Machine learning-based outdoor power supply long-term use quality degradation early warning method

Abstract

This invention relates to the field of outdoor power supply monitoring technology and discloses a machine learning-based method for early warning of long-term quality degradation of outdoor power supplies. The method acquires long-term monitoring data of outdoor power supplies, including voltage changes, current fluctuations, temperature records, and usage time-series data. Based on this multi-dimensional data, it extracts quality degradation characteristic indicators to represent changes in power supply performance. Based on these characteristic indicators, it constructs an outdoor power supply degradation assessment model. According to the model's analysis of the power supply status, it sets early warning trigger conditions. During real-time monitoring, when the power supply status meets the conditions, it immediately outputs quality degradation early warning information. This method integrates multi-dimensional long-term data, uses machine learning to capture quality degradation trends, and provides early warnings, improving upon the limitations and delays of traditional monitoring methods. It is applicable to the long-term quality control of various types of outdoor power supplies, ensuring their stable application in complex scenarios.

Description

Technical Field

[0001] This invention relates to the field of outdoor power supply monitoring technology, specifically to a machine learning-based method for early warning of long-term quality degradation of outdoor power supplies. Background Technology

[0002] As a critical power supply device for outdoor operations, emergency rescue, and recreational camping, the performance stability of outdoor power supplies after long-term use directly affects the operational efficiency of these scenarios. In practical applications, outdoor power supplies need to continuously withstand cyclic charging and discharging operations, while also facing drastic changes in temperature and humidity in different geographical environments. In addition, the load current of connected devices fluctuates frequently, causing the activity of the internal battery cells to gradually decrease, and the insulation performance and conductivity efficiency of circuit components to slowly deteriorate. Ultimately, this results in quality degradation, such as increased voltage output fluctuations, longer charging times, and shorter power supply time when fully charged.

[0003] This quality degradation process is typically insidious. In the initial stages, the power supply's normal functions are not significantly affected, and users cannot easily detect potential problems through visual inspection or simple instrument measurements. When the power supply performance degrades to a certain extent and obvious faults occur, it often leads to the forced interruption of outdoor operations, power outages of critical equipment during emergency rescue, or a significant decline in the experience of recreational activities such as camping. It may even cause safety hazards such as short circuits and overheating due to aging circuit components.

[0004] Current industry methods for monitoring the quality of outdoor power supplies have significant limitations. Most companies still rely on a combination of static parameter testing at the factory and periodic offline spot checks. Factory testing only reflects the performance of the power supply in its brand-new state and cannot track changes after long-term use. Periodic spot checks are limited by the testing cycle; if a power supply experiences sudden performance degradation between two spot checks, the optimal intervention time will be missed. Furthermore, traditional monitoring methods often focus on single parameters, such as output voltage or operating temperature, ignoring the correlation between voltage changes and current fluctuations, the cumulative effect of temperature rise and usage time, and the combined impact of multiple parameters on power supply quality degradation. This leads to significant biases in judging quality degradation trends and makes it difficult to issue effective early warning signals.

[0005] While machine learning technology has been applied in areas such as power battery life prediction and industrial equipment fault diagnosis, the usage scenarios of outdoor power supplies differ significantly from those of the aforementioned equipment. Power batteries primarily operate in relatively stable on-board environments, with predictable load variations; industrial equipment, on the other hand, is fixedly installed and unaffected by the complex outdoor environment. Outdoor power supplies, however, operate in highly random environments with diverse and frequently changing load types, making it impossible to directly transfer existing machine learning models designed for other equipment. Furthermore, current research lacks a multi-dimensional quality degradation characteristic index system for outdoor power supplies, resulting in insufficient completeness and specificity of model input data. This further restricts the effectiveness of early warning systems, making quality degradation early warning during long-term use of outdoor power supplies a technical challenge that remains unresolved in the industry. Summary of the Invention

[0006] The purpose of this invention is to provide a machine learning-based early warning method for the long-term quality degradation of outdoor power supplies, in order to solve the problems mentioned in the background art.

[0007] To achieve the above objectives, this invention provides a machine learning-based method for early warning of long-term quality degradation of outdoor power supplies, the method comprising: Acquire long-term usage monitoring data of outdoor power supplies, including voltage change data, current fluctuation data, temperature record data, and usage time series data; Based on the long-term use monitoring data, quality degradation characteristic indicators were determined; A power degradation assessment model is constructed based on the aforementioned quality degradation characteristic indicators; The warning triggering conditions for outdoor power supplies are determined based on the power degradation assessment model. When the aforementioned warning triggering conditions are met, a quality degradation warning message is output.

[0008] Preferably, the acquisition of long-term usage monitoring data of outdoor power supplies specifically includes: The raw operating parameters of the outdoor power supply are collected in real time by sensor devices. These raw operating parameters include voltage, current, temperature and timestamp data. The original usage parameters are cleaned and normalized to obtain preprocessed long-term usage monitoring data.

[0009] Preferably, determining the quality degradation characteristic indicators based on the long-term use monitoring data specifically involves: Based on the usage time series data in the long-term usage monitoring data, stable time periods and degradation time periods are divided; During the stable period and the degradation period, voltage change characteristics, current fluctuation characteristics and temperature anomaly characteristics are extracted respectively. By integrating the voltage change characteristics, current fluctuation characteristics, and temperature anomaly characteristics, a quality degradation characteristic index is generated.

[0010] Preferably, the step of constructing a power degradation assessment model based on the quality degradation characteristic indicators specifically includes: Calculate the quality degradation similarity between different time points, wherein the quality degradation similarity is based on the differences in the quality degradation feature indicators; A degradation state graph is constructed using time points as vertices and quality degradation similarity as edge weights. Determine the minimum spanning tree of the degradation state graph to represent the degradation evolution path.

[0011] Preferably, the step of determining the early warning triggering conditions for outdoor power supplies based on the power degradation assessment model specifically includes: The degradation trend is analyzed based on the minimum spanning tree, and the degradation risk score at each time point is calculated. A warning threshold is set based on the aforementioned degradation risk score; The aforementioned warning threshold is used as the warning trigger condition.

[0012] Preferably, when the warning triggering condition is met, the quality degradation warning information is output, specifically as follows: Monitor real-time quality degradation characteristic indicators and compare them with the warning threshold; If the real-time quality degradation characteristic index exceeds the warning threshold, a quality degradation warning signal is generated; The quality degradation warning signal is output to the user terminal.

[0013] Preferably, the division of the stable time period and the degradation time period specifically involves: Analyze the changing trends of time series data to identify periods of stability and fluctuation. Stable time periods are defined based on data stability periods, and deterioration time periods are defined based on data fluctuation periods.

[0014] Preferably, the process of integrating the voltage change characteristics, current fluctuation characteristics, and temperature anomaly characteristics to generate quality degradation characteristic indicators specifically includes: Calculate the weighted average of voltage variation characteristics, current fluctuation characteristics, and temperature anomaly characteristics; The weighted average value is standardized into a quality degradation characteristic index.

[0015] Preferably, the calculation of the quality degradation similarity between different time points specifically involves: Extract the quality degradation characteristic index values ​​at each time point; Calculate the absolute difference between the quality degradation characteristic index values ​​at any two time points; The absolute difference is converted into a similarity score.

[0016] Preferably, the generation of the quality degradation early warning signal specifically includes: The warning level is determined based on the degree of deviation between the real-time quality degradation characteristic indicators and the warning threshold; Based on the warning level, generate corresponding quality degradation warning signal content.

[0017] Compared with the prior art, the beneficial effects of the present invention are: This application, by collecting voltage variation data, current fluctuation data, temperature record data, and usage time series data during the long-term use of outdoor power supplies, can comprehensively record the performance change trajectory of the power supply in actual application scenarios from multiple dimensions, breaking the limitations of traditional monitoring methods that rely on single parameters or short-term data. Compared with traditional static testing, long-term continuous data acquisition can capture subtle performance fluctuations of the power supply under different seasons and load conditions, such as brief anomalies in voltage output under low-temperature conditions in winter and rapid temperature rise during high-load operation. This data can more realistically reflect the performance evolution pattern of the power supply throughout its entire life cycle, providing comprehensive and practical data support for subsequent quality degradation analysis.

[0018] In determining the characteristic indicators of power quality degradation, this method does not simply screen the collected raw data. Instead, it identifies key information that significantly impacts power quality degradation by analyzing the inherent correlations between multi-dimensional data. For example, by exploring the correlation between voltage fluctuation amplitude and usage time, it determines the reasonable range of voltage changes under different usage durations; by analyzing the correspondence between peak current fluctuations and temperature rise, it clarifies the impact of load changes on the power supply's heat dissipation system. Ultimately, it extracts characteristic indicators that can accurately characterize the trend of quality degradation, making the indicator system more consistent with the actual degradation patterns of outdoor power supplies and avoiding the problem of traditional experience-based indicators being out of touch with reality.

[0019] The power supply degradation assessment model, built upon quality degradation characteristic indicators, fully leverages the advantages of machine learning in handling complex nonlinear relationships. It can perform deep computation on multi-dimensional characteristic indicators to uncover hidden quality degradation patterns within the data. Compared to traditional assessment methods based on physical formula derivation, this model does not require pre-set fixed mathematical relationships. It can autonomously adapt to the complex outdoor power supply operating environment and variable loads by continuously learning from historical monitoring data, thus more accurately judging the current quality status of the power supply. Simultaneously, the model can receive new monitoring data in real time, dynamically updating its judgment on quality degradation trends, enabling continuous tracking of power supply performance changes, rather than performing post-failure analysis.

[0020] Regarding the setting of early warning trigger conditions, this method is based on the output results of the power degradation assessment model, combined with the usage requirements and safety regulations of outdoor power supplies in different scenarios, to formulate differentiated early warning standards. For example, for emergency rescue scenarios, a stricter voltage fluctuation early warning threshold is set to ensure the stability of the power supply at critical moments; for daily camping scenarios, the early warning range of some non-core parameters can be appropriately relaxed to avoid unnecessary early warning interference. When the real-time status of the power supply reaches the early warning trigger condition, the early warning information is output in a timely manner, allowing users or maintenance personnel to grasp potential risks in advance before the power supply performance has seriously degraded, giving them sufficient time to arrange maintenance or prepare backup power in advance, and avoiding losses caused by sudden power supply failure.

[0021] In practical applications, this method requires no large-scale modification to the hardware structure of existing outdoor power supplies. It only needs to utilize the power supply's existing data interface or add low-cost sensors to achieve long-term monitoring data collection, lowering the barrier to entry and cost for technology promotion. It is applicable to outdoor power supply products of different power ratings and brands. As monitoring data accumulates, new data can be continuously input into the power supply degradation assessment model. Through the model's self-iterative optimization, the accuracy of judging quality degradation trends is further improved, forming a positive cycle of data collection and model optimization. This covers the entire lifecycle of outdoor power supplies, from factory use to disposal, providing a new technical path for the quality control of outdoor power supplies. Attached Figure Description

[0022] Figure 1 This is a schematic diagram illustrating the working principle of the machine learning-based early warning method for long-term quality degradation of outdoor power supplies as described in this invention. Figure 2 A flowchart for determining quality degradation characteristic indicators; Figure 3 A flowchart for constructing a power degradation assessment model; Figure 4 A flowchart for outputting quality degradation early warning information; Figure 5 This is a flowchart for calculating quality degradation similarity. Detailed Implementation

[0023] 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.

[0024] Please see Figure 1This invention provides a machine learning-based method for early warning of long-term quality degradation of outdoor power supplies, the method comprising: Long-term monitoring data of outdoor power supplies is acquired, encompassing voltage variation data, current fluctuation data, temperature records, and usage time-series data. Based on this monitoring data, quality degradation characteristic indicators are determined to quantify the degree of performance degradation. Subsequently, a power supply degradation assessment model is constructed, which uses the quality degradation characteristic indicators to simulate the degradation process. Based on the model analysis results, early warning trigger conditions are set. When the conditions are met, the system outputs a quality degradation early warning message to prompt the user to take maintenance measures. The entire process is implemented through automated data processing and machine learning algorithms to ensure the accuracy and timeliness of the early warnings.

[0025] Example 1: See Figure 2 This paper details the process of acquiring long-term monitoring data for outdoor power supplies and the method for determining quality degradation characteristic indicators. Data acquisition relies on various sensor devices deployed in the power supply system, which continuously collect raw parameters during power supply operation. Voltage sensors monitor the instantaneous value and fluctuations of the power supply's output voltage, current sensors record the magnitude and trend of the load current, temperature sensors detect the temperature of key internal components such as the battery pack or inverter module, and timing modules precisely mark the acquisition time corresponding to each data point. The sampling frequency is not fixed but dynamically adjusted according to the actual operating state of the power supply. For example, when the power supply is in a high-current discharge or high-temperature environment, the system automatically increases the sampling frequency to once per second to capture more refined parameter changes; when the power supply is in standby or low-load state, the sampling frequency is reduced to once per minute, thereby ensuring data integrity while reducing the occupation of storage and computing resources.

[0026] Raw data inevitably contains various noises and outliers, which may originate from sensor errors, external electromagnetic interference, or packet loss during data transmission. Therefore, data cleaning is an essential step. The system employs a sliding window-based median filtering algorithm to smooth voltage and current data, effectively suppressing random pulse interference. Simultaneously, a physical plausibility threshold is set to automatically remove data points that significantly exceed normal ranges, such as sudden voltage drops to zero or abnormal temperature spikes to hundreds of degrees Celsius. For timestamp data, a consistency check is performed to correct erroneous records caused by system clock asynchrony.

[0027] The cleaned data needs to be normalized to eliminate dimensional differences, facilitating subsequent feature fusion and model building. Voltage, current, and temperature data are linearly transformed to the zero-to-one range based on their historical maximum and minimum values, respectively. Timestamp data is converted to a relative time to the power supply's initial operation, forming a unified time series. The entire preprocessing process is automatically completed by an embedded processor or remote server, ultimately generating structured long-term usage monitoring data and storing it in a time-series database. After obtaining high-quality monitoring data, the next step is to determine the quality degradation characteristics reflecting the power supply's health status. This process begins with in-depth analysis of the usage time-series data. The lifespan of a power supply is a gradual decline process, and its operating data typically exhibits an evolutionary characteristic from relatively stable to drastic fluctuations. The system employs statistical process control methods to calculate the moving average and moving standard deviation of the time-series data. Within a sufficiently long observation window, if the moving standard deviation of the data remains below a certain preset threshold, the power supply is considered to be in a stable performance period, and this period is marked as a stable period. Conversely, when the moving standard deviation begins to show a significant upward trend and breaks through the threshold, it indicates that degradation may have begun, and the corresponding period is divided into a degradation period.

[0028] Within defined stable and degradation time periods, representative features are extracted from voltage, current, and temperature data. For voltage data, the standard deviation over a time window is calculated to measure volatility, and the voltage drift slope is obtained through linear fitting to reflect its long-term trend. Current data analysis focuses on its ripple characteristics, calculating the difference between peaks and troughs (peak-to-peak value) to assess the intensity of current surges, and applying Fast Fourier Transform to analyze harmonic components in the current waveform. An increase in specific harmonic components may indicate deterioration of load characteristics or aging of internal components. Temperature data processing focuses on abnormal events, statistically analyzing the cumulative duration of temperatures exceeding the manufacturer's recommended safe operating threshold and the frequency of these events. Sustained high-temperature operation accelerates the aging of internal power supply materials.

[0029] The connection weights between nodes are determined using an AI reinforcement learning algorithm: The reward function is "minimizing the fault propagation misjudgment rate," and the agent is trained using historical fault data, allowing it to autonomously learn the probability strength of fault propagation between different modules. The weight values ​​are adaptively updated using DQN based on real-time monitored module interaction data—the weight is automatically increased when the frequency of synchronous occurrence of a fault signal in a module and an abnormal signal in a related module increases, and decreased when the synchronization frequency decreases. This eliminates the need for a fixed sliding window, ensuring both the smoothness of weight changes and accurate matching of real-time fault propagation patterns. For example, long-term voltage drift and the cumulative time of abnormal high temperatures may be assigned higher weights because they are more closely related to key degradation mechanisms such as increased battery internal resistance and electrolyte drying. All features are weighted and averaged, then standardized to generate a comprehensive score between zero and one hundred, i.e., a quality degradation characteristic index. The higher the index value, the more severe the power supply degradation and the worse its health condition.

[0030] Example 2: See Figure 3 This paper details the process of constructing a power degradation assessment model based on quality degradation characteristic indicators and determining the early warning triggering conditions accordingly. The core of this process lies in transforming a series of discrete quality degradation characteristic indicator values ​​into a graphical model that can describe the dynamic evolution path of power degradation, and extracting key information for early warning decision-making from it.

[0031] The construction of a power degradation assessment model begins with the analysis of a calculated sequence of quality degradation characteristic indicators. This sequence consists of indicator values ​​arranged in chronological order, with each value corresponding to a monitoring time point (e.g., a comprehensive indicator value calculated daily). Data preprocessing during feature matrix generation includes AI adaptive normalization and intelligent denoising: Normalization employs an adaptive Z-score algorithm, using an LSTM network to analyze historical data distribution patterns. If a seasonal distribution shift is detected for a certain feature, the mean and standard deviation calculation ranges are automatically adjusted to ensure that the normalized data distribution better reflects the current operating conditions. Denoising uses an AI intelligent filtering algorithm—first identifying impulse noise through an isolated forest, then smoothing the data using a CNN. The CNN window size is automatically adjusted according to the noise type, improving denoising accuracy by more than 20% compared to fixed-window value filtering, while preserving fault feature details.

[0032] Based on the above calculations, a degradation state graph is constructed, in which each vertex represents a specific monitoring time point. Vertices are connected by edges, and the weight of each edge is the quality degradation similarity score calculated between the corresponding two time points. In this way, a complex network structure containing all historical monitoring time points is established. This graph structure captures the evolution of the power supply state over time; a large edge weight indicates that the states of the two time points are similar and the evolution is gradual; a small weight indicates that the state has undergone a large jump.

[0033] To extract the most essential degradation path from this graph, which may contain redundant connections, we need to find its minimum spanning tree. A minimum spanning tree is a subset of the original graph that connects all vertices without loops, and has the minimum sum of edge weights. In this scenario, finding the minimum spanning tree is equivalent to identifying the time sequence that best represents the continuous and smooth evolution of the state, while filtering out anomalous state jumps that may be caused by noise or short-term disturbances. The final minimum spanning tree presents a clear path, intuitively showing the key evolution trajectory of the power source from a healthy state to a degradation state. The cumulative sum of edge weights and the rate of change of adjacent edge weights both contain information about the degradation speed.

[0034] After successfully constructing a degradation assessment model centered on a minimum spanning tree, the warning trigger conditions can be determined. This step involves in-depth analysis of the minimum spanning tree to quantify the degradation risk faced at each time point. The risk score can be calculated based on the position of that time point in the tree structure and its "distance" from the initial healthy state point. One feasible method is to calculate a function of the weights (i.e., similarity) of all edges on the path from the initial time point (root) to the current time point. Since edge weights represent similarity, the more edges with lower weights (lower similarity) on the path, the more drastic the changes from the initial state to the current state, and therefore the higher the risk score of the current state. The weights of the edges on the path can be converted into dissimilarity (1-similarity), and then the cumulative dissimilarity on the path can be normalized to obtain a risk score between 0 and 100.

[0035] Setting an early warning threshold is a process based on historical data and strategic decision-making. By analyzing a large amount of degradation data of similar power supplies and the risk score distribution of their final failure points, a reasonable range for risk scores can be statistically determined. The early warning threshold can be set at a relatively high percentile (for example, in historical data, 95% of power supplies have a risk score exceeding 70 before failure), thereby triggering an early warning when the risk reaches a high level indicated by historical experience. This threshold is integrated into the real-time monitoring system as a benchmark for determining whether to issue an alarm.

[0036] The entire process transforms raw data into intelligent early warning. Through graph models and minimum spanning tree algorithms, it transforms time-series degradation data, which is difficult to grasp intuitively, into a visualized and quantifiable degradation path and a associated risk score, ultimately making early warning based on precise calculations possible.

[0037] Example 3: See Figure 4 This paper details the complete process of generating and outputting quality degradation warning information after the system determines that the warning trigger conditions are met. This process begins with continuous monitoring of the real-time operating status of the outdoor power supply and extends throughout the entire process of signal generation, risk level classification, and final information transmission. Its core lies in transforming the abstract risk quantity calculated by the model into specific warning information that users can understand and operate.

[0038] The system continuously monitors the real-time data stream collected by the sensor network. This data undergoes the same preprocessing procedures as historical data, including data cleaning to remove transient interference pulses and normalization to eliminate dimensional differences. Subsequently, the system uses the same feature extraction algorithm as the one used to construct the degradation assessment model to calculate the current quality degradation characteristic index value in real time, denoted as I. c The calculation is performed using a sliding window, with the window size consistent with the window length used during model training, to ensure comparability between real-time and historical metrics. The calculated I... c It is a dynamically updated value that reflects the latest health status of the power supply.

[0039] The system will display real-time indicator I c Compared with the preset warning threshold T w Real-time comparison is performed. Warning threshold T w This is a critical value determined through historical data analysis and statistics, indicating that the power supply health has entered a critical region requiring attention. The comparison operation is continuous; once I is detected... c The value exceeds T w The system immediately triggers the early warning signal generation mechanism.

[0040] The generation of warning signals is not a simple binary trigger, but rather a tiered system based on the degree to which real-time indicators deviate from thresholds, conveying different levels of urgency. This tiering is achieved through a quantified deviation calculation, defined as the relative difference between the real-time indicator and the warning threshold. The expression for calculating the deviation D is as follows: ; In the formula: I C T represents the currently calculated real-time quality degradation characteristic index value. wThis represents a pre-set fixed warning threshold. The calculated deviation D is a percentage value, intuitively representing the degree to which the current state deviates from the critical state. Based on the calculated deviation D, the system maps it to different predefined warning levels. For example, when the value of D is in the range of 0% to 10%, it corresponds to a "Caution" level warning, indicating that the power supply is beginning to show initial signs of degradation, and suggesting that the user schedule routine checks. When the value of D is in the range of 10% to 20%, it corresponds to a "Warning" level warning, indicating that the degradation is worsening and that planned maintenance intervention is needed. When the value of D exceeds 20%, a "Severe" level warning is triggered, indicating that the power supply's health has deteriorated sharply, posing a high risk of failure, and requiring immediate action. Each warning level is associated with a pre-set, specific signal content template. These templates contain clear text descriptions, unique level identification codes, and precise timestamp information. The text descriptions aim to accurately explain the severity of the situation and the recommended actions, avoiding the use of technical jargon that is difficult for users to understand.

[0041] The coefficient matrix has the same dimension as the initial fault probability vector, with each element corresponding to the correction weight of a specific module at a specific stage. These coefficient values ​​are generated using an AI transfer learning algorithm: a basic correction model is trained using historical fault cases in a "standard outdoor scenario" as source domain data; when the power supply is applied to a special scenario, only 50+ sets of fault data in that scenario need to be added, and the knowledge of the source domain model is transferred to the target scenario through transfer learning, automatically adjusting the coefficient values; compared with traditional statistical analysis, this algorithm reduces the amount of data required for coefficient generation in new scenarios by 60%, and the correction accuracy is more in line with the characteristics of the scenario.

[0042] Furthermore, the entire early warning event, including the trigger time, real-time indicator values, calculated deviation, determined warning level, and signal content, is automatically recorded in the system database. This logging function not only provides users with a basis for audit trails but also provides a valuable data foundation for subsequent analysis of early warning accuracy, optimization of thresholds, and model development. The entire implementation embodies a closed-loop automated process from data to decision-making to action, aiming to provide users with timely, accurate, and actionable status information, enabling them to proactively respond to potential outdoor power supply failures.

[0043] Example 4: This example details a method for analyzing long-term operating time-series data of outdoor power supplies to identify stable and degrading periods, and further integrating multi-dimensional features to generate a comprehensive quality degradation index. The implementation of this method relies on the statistical characteristic analysis of time-series data and a feature weighting fusion strategy based on domain knowledge.

[0044] The implementation process begins with trend analysis of the collected power supply usage time series data. This data typically includes the power supply's cumulative operating hours, charge / discharge cycles, and the percentage of operating time under different loads. The primary task of the analysis is to identify the stable and fluctuating periods in the data. To achieve this, the system employs a moving window statistical method. A fixed-length window (e.g., 30 days) slides along the time axis, calculating statistics such as the moving average and moving standard deviation for each window. The moving average reflects the average level within that time period, while the moving standard deviation quantifies the degree of data fluctuation. By observing the trend of the moving standard deviation, different operating stages of the power supply can be effectively distinguished. When the moving standard deviation remains below a threshold set based on historical data or expert experience, it indicates that the power supply's operating parameters change only slightly and its performance is stable; therefore, this period is classified as a stable period. Conversely, when the moving standard deviation shows a continuous upward trend and eventually exceeds the threshold, it indicates that the power supply's operating status begins to fluctuate abnormally, and its performance may be entering a degradation phase; therefore, this period is marked as a degradation period. This classification method, based on the inherent volatility of the data, can objectively reveal the turning points in the health status of the power supply.

[0045] After successfully segmenting the time periods, key features need to be extracted from voltage, current, and temperature data within both stable and degradation periods. For voltage data, the standard deviation within an analysis sub-window (e.g., 24 hours) is calculated to measure the stability of the voltage output; simultaneously, the long-term drift slope of the voltage value is obtained through linear fitting to reflect its slow changing trend. For current data, the intensity of fluctuations is monitored, and the difference between peak and valley currents (peak-to-peak value) is calculated; furthermore, spectral analysis is performed to identify changes in harmonic components, as an increase in the amplitude of a specific harmonic may indicate aging of internal components. For temperature data, abnormal events are monitored, the cumulative duration of temperature values ​​exceeding the preset safe operating threshold is recorded, and the frequency of such abnormal events is calculated. These features characterize the operating status of the power supply from different perspectives. See Table 1 for examples of feature values ​​extracted from different time periods in a hypothetical analysis case.

[0046] Table 1: Feature extraction results for different time periods

[0047] After obtaining the above features, the next step is to fuse these heterogeneous features into a single comprehensive quality degradation feature index. Since each feature has different physical meanings and dimensions, they first need to be normalized to unify their numerical range. Normalization is based on the maximum and minimum values ​​of each feature in historical data, transforming each feature value to the range of zero to one.

[0048] The fusion process employs a weighted average method, assigning a weight coefficient to each normalized feature. This coefficient reflects the relative importance of the feature in the overall power supply degradation assessment. The weight allocation is not arbitrary but based on multiple considerations: First, historical fault data is analyzed to statistically determine the correlation strength between each feature and the eventual fault occurrence; features with higher correlation are assigned greater weight. Second, expert knowledge in power supply design and manufacturing is considered; for example, temperature is generally considered critical to battery life, thus features related to temperature exceedances may receive higher weights. Third, the characteristics of actual application scenarios are taken into account. Assuming that, based on the above principles, the following weights are assigned: voltage standard deviation 0.2, voltage drift slope 0.3, current peak-to-peak value 0.2, cumulative temperature exceedance time 0.15, and the number of temperature exceedance events 0.15. The weighted sum of all features is the initial fusion value.

[0049] To obtain the final quality degradation characteristic index, this initial fused value is standardized. Standardization linearly maps it to a more intuitive numerical range, such as zero to one hundred. Zero represents the best health state, and one hundred represents a severely degraded state. The comprehensive index generated in this process effectively aggregates multi-source information, providing a clear and quantitative input signal for subsequent machine learning models to assess the health status of the power supply and predict its remaining lifespan. The entire implementation emphasizes data-driven objective analysis and the reasonable fusion based on domain knowledge, aiming to achieve an accurate characterization of the power supply's degradation state.

[0050] Example 5: See Figure 5 This paper details the specific implementation method for calculating the similarity of quality degradation between different time points. This process is the foundation for constructing a power degradation state graph. Its core lies in quantifying the correlation strength between the power health states corresponding to any two time points in historical monitoring data, thereby providing accurate edge weight information for subsequent graph model analysis.

[0051] The implementation process begins with processing the generated time series of quality degradation characteristic indicators. This series consists of indicator values ​​arranged chronologically, each corresponding to a specific data collection time point, such as a comprehensive indicator calculated at midnight each day. These time points form the basic unit for subsequent similarity calculations. The system first extracts each time point t from the series. i and its corresponding quality degradation characteristic index value Q i Here, the subscript i represents the index number of the time point, and Q... i It is a comprehensive value that reflects the health status of the power supply at that point in time. The higher the value, the more severe the degradation usually is.

[0052] After obtaining the indicator values ​​at all time points, the system needs to calculate the values ​​at any two different time points t. i and tj The difference in state between them. This difference is obtained by calculating the absolute difference between their corresponding index values. Specifically, the calculation is as follows: This absolute difference Δ ij It is a non-negative real number that intuitively represents the absolute deviation of the power supply's health state at two points in time. The larger the difference, the greater the difference in state between the two points in time; the smaller the difference, the closer the states are.

[0053] Absolute differences are not directly suitable as edge weights in graph models because graph models typically require weights that represent the strength of the association between nodes (i.e., higher similarity corresponds to higher weights). Therefore, a transformation function is needed to convert the absolute difference Δ... ij Mapped to a similarity score S ij Above, and requires S ij The value falls within the range of 0 to 1. An efficient and computationally concise conversion method uses the following mathematical expression: ; In this formula, S ij Representing time point t i With time point t j The similarity score of quality degradation between them. Δ ij This is the absolute difference between the quality degradation characteristic index values ​​calculated in the previous step for the two time points. λ is a positive proportionality coefficient used to adjust the sensitivity of similarity to the difference. e is the base of the natural logarithm (approximately 2.71828).

[0054] This exponential function implements a non-linear mapping from difference to similarity. When the states at two time points are exactly the same, i.e., Δ... ij When = 0, the similarity score S ij =1 indicates perfect similarity. As the absolute difference Δ... ij The increase of S ij The value of λ decreases monotonically towards 0. The value of the scaling factor λ is crucial; it needs to be determined based on the actual range and historical fluctuations of the quality degradation characteristic index Q. A larger λ value causes similarity to decay more rapidly with increasing differences, meaning the system is more sensitive to state changes; while a smaller λ value allows for higher similarity scores even with greater state differences, making the model more inclusive. Optimization of λ is typically achieved by analyzing typical ranges of state changes in historical data.

[0055] After performing the above calculations on all the time point pairs that need to be analyzed, the system obtains a complete similarity matrix. This matrix is ​​a symmetric matrix because S ij =S jiFurthermore, the elements on the diagonal (i.e., when i=j) have a value of 1. This matrix precisely describes the pairwise state similarity relationships between all monitoring time points.

[0056] The similarity scores S obtained from these calculations ij The weights are used to construct the edges connecting vertices (time points) in the degenerate state graph. Larger weights indicate stronger edge connections, more similar states at two time points, and potentially closer proximity on the degenerate path. This process provides quantifiable and meaningful connections for transforming time-series data into a graph structure, and is a crucial prerequisite for implementing the entire graph theory analysis method.

[0057] It should be noted that, in this document, relational terms such as "first" and "second" are used only to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such process, method, article, or apparatus.

[0058] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their equivalents.

Claims

1. A machine learning-based outdoor power supply long-term use quality degradation early warning method, characterized in that, Includes the following steps: Acquire long-term usage monitoring data of outdoor power supplies, including voltage change data, current fluctuation data, temperature record data, and usage time series data; Based on the long-term use monitoring data, quality degradation characteristic indicators were determined; A power degradation assessment model is constructed based on the aforementioned quality degradation characteristic indicators; The warning triggering conditions for outdoor power supplies are determined based on the power degradation assessment model. When the aforementioned warning triggering conditions are met, a quality degradation warning message is output.

2. The machine learning-based early warning method for long-term quality degradation of outdoor power supplies according to claim 1, characterized in that, The acquisition of long-term usage monitoring data for outdoor power supplies specifically includes: The raw operating parameters of the outdoor power supply are collected in real time by sensor devices. These raw operating parameters include voltage, current, temperature and timestamp data. The original usage parameters are cleaned and normalized to obtain preprocessed long-term usage monitoring data.

3. The machine learning-based early warning method for long-term quality degradation of outdoor power supplies according to claim 1, characterized in that, The determination of quality degradation characteristic indicators based on the long-term use monitoring data specifically includes: Based on the usage time series data in the long-term usage monitoring data, stable time periods and degradation time periods are divided; During the stable period and the degradation period, voltage change characteristics, current fluctuation characteristics and temperature anomaly characteristics are extracted respectively. By integrating the voltage change characteristics, current fluctuation characteristics, and temperature anomaly characteristics, a quality degradation characteristic index is generated.

4. The machine learning-based early warning method for long-term quality degradation of outdoor power supplies according to claim 3, characterized in that, The construction of the power degradation assessment model based on the aforementioned quality degradation characteristic indicators is specifically as follows: Calculate the quality degradation similarity between different time points, wherein the quality degradation similarity is based on the differences in the quality degradation feature indicators; A degradation state graph is constructed using time points as vertices and quality degradation similarity as edge weights. Determine the minimum spanning tree of the degradation state graph to represent the degradation evolution path.

5. The machine learning-based early warning method for long-term quality degradation of outdoor power supplies according to claim 4, characterized in that, The determination of the early warning triggering conditions for outdoor power supplies based on the power degradation assessment model specifically includes: The degradation trend is analyzed based on the minimum spanning tree, and the degradation risk score at each time point is calculated. A warning threshold is set based on the aforementioned degradation risk score; The aforementioned warning threshold is used as the warning trigger condition.

6. The machine learning-based early warning method for long-term quality degradation of outdoor power supplies according to claim 5, characterized in that, When the warning triggering condition is met, the quality degradation warning information is output, specifically as follows: Monitor real-time quality degradation characteristic indicators and compare them with the warning threshold; If the real-time quality degradation characteristic index exceeds the warning threshold, a quality degradation warning signal is generated; The quality degradation warning signal is output to the user terminal.

7. The machine learning-based early warning method for long-term quality degradation of outdoor power supplies according to claim 3, characterized in that, The division of the stable period and the degradation period is as follows: Analyze the changing trends of time series data to identify periods of stability and fluctuation. Stable time periods are defined based on data stability periods, and deterioration time periods are defined based on data fluctuation periods.

8. The machine learning-based early warning method for long-term quality degradation of outdoor power supplies according to claim 7, characterized in that, The process of integrating the voltage change characteristics, current fluctuation characteristics, and temperature anomaly characteristics to generate quality degradation characteristic indicators is as follows: Calculate the weighted average of voltage variation characteristics, current fluctuation characteristics, and temperature anomaly characteristics; The weighted average value is standardized into a quality degradation characteristic index.

9. The machine learning-based early warning method for long-term quality degradation of outdoor power supplies according to claim 4, characterized in that, The calculation of the quality degradation similarity between different time points is specifically as follows: Extract the quality degradation characteristic index values ​​at each time point; Calculate the absolute difference between the quality degradation characteristic index values ​​at any two time points; The absolute difference is converted into a similarity score.

10. The machine learning-based early warning method for long-term quality degradation of outdoor power supplies according to claim 6, characterized in that, The generation of the quality degradation early warning signal is specifically as follows: The warning level is determined based on the degree of deviation between the real-time quality degradation characteristic indicators and the warning threshold; Based on the warning level, generate corresponding quality degradation warning signal content.

Images