Hydrogenation equipment component life monitoring and early warning sliding window statistics early warning method and system

Abstract

The application relates to the field of hydrogen energy infrastructure intelligent operation and maintenance, and discloses a hydrogenation equipment component life monitoring and early warning sliding window statistical early warning method and system. The method sequentially executes five core steps of multi-source data collection and environment adaptive preprocessing, three-level sliding window statistical feature extraction, multi-parameter coupling degradation index calculation, adaptive baseline dynamic calibration and residual life prediction, and multiple confirmation grading early warning and feedback regulation, and a complete cooperative architecture is formed through the closed-loop feedback of the early warning driven data collection frequency. The system solves the problems of early degradation trend of hydrogen refueling equipment being difficult to identify, high false alarm rate of single parameter monitoring and residual service life being unable to be quantitatively predicted.

Description

Technical Field

[0001] This invention belongs to the field of intelligent operation and maintenance and predictive maintenance technology of hydrogen energy equipment, specifically involving a method for monitoring and early warning of the life of hydrogen refueling equipment components based on multi-scale sliding window statistical analysis and multi-parameter coupled degradation modeling, and its corresponding early warning system. Background Technology

[0002] With the rapid development of the hydrogen energy industry, hydrogen refueling stations, as a key infrastructure for the promotion and application of hydrogen fuel cell vehicles, are receiving increasing attention for the operational reliability and safety of their core refueling equipment. Operating under high-pressure hydrogen conditions for extended periods, critical components such as compressors, valves, seals, and cooling systems inevitably undergo performance degradation. Failure to detect degradation trends and implement timely maintenance measures can lead to serious safety accidents such as hydrogen leaks and overpressure runaway. Hydrogen possesses unique properties, including flammability, explosiveness, low molecular weight leading to leakage, and hydrogen embrittlement of metallic materials. These characteristics make the safety monitoring requirements for hydrogen refueling equipment far higher than those for ordinary industrial equipment. Therefore, establishing an effective equipment component lifespan monitoring and early warning mechanism is crucial for ensuring the safe operation of hydrogen refueling stations.

[0003] Currently, the maintenance system for hydrogen refueling equipment mainly adopts either a periodic inspection model or a post-failure repair model. The periodic inspection model involves checking and maintaining the equipment at fixed intervals, regardless of its actual health condition. This approach often leads to a contradiction between over-maintenance and under-maintenance. Over-maintenance results in frequent equipment downtime, increasing operation and maintenance costs and reducing the service capacity of hydrogen refueling stations; under-maintenance may allow components showing signs of degradation to continue operating until they fail, causing unplanned downtime and safety risks. The post-failure repair model only addresses issues after equipment failure occurs, lacking any preventative capability. For refueling equipment involving high-pressure hydrogen, this passive response approach poses serious safety hazards.

[0004] At the monitoring technology level, existing hydrogen refueling equipment monitoring systems mostly adopt a single-parameter fixed threshold alarm method. This involves setting upper and lower thresholds for individual operating parameters such as hydrogen temperature and refueling pressure, triggering an alarm when real-time data exceeds these thresholds. However, this method has several significant shortcomings. First, single-parameter threshold alarms ignore the inherent physical coupling relationships between various operating parameters in the hydrogen refueling equipment. For example, there is an inherent correlation between hydrogen temperature and refueling pressure, and between cooling efficiency and cylinder temperature; monitoring only a single parameter cannot capture this coupled degradation pattern. Second, fixed thresholds cannot adapt to the normal fluctuation range of parameters under different operating conditions and environmental conditions. This leads to normal operating condition fluctuations such as seasonal temperature changes and switching between different refueling modes being misjudged as abnormal, resulting in a high false alarm rate. Third, fixed thresholds can only detect obvious anomalies that have exceeded safe limits, failing to identify slow degradation trends in equipment performance, thus missing the optimal opportunity for preventative maintenance before a failure occurs.

[0005] Furthermore, while fault prediction technologies based on machine learning methods such as deep learning have made some progress in recent years, these methods typically require large amounts of high-quality labeled training data, have poor model interpretability, and demand high computational resources, making them unsuitable for direct deployment at industrial hydrogen refueling stations. Meanwhile, while existing industrial equipment health assessment systems can integrate data from multiple sensors to score health, their scoring models are mostly based on simple linear weighting, failing to fully consider the dynamic evolution characteristics of time series and the physical coupling effects between parameters. This results in assessment results that are difficult to accurately reflect the gradual process of equipment degradation. In the actual operation of hydrogen refueling equipment, component degradation often manifests as a complex pattern of simultaneous small changes in multiple parameters. For example, a decrease in compressor efficiency may simultaneously cause increased fluctuations in exhaust pressure and a slow rise in exhaust temperature, while the individual changes in each parameter do not exceed their respective threshold ranges. This complex degradation pattern cannot be effectively captured by single-parameter monitoring methods alone. Moreover, existing methods generally lack adaptive feedback mechanisms; monitoring parameters and warning thresholds remain fixed after system deployment, unable to be dynamically adjusted according to the actual operating conditions and degradation stages of the equipment. This is particularly insufficient in application scenarios where the equipment lifecycle spans several years. The process of hydrogen refueling equipment going from commissioning to significant performance degradation can last from months to years. The operating characteristics and parameter sensitivities of the equipment differ at different degradation stages, making it difficult for fixed monitoring strategies to maintain optimal monitoring results throughout the entire lifecycle. In summary, current technologies lack a predictive maintenance method that can effectively identify degradation trends in hydrogen refueling equipment, while also possessing high interpretability, low implementation complexity, and adaptive capabilities. The industry urgently needs a degradation trend early warning technology solution tailored to the specific needs of hydrogen refueling equipment and capable of stable operation under industrial field conditions. Summary of the Invention

[0006] To address the aforementioned technical problems, this invention provides a sliding window statistical early warning method for monitoring and warning the lifespan of hydrogen refueling equipment components. This method constructs a sliding window statistical analysis framework with multiple time scales, combines multi-parameter coupled degradation modeling and adaptive feedback optimization mechanisms, and achieves early identification, quantitative assessment, and graded early warning of the performance degradation trend of hydrogen refueling equipment.

[0007] This invention discloses a sliding window statistical early warning method for monitoring and warning the lifespan of hydrogen refueling equipment components. The method includes the following steps: First, a data acquisition and preprocessing step is performed. Operating parameter time-series data are synchronously acquired from multiple sensors of the hydrogen refueling equipment at a preset acquisition frequency. The acquired raw data undergoes outlier removal, missing value imputation, and timestamp alignment to obtain preprocessed multi-parameter synchronous time-series data. Then, a multi-scale sliding window statistical analysis step is performed. Short-term, medium-term, and long-term sliding windows are applied to the preprocessed multi-parameter synchronous time-series data for statistical feature extraction. The short-term window uses hourly granularity to capture sudden anomalies and rapid changes; the medium-term window uses daily granularity to identify periodic fluctuations and medium-term degradation; and the long-term window uses monthly granularity to characterize the overall degradation trend of the equipment. The statistical feature results of each window are fused to form a multi-scale statistical feature vector sequence. Next, a multi-parameter coupled degradation analysis step is performed. A multi-parameter coupled correlation model is established based on a multi-scale statistical feature vector sequence. Principal component analysis is used for dimensionality reduction and feature reconstruction error calculation to achieve multi-parameter collaborative anomaly detection. Simultaneously, degradation trends are identified for each parameter, and single-parameter degradation indicators and multi-parameter coupled anomaly indicators are combined to form a comprehensive degradation feature vector. Then, a health measurement assessment and graded early warning step is performed. Based on the comprehensive degradation feature vector, a weighted scoring model is used to calculate the equipment health score, and a corresponding level of early warning response is triggered based on the health score and a preset grading threshold. Finally, an adaptive feedback optimization step is performed. Based on the historical trend of the health score and early warning trigger records, the sliding window parameters and weight coefficients are dynamically adjusted to form an adaptive monitoring feedback closed loop.

[0008] The present invention also provides a sliding window statistical early warning system for monitoring and warning the lifespan of hydrogen refueling equipment components, including a data acquisition and preprocessing module, a multi-scale sliding window statistical analysis module, a multi-parameter coupled degradation analysis module, a health quantification assessment and graded early warning module, and an adaptive feedback optimization module. Each module corresponds one-to-one with each step of the method described above.

[0009] This invention offers the following advantages: Multi-scale sliding window statistical analysis enables full-time-domain monitoring, covering everything from sudden anomalies to long-term degradation trends. The short-term window captures transient anomalies at an hourly granularity, the medium-term window tracks cyclical fluctuations at a daily granularity, and the long-term window assesses overall degradation trends at a monthly granularity. This three-level window collaboration significantly improves the sensitivity and timeliness of degradation trend identification. Multi-parameter coupled degradation modeling fully utilizes the inherent physical correlations between various operating parameters of the hydrogenation equipment. Principal component analysis dimensionality reduction and feature reconstruction error calculation achieve accurate capture of multi-parameter collaborative degradation, effectively compensating for the blind spots of single-parameter monitoring methods in recognizing complex degradation patterns. The false alarm rate is reduced by 40% to 60% compared to single-parameter threshold monitoring. An adaptive feedback optimization mechanism enables the system to dynamically adjust monitoring strategies and scoring weights based on the actual health status of the equipment. When the equipment is in good health, the calculation frequency is automatically reduced to save 50% of computing resources; when signs of degradation appear, the monitoring accuracy is automatically increased to ensure timely warnings. Methods based on classical statistical analysis principles offer high interpretability and low implementation complexity. All early warning results can be traced back to specific parameter change trends and statistical indicators, making it easier for maintenance personnel to understand the causes of anomalies and develop targeted maintenance measures. At the same time, they do not require a large amount of labeled data or a professional algorithm team, making them suitable for direct deployment and application in industrial sites of hydrogen refueling stations of various sizes. Attached Figure Description

[0010] Figure 1 This is a flowchart of the sliding window statistical early warning method for monitoring and warning the lifespan of hydrogen refueling equipment components provided in this embodiment of the invention.

[0011] Figure 2 This is an architecture diagram of the sliding window statistical early warning system for monitoring and warning the lifespan of hydrogen refueling equipment components provided in this embodiment of the invention. Detailed Implementation

[0012] The technical solution of the present invention will now be described in detail with reference to the accompanying drawings and specific embodiments. It should be noted that the following embodiments are for illustrative purposes only and are not intended to limit the scope of the invention.

[0013] See Figure 1 The hydrogen refueling equipment component life monitoring and early warning sliding window statistical early warning method provided in this embodiment of the invention is applied to the monitoring scenario of hydrogen refueling station refueling equipment equipped with multiple types of sensors. The method includes five core steps from step S1 to step S5, and each step forms a deeply coupled data-driven closed loop.

[0014] Step S1: Data Acquisition and Preprocessing. In one embodiment of the present invention, the data acquisition and preprocessing step is the basic link of the entire early warning method. Its main task is to obtain high-quality time series data of operating parameters from various sensors of the hydrogen refueling equipment, and to eliminate noise interference and inconsistencies in the original data through systematic preprocessing operations, so as to provide a reliable data foundation for subsequent statistical analysis and degradation modeling.

[0015] Specifically, in this embodiment, data acquisition synchronously obtains time-series data of operating parameters from multiple sensors deployed on the hydrogen refueling equipment at a preset acquisition frequency. Preferably, the acquisition frequency is set to 1Hz, meaning that the readings from each sensor are acquired once per second. This frequency ensures data timeliness while avoiding the storage and computational burden caused by excessively high sampling rates. In some alternative embodiments, the acquisition frequency can be flexibly adjusted within the range of 0.5Hz to 2Hz according to actual monitoring needs. The acquired operating parameters include at least the following five key indicators: hydrogen temperature... This characterizes the real-time temperature of hydrogen in the refueling pipeline, with a typical measurement range of -40℃ to 85℃; cylinder temperature. This characterizes the temperature change of the outer wall of the on-board hydrogen storage tank, with a typical measurement range of -20℃ to 70℃; filling pressure. This characterizes the hydrogen pressure in the refueling line, with a typical measurement range of 0 to 87.5 MPa; real-time flow rate. The instantaneous flow rate of hydrogen during the refueling process is characterized, with a typical measurement range of 0 to 5 kg / min; equipment vibration signal. Accelerometers, installed on the compressor and valve assemblies, are used to characterize the mechanical vibration characteristics of the equipment, with a typical measurement range of 0 to 50 mm / s. In actual deployments, auxiliary parameters such as cumulative flow, ambient temperature, and ambient humidity can also be collected, depending on the specific equipment configuration of the hydrogen refueling station.

[0016] After completing the raw data acquisition, this step performs three preprocessing operations on the time-series data of each channel. The first operation is outlier removal, which is implemented using a statistical filtering method based on standard deviation. Specifically, a length of [value missing] is maintained for each parameter channel. The moving average buffer, preferably The data points are set to 600, corresponding to a 10-minute data window. The moving average is calculated for the data within the buffer. and sliding standard deviation The deviation from the moving average will exceed Data points are marked as outliers. The marked outlier data point is replaced by the linear interpolation result of its two preceding and following valid data points, thus maintaining data continuity while eliminating sudden interference. In one embodiment of the invention, the standard deviation multiple threshold is set to 3. This value is based on the statistical characteristics of the normal distribution, ensuring that the probability of normal data being misjudged as outliers is controlled below 0.27%, effectively balancing the sensitivity and specificity of outlier removal.

[0017] The second preprocessing operation is missing value imputation. In actual operation, factors such as sensor failure and communication interruption may cause data loss at certain times. This step uses linear interpolation to fill in short-term data loss with a missing proportion of no more than 5%, and marks the corresponding interval as invalid data segment for long-term data loss with a missing proportion of more than 5% and excludes it in subsequent analysis. The third preprocessing operation is timestamp alignment. Since there may be slight deviations in the sampling clocks of different sensors, this step uses a unified system clock as a reference to resample and synchronize the data of each channel to ensure that the data of all parameters are strictly aligned at the same time. Preferably, the resampling adopts the zero-order hold method, and takes the effective sampled value closest to the target time as the alignment result. After the above three preprocessing operations, the preprocessed multi-parameter synchronized time series data is obtained, denoted as a matrix. ,in This is a discrete-time index. It's worth noting that the quality of data preprocessing directly affects the accuracy of subsequent statistical analysis and degradation modeling. In a preferred embodiment of the invention, the preprocessed data undergoes data smoothing before being written to the subsequent processing buffer, using a window length of... The moving average method is used to smooth the data of each channel, preferably... The data is represented by 10 data points, corresponding to a 10-second data window. Moving average smoothing effectively suppresses high-frequency noise interference while preserving data trend characteristics, making subsequent statistical feature calculations more robust. Furthermore, the preprocessing module maintains data quality monitoring indicators, providing real-time statistics on the data missing rate and outlier percentage for each channel. When the data quality indicator for a particular channel continuously exceeds a preset alarm threshold, a sensor fault report is automatically generated to notify maintenance personnel to perform sensor maintenance or replacement.

[0018] Step S2: Multi-scale sliding window statistical analysis. In one embodiment of the present invention, the multi-scale sliding window statistical analysis step receives the multi-parameter synchronous time-series data output from step S1. As input, a three-tiered sliding window is used to extract hierarchical statistical features from the data, aiming to capture the changing characteristics of equipment operating status at different time granularities. The collaborative working mechanism of the three-tiered window is one of the core innovations of this invention. Each window has a clear division of labor and complements the others, forming a complete time-domain monitoring system covering sudden anomalies to long-term degradation.

[0019] The first level is the short-term window, and its window size is... The preferred setting is 24 hours, corresponding to 86,400 data points at a sampling frequency of 1 Hz, with a sliding step size of... The preferred setting is 1 hour, or 3600 data points. The main function of the short-term window is to monitor sudden anomalies and rapid changes in equipment, suitable for scenarios requiring rapid response, such as compressor start-up anomalies and instantaneous valve failures. Within each short-term window, five statistical characteristics are calculated for each operating parameter. Window mean. Reflects parameters within the window period Average operating level: ,in, For parameters At the start of the window After that, the The value of each sampling point, The total number of data points included in the short-term window. For parameter index, These correspond to five parameter channels: hydrogen temperature, cylinder temperature, filling pressure, real-time flow rate, and equipment vibration signal. Window standard deviation. Reflects the fluctuation range of the parameter within the window period:

[0020] ,in, For the degrees of freedom correction term, an unbiased estimate is used to improve the accuracy of standard deviation calculation with finite samples. Window range Reflects the extreme fluctuation range of parameters within the window. Rate of change slope. The parameters are obtained by linearly fitting the data points within the window using the least squares method, which characterizes the direction and rate of change of the parameters within the window period.

[0021] ,in, This is the index of the data points within the window. The numerator and denominator are the cross product and sum of squares terms used in the slope calculation of least squares linear regression, respectively. Distribution skewness Used to characterize the asymmetry of data distribution within a window, positive skewness indicates a tendency for the parameter to shift towards higher values, while negative skewness indicates a shift towards lower values.

[0022] Among them, the denominator The standard deviation normalization factor makes the skewness a dimensionless index, which facilitates cross-sectional comparisons between different parameters.

[0023] The second level is the intermediate window, and its window size is... The preferred setting is 7 days, or 604,800 data points, with a sliding step size. The preferred setting is 6 hours, or 21,600 data points. The intermediate-term window is primarily used to identify periodic fluctuations and medium-term degradation trends in equipment performance. It is suitable for degradation patterns that require several days of observation to become apparent, such as a slow decline in refueling efficiency and gradual degradation of sealing performance. The statistical features extracted within the intermediate-term window are the same as those in the short-term window, but due to its longer time span, it can smooth out the impact of intraday fluctuations and more clearly reflect the performance change trend over a cross-day scale. Preferably, the intermediate-term window also additionally calculates a periodic intensity index. The data is obtained by performing autocorrelation analysis on the data within the window, and is used to determine whether the parameters exhibit regular fluctuations with a daily cycle.

[0024] The third level is the long-term window, and its window size is... The preferred setting is 30 days, or 2,592,000 data points, with a sliding step size. The preferred setting is 1 day, or 86,400 data points. The long-term window serves as a comprehensive assessment of the equipment degradation trend, with a time span sufficient to cover the gradual transition of the hydrogenation equipment from normal operation to degradation. Within the long-term window, in addition to calculating the five statistical characteristics mentioned above, the goodness of fit of the long-term trend line is also calculated. and health decay rate The health decay rate is defined as a normalized index of the rate of parameter degradation within a long-term window.

[0025] ,

[0026] in, The slope of the rate of change over the long-term window. This is the long-term window mean. This represents the number of data points within a long-term window. This metric amplifies the slope relative to the mean by multiplying it by the window length, making the degree of degradation comparable for parameters with different dimensions. When Exceeding the preset degradation threshold When, preferably A value of 0.05 indicates that the parameter The cumulative change over the long-term window has exceeded 5% of the mean, indicating a significant degradation trend.

[0027] After extracting statistical features from each level of the window, the statistical results from the three levels of windows are aligned by time and then concatenated into vectors to form a multi-scale statistical feature vector. It should be noted that because the sliding step size of the three-level windows is different, the time interval between the statistical results generated by each level of window is also different. Therefore, time alignment processing is required during feature fusion. Specifically, a sliding step size of 6 hours for the intermediate window is used as the unified feature update cycle. For the short-term window, the latest value among the multiple statistical results generated within each 6-hour interval is taken as the representative short-term feature at that moment. Statistical results that remain unchanged between two update cycles for the long-term window are directly reused. This time alignment strategy ensures that the three-level features are synchronously available while preserving the original temporal resolution characteristics of each level of window. For each parameter... Its multi-scale feature vectors are A total of 11 features are generated. The eigenvectors of the five parameter channels are concatenated to form a 55-dimensional multi-scale statistical feature vector. This constitutes the input data for subsequent degradation analysis.

[0028] Step S3: Multi-parameter coupled degradation analysis. In one embodiment of the present invention, the multi-parameter coupled degradation analysis step receives the multi-scale statistical feature vector sequence output from step S2. By establishing a multi-parameter coupled correlation model, the degradation status of the equipment is comprehensively evaluated from two dimensions: single-parameter degradation and multi-parameter synergistic anomalies. This step is the key difference between this method and the traditional single-parameter threshold monitoring method. It makes full use of the physical coupling relationship between various operating parameters of the hydrogenation equipment and can identify complex degradation modes that cannot be detected by single-parameter monitoring.

[0029] First, perform single-parameter degradation trend identification. For each operating parameter... Extract the slope of the rate of change from its long-term window statistical features. and health decay rate This serves as a criterion for degradation assessment. Simultaneously, the trend fit is calculated. This indicator reflects how well the long-term trend line fits the actual data. The parameter is determined when the following conditions are met. A degradation trend exists: the absolute value of the slope of the rate of change exceeds the preset slope threshold. Furthermore, the trend fit exceeds the preset fit threshold. Preferably, for the injection pressure parameter, Set to 0.02 MPa / day. Set to 0.6; for hydrogen temperature parameters, Set to 0.1℃ / day. Set to 0.6. The differences in threshold values ​​for different parameters reflect the varying degradation characteristics and dimensions of each parameter. Single-parameter degradation index. Defined as a normalized measure of degradation intensity: ,in, For parameters The corresponding slope threshold, For trend fit, The function truncates the upper limit of the degradation index to 100 to prevent extreme degradation values ​​from having an excessive impact on subsequent health calculations. This formula combines slope strength with fit reliability, producing higher degradation index values ​​only when the degradation trend is both significant and stable, thus effectively filtering out occasional slope fluctuations.

[0030] Secondly, multi-parameter coupling anomaly detection is performed. This invention employs a multi-parameter coupling correlation model based on principal component analysis. Its core idea is to establish a low-dimensional representation of the coupling relationships between parameters using normal operation data. When equipment degradation occurs, the coupling relationships between parameters deviate from the normal pattern, manifesting as an increase in reconstruction error in the low-dimensional space. The specific implementation process is as follows: First, a training dataset is selected from the multi-scale statistical feature vectors collected during the normal operation phase of the equipment. ,in For the number of training samples, preferably The dataset should contain at least 720 datasets, corresponding to at least 30 days of normal operation data. Principal component analysis should be performed on the training dataset after standardization, retaining the top performers with a cumulative variance contribution rate of 95%. Construct the projection matrix from the principal components. For the newly acquired feature vector First, project the model to the principal component space and then back project it back to the original space. Calculate the feature reconstruction error. ,in, This is the transpose of the projection matrix. The square of the L2 norm of the vector. The reconstructed vector is the feature vector projected onto the principal component space. When the coupling relationship between parameters is normal, the feature vector can be well represented by the principal component space, and the reconstruction error is small. However, when multiple parameters simultaneously undergo slight degradation, causing a shift in the coupling relationship, the reconstruction error increases significantly. The statistical distribution of the reconstruction error is calculated based on the training dataset, and an anomaly threshold is set. Add three times the standard deviation to the mean of the reconstruction error of the training samples. When calculated in real time... Exceed When this occurs, a multi-parameter collaborative anomaly is determined to have occurred.

[0031] Preferably, this step also establishes two sets of physical correlation analysis sub-models to enhance the interpretability of coupling degradation. The first set is a pressure-flow correlation analysis, which is performed under normal conditions when pressure is added. With real-time traffic There is an approximately linear relationship between them. ,in The correlation coefficient, The intercept is a constant. The correlation coefficient is obtained through linear regression within a long window. The temporal changes, when A significant shift indicates a decrease in compressor efficiency or an increase in piping resistance. The second group is a temperature-efficiency correlation analysis, defining cooling efficiency. for:

[0032] ,in, The temperature of hydrogen gas. The temperature of the bottle. The ambient temperature. When When the long-term window mean shows a continuous downward trend, it indicates that the cooling system performance is degrading.

[0033] Finally, the single-parameter degradation index and multi-parameter coupling anomaly indicators After normalization, the combined features form a comprehensive degenerate feature vector. ,in The normalized value for the reconstruction error ranges from 0 to 100. The normalization method uses the following formula: ,in, The anomaly threshold is determined based on the training data. This represents the ratio of the current reconstruction error to the anomaly threshold. Multiplying this by 50 results in a normalized value of 50 when the reconstruction error is exactly equal to the anomaly threshold, and a normalized value of 100 when the error exceeds twice the threshold. This normalization method ensures that the coupled anomaly index and the single-parameter degradation index are fused and calculated on the same numerical scale, guaranteeing the rationality of the health score. The comprehensive degradation feature vector carries both independent degradation information for each parameter and multi-parameter coupled anomaly information, providing a comprehensive description of the degradation status for subsequent health assessments.

[0034] Step S4: Health Measurement Assessment and Graded Early Warning. In one embodiment of the present invention, the health measurement assessment and graded early warning step receives the comprehensive degradation feature vector output from step S3. By using a weighted scoring model, multidimensional degradation information is integrated into a single health score, and differentiated early warning responses are triggered based on a tiered threshold system. This step transforms equipment status from a qualitative description to a quantitative assessment, providing maintenance personnel with an intuitive and actionable basis for decision-making.

[0035] Device health score The calculation uses a weighted subtraction model, and its calculation formula is as follows:

[0036] ,in, for The device health score at any given time, with an initial maximum score of 100, indicates that the device is in optimal operating condition; For running parameters The weighting coefficient reflects the importance of this parameter to the overall health of the equipment; For parameters The single-parameter degradation index value ranges from 0 to 100; The weighting coefficients for multi-parameter coupled anomaly indicators; This represents the normalized value of the multi-parameter coupling anomaly index, ranging from 0 to 100. All weight coefficients satisfy the normalization constraint. .

[0037] Preferably, the initial values ​​of each weighting coefficient are allocated based on the safety importance and degradation sensitivity of each component of the hydrogenation equipment. The weight corresponding to the stability of the refueling pressure is... The value is set to 0.25, reflecting the primary impact of pressure parameters on refueling safety; the weight corresponding to the accuracy of hydrogen temperature control. Set to 0.20 to reflect the important role of temperature control in refueling quality; the weight corresponding to flow consistency. The value is set to 0.15 to reflect the impact of flow rate parameters on refueling efficiency; the weight corresponding to the equipment vibration level. The value is set to 0.12 to reflect the impact of mechanical vibration on component lifespan; the weight corresponding to bottle temperature is... The value is set to 0.10 to reflect the impact of hydrogen storage tank temperature on the safety margin; the weight of multi-parameter coupling anomalies. The value is set to 0.18 to reflect the significant role of coupling anomalies in the comprehensive evaluation. This weighting allocation assigns higher scoring weights to safety-critical parameters, embodying the safety-first design philosophy.

[0038] To obtain a health score Subsequently, a corresponding level of early warning response is triggered based on a preset hierarchical threshold system. This invention sets four status levels: three early warning levels plus one normal level. The first level is an emergency warning, triggered when the health score... The first level is triggered when the device's health has severely deteriorated and poses a security risk. The system immediately sends multiple notifications via SMS, phone, and platform pop-ups, requiring maintenance personnel to immediately shut down the device for inspection and activate the emergency plan. The second level is a severe warning, triggered when... The first level is triggered when the device shows a clear trend of degradation but has not yet reached a dangerous level. The system notifies maintenance personnel via platform pop-ups and email pushes to schedule planned maintenance within 24 hours. The third level is a monitoring alert. When triggered, it indicates that the device shows slight signs of degradation and requires attention. The system will notify maintenance personnel via platform message to arrange an inspection within the week. If the equipment is in normal operating condition, routine monitoring is sufficient.

[0039] Preferably, this step also includes a false alarm suppression mechanism to reduce the system's false alarm rate. The false alarm suppression mechanism includes three layers of protection. The first layer is environmental factor compensation. By collecting ambient temperature and humidity data, an environmental compensation model is established to correct the monitoring data of hydrogen temperature and cylinder temperature, eliminating the interference of seasonal temperature changes and diurnal temperature differences on degradation judgment. Specifically, the temperature degradation index after environmental compensation... for: ,in, This is the ambient temperature compensation coefficient. This represents the deviation of the current ambient temperature from the standard reference temperature of 25°C. The value ranges from 0.05 to 0.15, and the specific value is determined through regression analysis of historical normal operation data. The second layer is operating mode recognition. The system automatically identifies the current operating mode based on the characteristics of changes in refueling flow rate and pressure, including fast refueling mode, slow refueling mode, and standby mode. Differentiated normal fluctuation ranges of parameters are set for different operating conditions to avoid parameter abrupt changes during operating mode switching being misjudged as degradation. Specifically, the characteristic of fast refueling mode is real-time flow rate... kg / min and injection pressure The temperature shows a continuous upward trend, and under this mode, the normal fluctuation range of hydrogen temperature expands to... ℃; the characteristics of the slow filling mode are: kg / min, normal temperature fluctuation range set to ℃; the characteristics of standby mode are kg / min, at this point the parameter should remain basically stable, and the normal fluctuation range should be narrowed to ℃. Automatic identification of operating conditions enables the system to adaptively adjust anomaly judgment criteria under different operating scenarios, fundamentally reducing false alarms caused by changes in operating conditions. The third layer is multi-window cross-confirmation, requiring at least two windows at different time scales to indicate anomalies before triggering an alert. That is, after an anomaly is detected in the short-term window, it needs to be corroborated by statistical indicators in the medium-term or long-term window, thereby filtering out false alarms from single windows caused by occasional disturbances.

[0040] Step S5: Adaptive Feedback Optimization. In one embodiment of the present invention, the adaptive feedback optimization step dynamically adjusts the window parameters of each sliding window in step S2 and the weight coefficients of each operating parameter in step S4 based on the historical trend of the health score output in step S4 and the early warning trigger records, thereby forming a closed-loop adaptive monitoring mechanism. This step enables the entire early warning system to automatically optimize the monitoring strategy according to the actual health status of the equipment, reducing the computational load to save resources when the equipment is in good health, and improving monitoring accuracy to ensure timely early warning when the equipment shows signs of degradation.

[0041] The dynamic adjustment strategy for window parameters is as follows: The system maintains a window with a length of... A health score history buffer, preferably The value is 168, corresponding to a health record updated hourly for 7 consecutive days. Adjustment decisions are made based on the health score sequence in the historical buffer. When all health scores in the buffer are greater than or equal to 90, indicating the device is continuously operating normally, the sliding step size of the short-term window is adjusted. The sliding step size of the intermediate window, scaling up from an initial value of 1h to 2h. The time frame was increased from 6 hours to 12 hours, reducing the statistical calculation frequency by 50% and significantly reducing computational resource consumption. When the health score enters the warning range... At that time, the sliding step size of the short-term window will be adjusted. The timeframe is restored to 1 hour, and the weighting coefficients of degradation parameters for the medium-term and long-term windows are increased to strengthen the monitoring of trend-based degradation. When the health score enters the severe or emergency warning range... At the same time, the sliding step size of all windows is reduced to the minimum allowable value. Preferably, the sliding step size of the short-term window is reduced to 0.5h, the sliding step size of the medium-term window is reduced to 4h, and the sliding step size of the long-term window is reduced to 0.5 days, so as to achieve the highest accuracy of all-round monitoring.

[0042] The dynamic adjustment strategy for weighting coefficients is as follows: The system records the contribution value of each parameter's degradation index at each warning trigger, and determines the dominant degradation parameter causing the warning through statistical analysis. For past... For parameters contributing more than 30% to the initial warning, their weight coefficients should be appropriately increased, preferably by 10% to 20% of the current weight. Conversely, for parameters that have not degraded over a long period, their weight coefficients should be appropriately reduced to maintain the weight normalization constraint. The upper and lower limits for weight adjustment are set to 0.5 to 2.0 times the initial weight to prevent extreme adjustments from causing an imbalance in the assessment.

[0043] Preferably, adaptive feedback optimization also includes a slow evolution mechanism for the window size. Once a long-term degradation trend of a parameter is confirmed, the corresponding medium-term window size can be shortened from the initial 7 days to 5 days to improve the sensitivity to changes in the parameter's degradation rate. Conversely, for parameters that have not shown a degradation trend for a long time, the long-term window size can be extended from 30 days to 45 days to obtain trend information over a longer time span and improve the robustness of trend judgment. The adjustment of the window size is limited to no more than 50% of the initial value to ensure that the time scale differences between different window levels are not disrupted.

[0044] Preferably, step S5 also introduces an early warning effect evaluation mechanism as a quality assurance for feedback optimization. The system periodically statistically analyzes the actual response results of maintenance personnel after an early warning is triggered, marking early warnings confirmed as valid as "true positives" and early warnings confirmed as false alarms as "false positives." When the false positive rate exceeds a preset quality control threshold (preferably 25%) within a continuous statistical period, the system automatically tightens the conditions for multi-window cross-confirmation, increasing the requirement from at least two windows indicating anomalies to at least three windows indicating anomalies, in order to reduce the false alarm rate. Conversely, when the true positive rate remains at a high level (preferably exceeding 85%), the cross-confirmation conditions can be appropriately relaxed to improve the sensitivity of degradation detection. Through this dynamic threshold adjustment based on actual maintenance feedback, the system continuously optimizes the balance between early warning accuracy and degradation identification sensitivity during long-term operation, achieving continuous improvement in early warning performance.

[0045] Through the coordinated work of steps S1 to S5, this invention constructs a complete processing link from data acquisition to degradation analysis and then to early warning decision-making. The feedback adjustment mechanism in step S5 transmits the evaluation results of step S4 back to steps S2 and S4 themselves, forming a deeply coupled closed loop between data acquisition, statistical analysis, degradation modeling, early warning assessment and parameter optimization. This enables the system to continuously self-optimize to adapt to the changing monitoring needs of the equipment at different stages of its life cycle.

[0046] See Figure 2 The present invention also provides a sliding window statistical early warning system for monitoring and warning the lifespan of hydrogen refueling equipment components. This system corresponds one-to-one with each step of the above method embodiment, including the coordinated operation of five functional modules, realizing end-to-end automated processing from data acquisition to degradation early warning.

[0047] The data acquisition and preprocessing module corresponds to step S1 in the method embodiment. This module is configured to establish data communication connections with various sensors deployed on the hydrogen refueling equipment, and synchronously acquire time-series data of operating parameters such as hydrogen temperature, cylinder temperature, filling pressure, real-time flow rate, and equipment vibration signals according to a preset acquisition frequency. In one embodiment of the present invention, the data acquisition and preprocessing module includes two sub-units: a sensor interface unit and a data preprocessing engine. The sensor interface unit interacts with various sensors through industrial fieldbus or wireless communication protocols, supporting mainstream industrial communication protocols such as Modbus RTU, Modbus TCP, and OPC UA, ensuring compatibility with sensors of different brands and models. The data preprocessing engine receives the raw data stream forwarded by the sensor interface unit, and sequentially performs outlier removal based on standard deviation, missing value filling by linear interpolation, and synchronous processing by unified time base resampling, outputting the preprocessed multi-parameter synchronous time-series data to subsequent modules. Preferably, the data preprocessing engine adopts a streaming processing architecture, where data enters the preprocessing pipeline immediately upon acquisition, with a processing delay of no more than 100ms, meeting the requirements of near real-time monitoring.

[0048] The multi-scale sliding window statistical analysis module corresponds to step S2 in the method embodiment. This module is configured to apply three levels of sliding windows—short-term, medium-term, and long-term—to the received multi-parameter synchronous time-series data for statistical feature extraction. Internally, this module maintains three independent sliding window data buffers, corresponding to a 24-hour short-term window, a 7-day medium-term window, and a 30-day long-term window, respectively. Each buffer uses a circular buffer data structure, automatically overwriting the oldest data upon arrival of new data, achieving efficient window sliding without frequent memory allocation and release operations. The calculation of statistical features for each level of window is performed according to the formulas described in the method embodiment, including indicators such as window mean, window standard deviation, window range, rate of change slope, and distribution skewness. After calculation, the statistical results of the three levels of windows are merged to form a 55-dimensional multi-scale statistical feature vector sequence, which is then passed to the multi-parameter coupled degradation analysis module.

[0049] The multi-parameter coupled degradation analysis module corresponds to step S3 in the method embodiment. This module is configured to perform two core analysis tasks based on the received multi-scale statistical feature vector sequence: single-parameter degradation trend identification and multi-parameter coupled anomaly detection. In one embodiment of the present invention, this module includes a single-parameter degradation evaluation subunit and a multi-parameter coupled analysis subunit. The single-parameter degradation evaluation subunit performs threshold comparison judgment on the slope of the long-term window change rate and the trend fit of each parameter, and outputs the degradation index value of each parameter. The multi-parameter coupled analysis subunit has a built-in principal component analysis model trained with normal operating data, performs projection and reconstruction error calculation on the real-time feature vector, and determines whether multi-parameter coordinated anomalies have occurred. Preferably, this module also integrates two physical correlation analysis sub-models: pressure-flow correlation analysis and temperature-efficiency correlation analysis, to enhance the interpretability of degradation diagnosis. The output results of the two subunits are combined to form a comprehensive degradation feature vector.

[0050] The health measurement assessment and graded early warning module corresponds to step S4 in the method embodiment. This module is configured to calculate the equipment health score using a weighted scoring model based on the comprehensive degradation feature vector, and trigger an early warning response of the corresponding level based on the graded threshold. This module includes a health calculation subunit, a false alarm suppression subunit, and an early warning triggering subunit. The health calculation subunit updates the health score in real time according to the weighted deduction model formula in the method embodiment. The false alarm suppression subunit performs triple false alarm suppression processing: environmental factor compensation, operating condition pattern recognition, and multi-window cross-confirmation. The early warning triggering subunit compares the health score after false alarm suppression with preset three-level thresholds for emergency early warning, severe early warning, and attention early warning, triggers an early warning response of the corresponding level, and delivers the early warning information to maintenance personnel through SMS, email, and platform message push interfaces.

[0051] The adaptive feedback optimization module corresponds to step S5 in the method embodiment. This module is configured to dynamically adjust the sliding step size and window size of each level of the sliding window in the multi-scale sliding window statistical analysis module, as well as the weight coefficients of each operating parameter in the health quantification assessment and hierarchical early warning module, based on the historical trend of the health score and the early warning trigger records. This module maintains a health history buffer and an early warning trigger log database. By analyzing the temporal change pattern of the health score and the degradation contribution ratio of each parameter, it generates window parameter adjustment instructions and weight update instructions according to the dynamic adjustment strategy described in the method embodiment, and sends them back to the multi-scale sliding window statistical analysis module and the health quantification assessment and hierarchical early warning module, respectively, forming a complete adaptive feedback closed loop. Preferably, the adaptive feedback optimization module also provides a visualization function for parameter adjustment logs and optimization effect reports, facilitating the review and supervision of the system's adaptive behavior by operation and maintenance personnel.

[0052] The data flow between the above modules is as follows: data acquisition and preprocessing module → multi-scale sliding window statistical analysis module → multi-parameter coupled degradation analysis module → health measurement assessment and hierarchical early warning module, forming a forward data processing link; the adaptive feedback optimization module receives the output of the health measurement assessment and hierarchical early warning module and sends adjustment instructions back to the multi-scale sliding window statistical analysis module and the health measurement assessment and hierarchical early warning module, forming a reverse feedback adjustment link. The forward and reverse links together form a deeply coupled closed-loop system architecture, enabling the entire early warning system to have the ability to continuously self-optimize.

[0053] In one embodiment of the present invention, the aforementioned functional modules are deployed on industrial-grade edge computing devices at the hydrogen refueling station site. The edge computing devices are connected to the electrical control system of the hydrogen refueling equipment via Ethernet or RS-485 serial bus, and connected to the upper-level operation and maintenance management platform via 4G / 5G wireless network or fiber optic broadband. This edge-cloud collaborative deployment architecture enables data acquisition and real-time early warning processing to be completed locally, ensuring response timeliness and data security. Historical health data and early warning records are synchronously uploaded to the cloud-based operation and maintenance management platform, supporting remote monitoring, historical analysis, and multi-site comparative management. Preferably, the system also provides a visual display interface, displaying the operating status of each parameter in real time in the form of a dashboard, sliding window statistical curves, health score trend charts, and early warning history records. Operation and maintenance personnel can view the equipment health status at any time via PC and mobile terminals.

[0054] Furthermore, the system boasts excellent scalability. When new equipment or sensors are added to a hydrogen refueling station, only the corresponding sensor configuration information needs to be added to the data acquisition and preprocessing module, and the training data of the principal component analysis model needs to be updated in the multi-parameter coupled degradation analysis module; no major modifications to the system architecture are required. The system also supports expanding the monitoring scope from refueling machines to other key equipment within the hydrogen refueling station, such as hydrogen storage tank groups, compressor units, and cooling systems. By configuring differentiated window parameters, weighting coefficients, and early warning thresholds for different equipment types, a comprehensive equipment health management platform covering the entire hydrogen refueling station can be built.

[0055] To verify the effectiveness of the proposed hydrogen refueling equipment component life monitoring and early warning method, a six-month field deployment test was conducted at a typical 70MPa hydrogen refueling station. The test environment included two membrane compressors, four sets of hydrogen storage cylinder bundles, two refueling machines, and a supporting cooling system. A total of 32 sensors were used, covering the five types of operating parameters mentioned above. The total amount of test data exceeded 500 million sampling records.

[0056] Regarding the ability to identify degradation trends, the method of this invention successfully identified three equipment degradation events during the test: The first case was the degradation of the exhaust valve seal of compressor No. 1, manifested as a slope of the long-term window change rate of the filling pressure reaching -0.035 MPa / day with a trend fit of 0.78. The system triggered a warning 18 days before the failure occurred, and the maintenance personnel replaced the exhaust valve seal ring within the planned maintenance window based on the warning prompt, avoiding an unplanned downtime event that could have caused a filling interruption; The second case was the decrease in the efficiency of the cooling system of filling machine No. 2, manifested as a decrease in cooling efficiency. Within 30 days, the efficiency dropped from 0.82 to 0.71. The multi-parameter coupling analysis module simultaneously detected a correlation shift in the statistical characteristics of hydrogen temperature and cylinder temperature. Within 12 days of the warning being triggered, the system arranged for maintenance, including coolant replacement and heat exchanger cleaning, restoring the cooling efficiency to above 0.80. The third case involved a multi-parameter coupling anomaly and feature reconstruction error. The fact that the degradation rate exceeded 1.5 times the abnormal threshold for five consecutive days, while each individual parameter degradation index did not exceed its respective threshold, demonstrates the unique advantage of the multi-parameter coupled detection method in detecting complex degradation patterns. Investigation revealed that the pressure and flow parameters were synchronously abnormal due to a deviation in the opening of the pressure reducing valve of hydrogen storage cylinder bundle No. 1. All three degradation events were successfully identified and handled before functional failures occurred, verifying the effectiveness of the method of this invention in early identification of degradation trends.

[0057] Regarding the false alarm rate, the method of this invention generated 8 warnings during the 6-month testing period. Of these, 6 were valid warnings caused by actual degradation or abnormal events, and 2 were false alarms caused by external factors (sudden changes in ambient temperature due to extreme weather), resulting in an effective warning rate of 75%. In contrast, a traditional single-parameter fixed threshold monitoring system operating during the same period generated 47 alarms, of which only 9 were related to actual equipment anomalies, resulting in an effective warning rate of only 19.1%. The method of this invention reduced the number of false alarms by approximately 83% compared to the traditional method, and improved the effective warning rate by approximately 294%.

[0058] Regarding the accuracy of health assessment, the equipment health score calculated by this invention was compared with the results of manual assessments of equipment status by professional maintenance engineers. Using a monthly statistical period, the consistency rate between the health score and the manual assessment level reached 87.5%. In actual cases of equipment degradation, the average lead time for the decline in health score was 14 days, meaning the system could detect signs of degradation approximately two weeks earlier than manual inspections.

[0059] Regarding computational resource consumption, the method of this invention, running on an industrial-grade edge computing device equipped with an Intel Core i5-12400 processor and 16GB of memory, took approximately 120ms for a single complete statistical calculation, with a peak memory usage of approximately 580MB. Thanks to the adaptive feedback optimization mechanism, the average computational load during normal device operation was reduced by approximately 48% compared to the highest precision mode, validating the design goal of the adaptive strategy to effectively conserve computational resources while ensuring monitoring effectiveness.

[0060] In summary, the actual deployment test results show that the multi-scale sliding window statistical early warning method proposed in this invention has significant advantages in early identification of degradation trends, false alarm rate control, accuracy of health assessment, and computational efficiency, and can meet the practical application needs of predictive maintenance of hydrogen refueling station equipment.

[0061] In terms of economic benefits, the three preventative maintenance procedures scheduled during the testing period, based on early degradation warnings, saved approximately 120 hours of unplanned downtime compared to post-failure repair. Based on the average daily refueling volume of 50 units at the hydrogen refueling station and the revenue per refueling service, this is estimated to have prevented direct economic losses of approximately 150,000 yuan. Furthermore, because preventative maintenance allows for repairs or component replacements before equipment is severely damaged, maintenance costs are reduced by approximately 30% to 50% compared to post-failure overhauls. The introduction of an adaptive feedback optimization mechanism automatically reduces the system's computing frequency during normal equipment operation, lowering the average annual power consumption of the edge computing device by approximately 35%, further reducing system operating costs. The above test data fully validates the technical feasibility and economic value of the method presented in practical hydrogen refueling station applications, providing a practical and efficient comprehensive technical solution for predictive maintenance of hydrogen energy infrastructure.

[0062] The embodiments of the present invention are not limited to the specific embodiments described above. Those skilled in the art can make various equivalent changes or substitutions based on the technical solutions of the present invention, and all such changes or substitutions should be included within the protection scope of the present invention.

Claims

1. A sliding window statistical early warning method for monitoring and warning the lifespan of hydrogen refueling equipment components, characterized in that, Includes the following steps: Step S1, Data Acquisition and Preprocessing: Simultaneously acquire time-series data of operating parameters from multiple sensors of the hydrogen refueling equipment at a preset acquisition frequency. The time-series data of operating parameters includes at least hydrogen temperature, cylinder temperature, filling pressure, real-time flow rate and equipment vibration signal. The time series data of the operating parameters are processed by outlier removal, missing value imputation and timestamp alignment to obtain preprocessed multi-parameter synchronous time series data; Step S2, Multi-scale sliding window statistical analysis: Apply three levels of sliding windows—short-term window, medium-term window, and long-term window—to the multi-parameter synchronous time-series data to extract statistical features. The short-term window is used to capture sudden anomaly features, the medium-term window is used to identify periodic fluctuation features, and the long-term window is used to characterize degradation trend features. The statistical feature results of each window are fused to form a multi-scale statistical feature vector sequence; Step S3, Multi-parameter Coupling Degradation Analysis: Based on the multi-scale statistical feature vector sequence, a multi-parameter coupling correlation model is established. By performing correlation quantification analysis on the physical coupling relationship between various operating parameters, single-parameter degradation trends and multi-parameter collaborative anomalies are identified. Single-parameter degradation indicators and multi-parameter coupling anomaly indicators are combined to form a comprehensive degradation feature vector. Step S4, Health Measurement Assessment and Graded Early Warning: Based on the comprehensive degradation feature vector, calculate the equipment health score using a weighted scoring model, and trigger the corresponding level of early warning response based on the health score and the preset graded threshold. Step S5, Adaptive Feedback Optimization: Based on the historical trend of the health score and the early warning trigger records, dynamically adjust the window parameters of each sliding window in step S2 and the weight coefficients of each running parameter in step S4 to form an adaptive monitoring feedback closed loop.

2. The method according to claim 1, characterized in that, In step S1, the preset acquisition frequency is 0.5Hz to 2Hz; the outlier removal adopts a statistical filtering method based on standard deviation, marking data points that deviate from the moving mean by more than a preset standard deviation multiple as outliers and replacing them with linear interpolation of adjacent data points; the timestamp alignment process uses a unified time reference to resample and synchronize multi-source sensor data.

3. The method according to claim 1, characterized in that, In step S2, the short-term window has a window size of 12 to 48 hours and a sliding step size of 0.5 to 2 hours; the medium-term window has a window size of 5 to 14 days and a sliding step size of 4 to 12 hours; and the long-term window has a window size of 20 to 60 days and a sliding step size of 0.5 to 2 days.

4. The method according to claim 1, characterized in that, In step S4, the grading thresholds include: triggering an emergency warning when the health score is lower than a first threshold, triggering a severe warning when the health score is between the first threshold and a second threshold, and triggering a concern warning when the health score is between the second threshold and a third threshold; wherein the first threshold ranges from 55 to 65 points, the second threshold ranges from 75 to 85 points, and the third threshold ranges from 88 to 92 points.

5. The method according to claim 1, characterized in that, In step S2, the statistical features extracted from the data within each sliding window include: window mean, window standard deviation, window range, rate of change slope, and distribution skewness; wherein, the rate of change slope is obtained by linearly fitting the data points within the window using the least squares method, and the distribution skewness is used to characterize the degree of asymmetry in the distribution of data within the window.

6. The method according to claim 1, characterized in that, In step S3, the multi-parameter coupling correlation model includes: performing principal component analysis to reduce the dimensionality of the multi-scale statistical feature vector sequence, calculating the feature reconstruction error at each time step in the reduced low-dimensional space as a multi-parameter coupling anomaly index; and determining that a multi-parameter collaborative anomaly has occurred when the feature reconstruction error exceeds the anomaly threshold determined based on historical data statistics.

7. The method according to claim 1, characterized in that, In step S3, the identification of the single-parameter degradation trend includes: calculating the trend slope value and trend fit degree of the long-term window statistical characteristics of each operating parameter; when the absolute value of the trend slope value exceeds the preset slope threshold and the trend fit degree exceeds the preset fit degree threshold, it is determined that the operating parameter has a degradation trend.

8. The method according to claim 1, characterized in that, Step S4 also includes false alarm suppression processing, which includes: environmental factor compensation, which compensates and corrects the monitoring data according to the ambient temperature and humidity; operating condition mode identification, which distinguishes the normal parameter fluctuation range under different refueling conditions; and multi-window cross-confirmation, which requires that at least two windows of different time scales indicate abnormalities before triggering an early warning.

9. The method according to claim 1, characterized in that, In step S5, the dynamic adjustment includes: when the health score is within the normal range for a consecutive preset number of times, increasing the sliding step size of the short-term window to reduce computational resource consumption; when the health score enters the attention warning range, decreasing the sliding step size of the short-term window and increasing the weight coefficient of the running parameter corresponding to the medium-term window; when the health score enters the severe warning or emergency warning range, simultaneously decreasing the sliding step size of each level of window to the minimum value.

10. A sliding window statistical early warning system for monitoring and warning the lifespan of hydrogenation equipment components, used to implement the method described in any one of claims 1-9, characterized in that, include: The data acquisition and preprocessing module is configured to synchronously acquire time-series data of operating parameters from multiple sensors of the hydrogenation equipment at a preset acquisition frequency, and perform outlier removal, missing value interpolation and timestamp alignment on the time-series data of operating parameters to obtain preprocessed multi-parameter synchronous time-series data. The multi-scale sliding window statistical analysis module is configured to apply three levels of sliding windows—short-term window, medium-term window, and long-term window—to the multi-parameter synchronous time-series data to extract statistical features and fuse them to form a multi-scale statistical feature vector sequence. The multi-parameter coupled degradation analysis module is configured to establish a multi-parameter coupled correlation model based on the multi-scale statistical feature vector sequence, identify single-parameter degradation trends and multi-parameter collaborative anomalies, and form a comprehensive degradation feature vector; The health measurement assessment and graded early warning module is configured to calculate the device health score based on the comprehensive degradation feature vector and trigger an early warning response of the corresponding level based on the graded threshold. The adaptive feedback optimization module is configured to dynamically adjust the window parameters of each level of sliding window in the multi-scale sliding window statistical analysis module and the weight coefficients of each operating parameter in the health quantification assessment and graded early warning module based on the historical trend of the health score and the early warning trigger records.

Images