on historical data, is set to 60 after analysis. For example, the total deviation is 1.4875, which translates to a consistency score of 10.75.
[0048] If the data consistency score is below 80, there is a certain deviation between the real-time anomaly of the mobile device and the anomaly template. The difference between the real-time status information and the anomaly template status information is divided by the anomaly template status information to obtain the relative deviation rate of each feature dimension of the status information. The average of each relative deviation rate is then used as the adjustment range for the power management parameters. Specifically, the power management parameters are multiplied by the adjustment range and added to the power management parameters to obtain the adjusted power management parameters, thus obtaining the real-time adjustment strategy. Statistical analysis of a large amount of historical operating data is conducted. Specifically, a normal distribution analysis is performed on all scores to calculate the mean, standard deviation, and cumulative distribution of scores under different operating states. The scores are matched with recorded power management anomaly events to calculate the proportion of anomaly events occurring within each score interval. By analyzing the relationship between the score distribution and the anomaly event matching rate, it is found that when the consistency score is above 80, the difference between the operating state and the anomaly template is small and usually does not have a substantial impact on device performance. However, when the score is below 80, the deviation of the device's real-time status increases significantly and has a high correlation with the occurrence of actual power management anomalies in simulation tests. Therefore, 80 is selected as the data consistency score threshold.
[0049] In step S16, the real-time adjustment strategy is validated using a preset support vector machine model, and the validation accuracy is calculated. If the validation accuracy is higher than a preset accuracy threshold, a fault detection report is generated, including: Obtain the mobile device's runtime data and runtime environment parameters; Using a preset support vector machine model, the verification accuracy of the real-time adjustment strategy is calculated based on the running data and the running environment parameters. The verification accuracy rate is compared with a preset accuracy threshold. If the accuracy rate exceeds the preset accuracy threshold, a fault detection report is generated.
[0050] It should be noted that the operation data of the mobile device and the corresponding operation environment parameters, including voltage, current and load rate, are obtained. Then, the above operation data and operation environment parameters are input into a preset support vector machine model, and the verification accuracy of the real-time adjustment strategy is calculated through the model.
[0051] The pre-defined support vector machine model includes an input layer, a kernel function mapping layer, and an output discriminant layer. The input layer receives a multi-dimensional vector consisting of amplitude features (including peak value, mean, local extrema, amplitude standard deviation, and RMS value), duration features, and environmental parameters (including voltage, current, temperature, and load rate). For each feature, the minimum value is subtracted from the maximum value of each sample point, and then divided by the difference between the maximum and minimum values. This transforms all feature values to the range of 0 to 1, eliminating differences in amplitude between different features and allowing them to be compared and combined on the same order of magnitude, thus eliminating differences in the magnitude of different features. The kernel function mapping layer performs feature transformation through radial basis functions. Specifically, for each input vector, the distance between the vector and each support vector in the training set is calculated sequentially as a similarity value. The width parameter of the kernel function controls the rate of similarity decay, the bias parameter adjusts the position of the classification boundary, and the regularization parameter balances training error and model complexity.
[0052] The system adjusts the above parameters using cross-validation. The search range is typically set as follows: regularization parameter [10−1, 100, 101, 102], kernel width parameter [10−3, 10−2, 10−1, 100]. Different parameter combinations are repeatedly tried on the training data, and the classification accuracy or regression error is compared. The parameter combination that achieves the best performance on the validation data is selected, thereby optimizing the kernel function and hyperparameters to ensure that the model can accurately determine the operating status of the equipment. The output discriminant layer generates classification results based on similarity values, obtaining the probability of each category.
[0053] The training process of the pre-defined support vector machine model is based on labeled historical sample data, which includes normal states and various known fault states. Through an optimization process that minimizes structural risk, the model can learn and determine the optimal classification hyperplane during the training phase. The sample data consists of 100,000 state sampling points collected by the system from the long-term operating records of the mobile device. Normal state samples account for 70% of the total data, and known fault state samples account for 30%, ensuring that the model can fully learn the characteristics of normal operating modes while covering various possible fault types. Each sample data point contains multidimensional information including amplitude features, frequency features, duration features, and environmental parameters, and is labeled with its corresponding state category.
[0054] During use, the operational data and environmental parameters corresponding to the real-time adjustment strategy to be verified are input into a pre-trained support vector machine (SVM) model to obtain the model's discriminative output. If the anomaly classification with the highest probability in the discriminative output matches the anomaly classification result, the predicted probability of the anomaly classification is used as the verification accuracy of the real-time adjustment strategy. Specifically, the operational data and environmental parameters corresponding to the strategy to be verified are organized into multi-dimensional feature vectors according to the sampling order. Each vector contains amplitude features, frequency features, duration features, and environmental parameter information. Then, each feature vector is input into the pre-trained SVM model to obtain the predicted probability value of each anomaly category, and the category with the highest probability is selected as the model's judgment result. During comparison, the system calculates the difference between the category with the highest probability output by the model, the average feature value of this type of anomaly in history, and the feature value corresponding to the strategy to be verified, and divides the difference by the average feature value to obtain the relative deviation value. If the relative deviation value is less than 5%, it is considered consistent, and the maximum probability value is recorded as the verification score of the sample. In the backtesting, it was found that when the relative deviation between the predicted category and the actual anomaly category was less than 5%, the corresponding prediction probability accuracy was higher than 80%, and the matching rate with the actual fault event was significantly improved, reliably reflecting the effectiveness of the strategy. However, when the relative deviation was greater than 5%, even if some features were similar, it might lead to false alarms or false negatives. Therefore, only when the relative deviation was less than 5% was the verification score included to ensure the accuracy of the evaluation and the reliability of the engineering. The system then averaged the single-sample verification scores of all sampling points to obtain the overall verification accuracy of the real-time adjustment strategy, thereby quantifying the effectiveness of the strategy in fault detection. If the verification accuracy exceeded 90%, the fault detection was considered correct and a fault detection report was generated. The 90% accuracy threshold was selected because when the accuracy reached 90% or higher, the matching rate between the detection results and the actual fault records was significantly improved, meeting the high reliability requirements of engineering applications. Secondly, from the perspective of safety and robustness, 90% as a critical value can achieve a balance between detection sensitivity and stability, ensuring the timely detection of potential faults while avoiding overly conservative judgments due to excessively high thresholds. Finally, in the comparative experiment between simulation test and actual application scenario, when the accuracy threshold was set to 90%, the consistency between the fault detection report and the manual review result exceeded 92%, while the false alarm rate was controlled within 10%, indicating that the threshold has high engineering feasibility and practical rationality.
[0055] In summary, this invention provides a mobile terminal power failure detection method. By performing unified preprocessing and refined feature extraction on multi-source real-time data, it retains key dynamic features while eliminating redundant and noise information, enabling the input data to more realistically reflect the core state of the transportation environment and solving the problem of insufficient accuracy in mobile terminal power failure detection under dynamic usage scenarios.
[0056] Reference Figure 2 The second embodiment of the present invention provides a mobile terminal power failure detection system, comprising: The data acquisition module is used to acquire polar signal data and preprocess it to obtain a polar signal sequence; The feature extraction module is used to obtain the amplitude value and duration value of each time point in the polar signal sequence and extract features to obtain amplitude change trajectory features and duration features, and generate a high-dimensional vector output. An anomaly labeling module is used to calculate the Euclidean distance of the high-dimensional vector output. If the Euclidean distance exceeds a preset distance threshold, it is labeled as a potential anomaly, and an anomaly candidate list is obtained. An anomaly classification module is used to obtain the amplitude change trajectory features and duration features of each potential anomaly based on the anomaly candidate list, calculate the anomaly score of the potential anomaly according to a preset anomaly judgment mechanism, compare the anomaly score with a preset anomaly score threshold, confirm the anomaly type, and obtain the anomaly classification result. The adjustment strategy generation module is used to obtain preset power management parameters and real-time status information of the mobile device based on the anomaly classification results, optimize the preset power management parameters based on the real-time status information and preset status thresholds, and obtain a real-time adjustment strategy. The fault report generation module is used to verify the real-time adjustment strategy using a preset support vector machine model, calculate the verification accuracy, and generate a fault detection report if the verification accuracy is higher than a preset accuracy threshold.
[0057] It should be noted that the mobile power failure detection device provided in this embodiment of the invention is used to execute all the process steps of the mobile power failure detection method in the above embodiment. The working principle and beneficial effects of the two are one-to-one, so they will not be described again.
[0058] This invention also provides an electronic device. The electronic device includes a processor, a memory, and a computer program stored in the memory and executable on the processor, such as a data acquisition program. When the processor executes the computer program, it implements the steps described in the various embodiments of the mobile terminal power failure detection method, for example... Figure 1 The step S11 shown. Alternatively, when the processor executes the computer program, it implements the functions of each module in the above-described device embodiments, such as the data acquisition module.
[0059] For example, the computer program may be divided into one or more modules, which are stored in the memory and executed by the processor to complete the present invention. The one or more modules may be a series of computer program instruction segments capable of performing a specific function, which describe the execution process of the computer program in the electronic device.
[0060] The electronic device may be a desktop computer, laptop, handheld computer, or smart tablet, etc. The electronic device may include, but is not limited to, a processor and memory. Those skilled in the art will understand that the above components are merely examples of electronic devices and do not constitute a limitation on the electronic device. It may include more or fewer components than described above, or combine certain components, or different components. For example, the electronic device may also include input / output devices, network access devices, buses, etc.
[0061] The processor can be a Central Processing Unit (CPU), or 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 can be a microprocessor or any conventional processor. The processor is the control center of the electronic device, connecting all parts of the electronic device via various interfaces and lines.
[0062] The memory can be used to store the computer programs and modules. The processor implements various functions of the electronic device by running or executing the computer programs and modules stored in the memory and by calling data stored in the memory. The memory may mainly include a program storage area and a data storage area. The program storage area may store the operating system, at least one application program required for a function (such as sound playback function, image playback function, etc.), etc.; the data storage area may store data created according to the use of the mobile phone (such as audio data, phonebook, etc.). In addition, the memory may include high-speed random access memory, and may also include non-volatile memory, such as hard disk, RAM, plug-in hard disk, smart media card (SMC), secure digital (SD) card, flash card, at least one disk storage device, flash memory device, or other volatile solid-state storage device.
[0063] If the modules integrated into the electronic device are implemented as software functional units and sold or used as independent products, they can be stored in a computer-readable storage medium. Based on this understanding, all or part of the processes in the methods of the above embodiments can also be implemented by a computer program instructing related hardware. The computer program can be stored in a computer-readable storage medium, and when executed by a processor, it can implement the steps of the various method embodiments described above. The computer program includes computer program code, which can be in the form of source code, object code, executable files, or certain intermediate forms. The computer-readable medium can include: any entity or device capable of carrying the computer program code, recording media, USB flash drives, portable hard drives, magnetic disks, optical disks, computer memory, read-only memory (ROM), random access memory (RAM), electrical carrier signals, telecommunication signals, and software distribution media, etc. It should be noted that the content included in the computer-readable medium can be appropriately added or removed according to the requirements of legislation and patent practice in the jurisdiction. For example, in some jurisdictions, according to legislation and patent practice, computer-readable media do not include electrical carrier signals and telecommunication signals.
[0064] It should be noted that the device embodiments described above are merely illustrative. The units described as separate components may or may not be physically separate, and the components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the modules can be selected to achieve the purpose of this embodiment according to actual needs. Furthermore, in the accompanying drawings of the device embodiments provided by this invention, the connection relationships between modules indicate that they have communication connections, which can be specifically implemented as one or more communication buses or signal lines. Those skilled in the art can understand and implement this without any creative effort.
[0065] The specific embodiments described above further illustrate the purpose, technical solution, and beneficial effects of the present invention. It should be understood that the above descriptions are merely specific embodiments of the present invention and are not intended to limit the scope of protection of the present invention. In particular, it should be noted that any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the scope of protection of the present invention for those skilled in the art.
Claims
1. A method for detecting power failure in a mobile terminal, characterized in that, include: The polarity signal data is acquired and preprocessed to obtain the polarity signal sequence; The amplitude and duration values at each time point in the polar signal sequence are obtained and features are extracted to obtain amplitude change trajectory features and duration features, and a high-dimensional vector output is generated. Calculate the Euclidean distance of the high-dimensional vector output. If the Euclidean distance exceeds a preset distance threshold, it is marked as a potential anomaly, and an anomaly candidate list is obtained. Based on the anomaly candidate list, the amplitude change trajectory features and duration features of each potential anomaly are obtained. An anomaly score of the potential anomaly is calculated according to a preset anomaly judgment mechanism. The anomaly score is compared with a preset anomaly score threshold to confirm the anomaly type and obtain the anomaly classification result. Based on the anomaly classification results, preset power management parameters and real-time status information of the mobile device are obtained. Based on the real-time status information and preset status thresholds, the preset power management parameters are optimized to obtain a real-time adjustment strategy. The real-time adjustment strategy is validated using a preset support vector machine model, and the validation accuracy is calculated. If the validation accuracy is higher than a preset accuracy threshold, a fault detection report is generated.
2. The mobile terminal power failure detection method according to claim 1, characterized in that, The process of acquiring and preprocessing polarity signal data to obtain a polarity signal sequence includes: Acquire polarity signal data; The polarity signal data is converted into a digital signal to obtain the original signal sequence; The features of the original signal sequence are extracted and noise is suppressed to obtain a denoised signal sequence. If the denoised signal sequence has fluctuations, then the denoised signal sequence is smoothed to obtain a polar signal sequence.
3. The mobile terminal power supply fault detection method according to claim 1, characterized in that, The amplitude and duration values at each time point in the polar signal sequence are obtained and features are extracted to obtain amplitude change trajectory features and duration features, generating... High-dimensional vector output, including: Obtain the amplitude and duration values at each time point in the polar signal sequence to obtain a feature set; The feature set is traversed point by point, and the amplitude change trajectory features and duration features are extracted according to the preset convolution kernel to obtain the initial feature map; The initial feature map is subjected to dimensionality reduction processing using a preset pooling layer to obtain a dimensionality-reduced feature map; The reduced-dimensional feature map is used to generate a high-dimensional vector output by calculating a preset activation function.
4. The mobile terminal power supply fault detection method according to claim 1, characterized in that, The process involves calculating the Euclidean distance of the high-dimensional vector output. If the Euclidean distance exceeds a preset distance threshold, it is marked as a potential anomaly, resulting in an anomaly candidate list, including: Obtain the preset high-dimensional vector of the normal state; Calculate the Euclidean distance between the high-dimensional vector output and the high-dimensional vector in the normal state; If the Euclidean distance exceeds a preset distance threshold, it is marked as a potential anomaly, and the state characteristics of the potential anomaly are recorded. By integrating the aforementioned state features, an anomaly candidate list is obtained.
5. The mobile terminal power supply fault detection method according to claim 1, characterized in that, Based on the candidate anomaly list, the amplitude change trajectory features and duration features of each potential anomaly are obtained. An anomaly score is calculated for each potential anomaly according to a preset anomaly judgment mechanism. The anomaly score is compared with a preset anomaly score threshold to confirm the anomaly type and obtain an anomaly classification result, including: Obtain the amplitude change trajectory features and duration features of each potential anomaly in the anomaly candidate list to obtain an anomaly feature set; The abnormal feature set is weighted to obtain an abnormal score, and the abnormal score is compared with a preset abnormal score threshold to obtain a preliminary abnormal classification set. The preliminary anomaly classification set is optimized to obtain the anomaly classification results.
6. The mobile terminal power supply fault detection method according to claim 1, characterized in that, The step of obtaining preset power management parameters and real-time status information of the mobile device based on the anomaly classification result, and optimizing the preset power management parameters based on the real-time status information and preset status thresholds to obtain a real-time adjustment strategy includes: Based on the anomaly classification results, preset power management parameters and anomaly status information are extracted from the preset anomaly classification template database. The real-time status information of the mobile device is obtained and compared with the abnormal status information. The data consistency between the real-time status information and the abnormal status information is analyzed to obtain a data consistency score. If the data consistency score exceeds a preset consistency threshold, the preset power management parameters are adjusted based on the real-time status information to obtain a real-time adjustment strategy.
7. The mobile terminal power supply fault detection method according to claim 1, characterized in that, The real-time adjustment strategy is validated using a preset support vector machine model, and the validation accuracy is calculated. If the validation accuracy is higher than a preset accuracy threshold, a fault detection report is generated, including: Obtain the mobile device's runtime data and runtime environment parameters; Using a preset support vector machine model, the verification accuracy of the real-time adjustment strategy is calculated based on the running data and the running environment parameters. The verification accuracy rate is compared with a preset accuracy threshold. If the accuracy rate exceeds the preset accuracy threshold, a fault detection report is generated.
8. A mobile terminal power supply fault detection system, characterized in that, include: The data acquisition module is used to acquire polar signal data and preprocess it to obtain a polar signal sequence; The feature extraction module is used to obtain the amplitude value and duration value of each time point in the polar signal sequence and extract features to obtain amplitude change trajectory features and duration features, and generate a high-dimensional vector output. An anomaly labeling module is used to calculate the Euclidean distance of the high-dimensional vector output. If the Euclidean distance exceeds a preset distance threshold, it is labeled as a potential anomaly, and an anomaly candidate list is obtained. An anomaly classification module is used to obtain the amplitude change trajectory features and duration features of each potential anomaly based on the anomaly candidate list, calculate the anomaly score of the potential anomaly according to a preset anomaly judgment mechanism, compare the anomaly score with a preset anomaly score threshold, confirm the anomaly type, and obtain the anomaly classification result. The adjustment strategy generation module is used to obtain preset power management parameters and real-time status information of the mobile device based on the anomaly classification results, optimize the preset power management parameters based on the real-time status information and preset status thresholds, and obtain a real-time adjustment strategy. The fault report generation module is used to verify the real-time adjustment strategy using a preset support vector machine model, calculate the verification accuracy, and generate a fault detection report if the verification accuracy is higher than a preset accuracy threshold.