A method and system for detecting faults in a probe device

Abstract

The present application relates to the technical field of detection device fault detection, in particular to a detection device fault detection method and system, which comprises calculating the mean value and standard deviation of the information entropy value of the healthy detection device subset, and then calculating the standard score corresponding to the information entropy value of each detection device in the current sliding window based on the mean value and standard deviation; taking the absolute value of the standard score corresponding to each detection device to obtain the abnormality coefficient of each detection device, and implementing graded early warning on each detection device according to the abnormality coefficient. The present application takes the dynamically screened healthy detection device group as the calculation reference, significantly improves the diagnostic robustness in multiple fault concurrent scenarios, prevents mutual interference between faulty detection devices, avoids missed or mistaken judgments, and ensures the continuous monitoring capability of the system under the condition of partial component failure.

Description

Technical Field

[0001] This invention relates to the field of fault detection technology for detection devices, specifically to a fault detection method and system for a composite fire detection device in an energy storage power station based on information entropy. In particular, it relates to early, online, unsupervised diagnosis and warning of faults in the detection device itself. Background Technology

[0002] Electrochemical energy storage power stations are critical infrastructure of energy systems, and their fire safety is of paramount importance. Composite fire detectors (typically integrating multiple sensing units such as smoke, temperature, and CO) are widely deployed in energy storage compartments to detect potential fire hazards at an early stage. Here, composite fire detectors are collectively referred to as detection devices (hereinafter referred to as detection devices). However, the detection devices themselves may also malfunction due to the harsh environment inside the energy storage compartment, such as high temperature, high humidity, and chemical corrosion, as well as aging and component failure caused by long-term operation. These malfunctions may manifest as reading drift, decreased sensitivity, intermittent abnormalities, or even complete freezing at a fixed value.

[0003] To address the aforementioned issues, existing technology CN121026185A proposes a dynamic detection and fault-tolerance method for navigation anomaly data. This method acquires real-time time-series observation data from the navigation system using multiple sensors, then calculates the information entropy perturbation rate of each sensor's data stream within a sliding window. By comparing the deviation of the current entropy value with historical baseline entropy values, it determines abnormal trends such as data mutations, drift, or oscillations. The core of this method lies in introducing anomaly type reasoning, repair result evaluation, and dynamic feedback mechanisms, achieving adaptive optimization of data anomaly repair and improving repair accuracy and long-term system stability.

[0004] In existing technologies, Z-scores (standard scores) are also a commonly used method for unsupervised / semi-supervised anomaly detection in the field of fault diagnosis. The core principle is to standardize the features of a single sample using the statistical mean of the features under normal operating conditions as a benchmark and the statistical standard deviation as a unified unit of measurement, eliminating differences in the dimensions and magnitudes of different features and achieving unified quantitative comparison of data. Under normal conditions, the absolute value of the sample Z-score will fluctuate slightly around 0; when the object experiences performance degradation or failure, the features deviate from the normal statistical benchmark, and the corresponding absolute value of the Z-score will increase significantly. By judging whether this value exceeds a preset threshold, it is possible to identify whether the sample is abnormal and achieve preliminary fault diagnosis.

[0005] However, existing technologies still have the following limitations:

[0006] Data-driven diagnostic methods require a large amount of known normal and fault labeled data for model training, which is difficult to obtain in actual engineering, especially for diverse fault modes, where the model's generalization ability faces challenges.

[0007] Methods based on statistical comparisons within a population (such as calculating Z-scores) assume that most detection devices are functioning correctly. However, when multiple detection devices within the energy storage compartment fail successively or simultaneously, the abnormal data from these failed devices can "contaminate" the statistical benchmarks used for comparison (such as the overall mean and standard deviation), leading to distorted Z-score calculations. This can result in misjudgments of normal detection devices or underjudgments of failed detection devices, thus reducing the robustness of the system.

[0008] The aforementioned issues may lead to missed alarms (the faulty detection device fails to alarm for a real fire) or false alarms (normal detection devices falsely alarm due to environmental interference) in fire early warning systems, seriously threatening the safe operation of energy storage power stations and increasing operation and maintenance costs. Therefore, there is a need for an online, adaptive detection device self-diagnostic technology that can maintain high reliability even under multiple fault conditions, providing a more reliable guarantee for the safe operation of energy storage power stations. Summary of the Invention

[0009] This summary section is provided to briefly introduce the concepts, which will be described in detail in the detailed description section below. This summary section is not intended to identify key or essential features of the claimed technical solution, nor is it intended to limit the scope of the claimed technical solution.

[0010] In a first aspect, the present invention provides a fault detection method for a detection device, which calculates the anomaly coefficient and provides graded early warning based on the current sliding window, including:

[0011] Acquire multi-dimensional time-series monitoring data from each detection device in the health detection device subset;

[0012] Calculate the information entropy value of each detection device in the health detection device subset within the current sliding window;

[0013] The mean and standard deviation are calculated based on the information entropy values ​​of a subset of health detection devices. Then, based on the mean and standard deviation, the standard scores corresponding to the information entropy values ​​of all detection devices within the current sliding window are calculated.

[0014] The absolute value of the standard score corresponding to each detection device is taken to obtain the anomaly coefficient of each detection device, and the detection device is given a graded early warning based on the anomaly coefficient.

[0015] Furthermore, the step of generating the subset of health detection devices includes:

[0016] Before processing the current sliding window or after processing each sliding window, check the historical anomaly coefficients of each detection device;

[0017] If the abnormality coefficient of any detection device reaches the alarm threshold within a consecutive preset number of sliding windows, then the detection device will be moved from the healthy list to the fault isolation list.

[0018] The health detection devices constitute a subset of the detection devices listed in the health list;

[0019] In the initial state, all detection devices are included in the health list.

[0020] In this technical solution, when calculating the anomaly coefficient, only the information entropy value of the selected health detection devices is used as the benchmark to solve for the mean and standard deviation. Because detection devices that have been judged as abnormal and moved to the fault isolation list are excluded, the anomaly entropy value of the fault detection devices is prevented from skewing the statistical benchmark. This accurately solves the problem of diagnostic distortion in existing standard scores under group fault scenarios, directly improving the accuracy of the anomaly coefficient calculation and providing a reliable quantitative basis for subsequent fault determination.

[0021] Specifically, the calculation of multi-level entropy values ​​is not isolated and parallel, but rather presents a progressive and mutually supportive logical relationship. The information entropy values ​​include single-sensor information entropy, conditional information entropy, information gain, joint information entropy, and the sliding information entropy change rate. Single-sensor information entropy serves as the basic level, first accurately quantifying the inherent uncertainty of the data monitored by a single detection device to construct the basic features for anomaly judgment. Based on this, the joint information entropy is derived layer by layer through the single-sensor information entropy of each pair of detection devices, capturing the data correlation features between multiple detection devices. Then, the conditional information entropy and information gain are further derived from the joint information entropy and single-sensor information entropy to quantify the efficiency and dependence of information transmission between detection devices. Finally, based on the entropy values ​​of each level of the continuous sliding window, the sliding information entropy change rate is calculated to capture the temporal mutation features of the entropy values.

[0022] This layered calculation of high-quality entropy features, through dynamic filtering based on recent continuous sliding window historical anomaly coefficients, moves abnormal detection devices to a fault isolation list, retaining only healthy detection devices in the healthy list. This ensures that the single-sensor information entropy of all devices within the healthy detection device subset represents the true uncertainty quantification value under normal operating conditions, eliminating entropy distortion caused by abnormal data from faulty devices. Simultaneously, the joint information entropy of devices within the healthy subset reflects the inherent and stable data correlation between healthy detection devices, preventing abnormal entropy values ​​from faulty devices from disrupting the normal correlation characteristics between multiple devices and causing deviations in the joint information entropy calculation. This makes the entropy calculations at the two fundamental levels—single-sensor information entropy and joint information entropy—more accurate, effectively eliminating abnormal interference factors. These two levels of entropy values ​​are the core basis for subsequent calculations of the mean and standard deviation. The mean value calculated based on the high-quality base entropy values ​​of all health devices can accurately represent the benchmark level of entropy values ​​under normal working conditions within the current sliding window, rather than a false benchmark skewed by faulty devices; the calculated standard deviation can accurately reflect the normal fluctuation range of the entropy values ​​of health detection devices, reflecting the inherent differences of health devices, rather than the fluctuation distortion caused by abnormal data.

[0023] Because the mean and standard deviation are highly representative and reliable, they can not only accurately calculate the standard scores of devices in the health list, but also be effectively used to calculate the standard scores of detection devices in the fault isolation list.

[0024] Secondly, this invention provides a fault detection system for a detection device, which calculates the anomaly coefficient and provides graded early warning based on the current sliding window, including:

[0025] The data acquisition module is configured to acquire multi-dimensional time-series monitoring data of each detection device in the health detection device subset in real time;

[0026] The entropy calculation module is connected to the data acquisition module and is configured to calculate the information entropy value of each detection device in the health detection device subset within the current sliding window.

[0027] The anomaly detection module is connected to the entropy calculation module. It is configured to calculate the mean and standard deviation based on the information entropy values ​​of a subset of health detection devices. Then, based on the mean and standard deviation, it calculates the standard score corresponding to the information entropy value of each detection device in the current sliding window. The absolute value of the standard score corresponding to each detection device is taken to obtain the anomaly coefficient of each detection device.

[0028] The fault warning module is connected to the anomaly detection module and is configured to perform graded warnings based on the anomaly coefficient.

[0029] Compared with the prior art, the beneficial effects of the present invention are as follows:

[0030] 1. This invention detects the performance degradation trend of the detection device before it completely fails, transforming passive response into proactive maintenance.

[0031] 2. By comparing standard scores within a group, this invention can effectively offset the combined effects of environmental factors (such as diurnal temperature variation and seasonal changes) on all detection devices, focusing on individual anomalies.

[0032] 3. This invention uses a dynamically selected group of healthy detection devices as the calculation benchmark, which significantly improves the diagnostic robustness in scenarios with multiple concurrent faults, prevents mutual interference between fault detection devices, avoids missed or false judgments, and ensures the system's continuous monitoring capability in the event of partial component failure. Attached Figure Description

[0033] Figure 1 This is a flowchart of the fault detection method of the detection device of the present invention;

[0034] Figure 2 This is a schematic diagram of the fault detection system of the detection device of the present invention.

[0035] Definition of labels in the diagram:

[0036] 100. Fault detection system; 101. Data acquisition module; 102. Entropy calculation module; 103. Anomaly detection module; 104. Fault early warning module. Detailed Implementation

[0037] Embodiments of the invention will be described more fully below with reference to the accompanying drawings. The invention may be embodied in many different forms and should not be construed as limited to the embodiments set forth herein. Rather, these embodiments are provided so that this disclosure will be thorough and complete, and will fully convey the scope of the invention to those skilled in the art.

[0038] The present invention discloses a method and system for detecting faults in a detection device. Existing composite fire detection devices (such as smoke, heat and gas composite detection devices) lack effective online self-diagnosis and early performance degradation warning mechanisms, and cannot identify sensitivity decline, drift or partial functional abnormalities before the device completely fails.

[0039] To address the aforementioned issues, the technical solution of this invention does not require prior fault data. By adaptively selecting a reference benchmark, it achieves unsupervised, early, and accurate diagnosis and warning of the working status of the composite fire detection device itself, effectively preventing missed or false alarms caused by detection device failure, and improving the overall fire safety reliability of the energy storage power station.

[0040] First embodiment, Figure 1 The flowchart of the fault detection method for the detection device is shown. Figure 1In this method, the anomaly coefficient calculation and graded early warning of the detection device are realized based on the current sliding window, specifically including:

[0041] S1. Data acquisition and preprocessing: Specifically, this involves real-time acquisition of multi-dimensional time-series monitoring data from multiple composite fire detection devices within the energy storage power station, such as smoke concentration, temperature, and CO concentration, and preprocessing the data, including cleaning and filling, to form a multi-dimensional feature data stream.

[0042] S2. Sliding window division, wherein the preset sliding window length can be adjusted according to the data characteristics of the detection device and the requirements of early warning timeliness, preferably ranging from 10 seconds to 60 seconds, and the sliding step size is preferably from 1 second to 5 seconds. For example, the length of the sliding window is set to 60 seconds (60 frames), and the sliding step size is 1 second (1 frame).

[0043] S3. Information entropy calculation: Specifically, this involves calculating the information entropy value of each detection device within the current sliding window. More specifically, it utilizes the sliding window to divide the multi-dimensional feature data stream, calculating the information entropy value of the multi-dimensional feature data stream of each detection device within each sliding window. The information entropy value includes single-sensor information entropy, conditional information entropy, information gain, joint information entropy, and the sliding information entropy change rate. Specifically:

[0044] In single-sensor information entropy, "single sensor" refers to a single detection device (hereinafter, "sensor" refers to the detection device). This entropy describes the data sequence of detection device X within a sliding window T. First, the data is discretized into Given a set of discretized intervals, calculate the probability distribution for each interval. ,in:

[0045] ;

[0046] Then, the single-sensor information entropy value of the detection device X within the sliding window T is calculated:

[0047] ;

[0048] In the formula, denoted as the single-sensor information entropy value of the detection device X within the sliding window T; n represents the total number of multi-dimensional time-series monitoring data of the detection device X within the sliding window T. Let X be the probability that the multidimensional time-series monitoring data of the detection device X falls into the i-th discretized interval; Let be the number of data points in the multidimensional time-series monitoring data of the detection device X that fall into the i-th discretized interval.

[0049] Joint information entropy is a bivariate extension of single-sensor information entropy. Its core function is to quantify the overall uncertainty of the joint distribution of data from two detection devices X and Y within the same sliding window T. It reflects the comprehensive dispersion of the data from both devices and is an important indicator for analyzing the correlation of data from multiple detection devices and assisting in collaborative fault diagnosis. Specifically:

[0050] ;

[0051] In the formula, Let be the joint information entropy value of detection devices X and Y within the same sliding window T, where the data sequence of detection device X within the sliding window T is denoted as . The data sequence of the detection device Y within the sliding window T z represents the total number of intervals obtained by discretizing the multidimensional time-series monitoring data of the detection device Y. Let be the i-th data interval (i=1,2,…,m) after discretization of the detection device X. Let Y be the j-th data interval (j=1,2,…,z) after discretization of the detection device Y. Let be the two-dimensional joint probability, that is, the joint probability that the data from detection device X falls into the i-th interval and the data from detection device Y falls into the j-th interval, where:

[0052] ;

[0053] In the formula, Let X be the number of data points where X falls in the i-th interval and Y falls in the j-th interval. This represents the total number of valid combined data points for X and Y within window T.

[0054] Conditional information entropy and information gain, where conditional information entropy and information gain are derived quantitative indicators based on single-sensor information entropy and joint information entropy, are primarily used to analyze the dependency and information transmission degree of data from two detection devices within the same sliding window, and are key indicators for judging the correlation of data from multiple detection devices. Specifically:

[0055] ;

[0056] ;

[0057] In the formula, Within the sliding window T, the conditional information entropy value of X with Y as a condition represents the degree of uncertainty of the monitoring data of X under the premise that the monitoring data information of the detection device Y is known. It reflects the remaining information complexity of X after the information of Y is known. The smaller the value, the stronger the interpretability of the information of Y to the information of X, and the higher the correlation between the two data. The joint information entropy value of detection devices X and Y within the sliding window T is used to quantify the overall uncertainty of the joint distribution of their data. To quantify the uncertainty of Y's own data by measuring the single-sensor information entropy value of the detection device Y within the sliding window T; Within the sliding window T, the information gain value of Y on X represents the degree to which the uncertainty of the monitoring data of the detection device X is reduced after the information of X is interpreted by the information of the detection device Y, that is, the effective information increment provided by Y to X; the larger the value, the more the information of Y can eliminate the uncertainty of X, and the stronger the correlation between the two data. The entropy value of the single sensor information of the detection device X within the sliding window T is used to quantify the uncertainty of X's own data.

[0058] The sliding entropy change rate is an indicator that quantifies the dynamic change of the entropy value of a single detection device within a continuous sliding window. It primarily reflects the degree of abrupt changes in the distribution of monitored data and can capture abnormal fluctuations in entropy values. When a detection device experiences performance drift, parameter anomalies, or malfunctions, its entropy will show significant abrupt changes. Therefore, this indicator is an important temporal characteristic for assisting fault diagnosis, as detailed below;

[0059] ;

[0060] The sliding information entropy change rate of the detection device at time node t is expressed as a percentage; t is the current time node being calculated, representing the current sliding window (consistent with the previous sliding window T time sequence); t-1 is the previous time node of t, representing the previous sliding window (a continuous time node with the same step size as t). The information entropy value of the detection device at time node t (which can be the information entropy value of a single sensor or the joint information entropy value, depending on the diagnostic requirements); The information entropy value of the detection device at time node t-1, and... They are the same type of entropy index.

[0061] In summary, for the multidimensional time-series monitoring data of the detection devices within a sliding window, the single-sensor information entropy is first calculated after discretization to quantify the inherent uncertainty of the data from a single detection device. Then, the joint information entropy is calculated based on the discretized data from two detection devices to characterize the overall uncertainty of the joint distribution of their data. Subsequently, the conditional information entropy is derived from the joint information entropy and the single-sensor information entropy to reflect the remaining uncertainty of the other device after the information of one device is known. The information gain is then calculated to quantify the effective information increment by which one device eliminates uncertainty for the other device. Finally, the rate of change of the sliding information entropy is calculated based on the entropy value results of the continuous sliding window to capture the dynamic abrupt change amplitude of the entropy value of the detection device in the time dimension.

[0062] S4. Health Detector Subset Filtering: Specifically, a subset of health detection devices is generated through dynamic filtering. This subset is based on the historical anomaly coefficients of each detection device within a recent continuous sliding window. The health detection device subset is generated after dynamic identification and filtering of healthy detection devices. In other words, the health detection device subset is not a fixed set of detection devices. Its generation is based on the historical anomaly coefficients of each detection device within a recent continuous sliding window. Based on these historical anomaly coefficients, the health status of all detection devices is continuously and dynamically identified and judged. Finally, detection devices with abnormal status are eliminated, and only those determined to be healthy are selected. These healthy detection devices together constitute the health detection device subset.

[0063] In some embodiments, step (S4) maintains two lists: a health list and a fault isolation list. Initially, all detection devices are in the health list. After each sliding window is processed or before the current sliding window is processed, the anomaly coefficient of each detection device is checked. For example, if the anomaly coefficient of a detection device is ≥4.0 for three consecutive windows, it is determined that it has a high probability of failure and is moved from the health list to the fault isolation list; when calculating statistical characteristics in the next sliding window, only the information entropy value of the detection devices in the current health list is used.

[0064] As can be seen, the steps for generating the subset of health detection devices include:

[0065] Before processing the current sliding window or after processing each sliding window, check the historical anomaly coefficients of each detection device;

[0066] If the abnormality coefficient of any detection device reaches the alarm threshold within a consecutive preset number of sliding windows, then the detection device will be moved from the healthy list to the fault isolation list.

[0067] The health detection devices constitute a subset of the detection devices listed in the health list;

[0068] In the initial state, all detection devices are included in the health list.

[0069] In some embodiments, any detection device in the fault isolation list is re-included in the health list during periodic periods or when a specific trigger condition is met. Specifically, combined with the dynamic filtering logic of the subset of health detection devices described above, this re-inclusion mechanism is used to avoid false isolation of detection devices due to non-permanent faults such as temporary interference or instantaneous parameter drift, thereby improving flexibility and detection device utilization. The specific technical solution is as follows:

[0070] The periodic time period can be set to a preset number of consecutive sliding windows (consistent with the timing of the sliding windows for anomaly detection, such as 10 consecutive sliding windows constituting a periodic check period). After each periodic time period ends, the status of all detection devices in the fault isolation list is automatically checked. The specific trigger condition is as follows: if a detection device in the fault isolation list has a calculated anomaly coefficient that is consistently below the alarm threshold within a preset number of consecutive sliding windows (e.g., 5 consecutive sliding windows), and the fluctuation range is within a preset normal range, then the detection device is determined to have recovered to a healthy state and meets the re-inclusion condition. Regardless of whether the re-inclusion is based on periodic time period checks or specific trigger conditions, when a detection device meets the re-inclusion requirement, it will be removed from the fault isolation list and re-added to the healthy list. After being re-included in the healthy list, the detection device will participate in the anomaly coefficient calculation for the next sliding window, simultaneously becoming part of the healthy detection device subset and participating in subsequent calculations.

[0071] S5. Anomaly Coefficient Calculation: For the current sliding window, calculate the mean μ and standard deviation σ using the information entropy values ​​of the subset of health detection devices determined in step (S4). Then, use these μ and σ to calculate the standard scores of all detection devices in the cabin (including those already in the fault isolation list), and take their absolute values ​​to obtain the anomaly coefficient.

[0072] The specific steps are as follows: calculate the mean μ:

[0073] ;

[0074] In the formula, d is the total number of health detection devices contained in the subset of health detection devices within the current sliding window t, which is determined by the filtering in step (S4) and is dynamically updated with the sliding window; , where is the target information entropy value (g=1,2,...,d) of the g-th health detection device in the subset of health detection devices within the current sliding window t. It can represent single sensor information entropy, joint information entropy, conditional information entropy, information gain, or sliding information entropy change rate; t is the currently processing sliding window (consistent with the time sequence node t, and unified with the time sequence dimension of the sliding information entropy change rate mentioned above).

[0075] Then calculate the standard deviation σ:

[0076] .

[0077] Then, the standard score for each detection device is calculated sequentially:

[0078] ;

[0079] In the formula, is the standard score of the f-th detection device in the cabin within the current sliding window t (f is the unique number of all detection devices). It can be positive or negative. Positive and negative only indicate the direction of the entropy value deviating from the mean (greater than the mean is positive, less than the mean is negative). The target information entropy value of the f-th detection device (including detection devices in the fault isolation list) within the current sliding window t, compared with... For the same type of information entropy value, ensure a consistent calculation benchmark.

[0080] Finally, calculate the anomaly coefficient:

[0081] ;

[0082] In the formula, is the anomaly coefficient of the f-th detection device in the cabin within the current sliding window t. It is a non-negative value and only reflects the degree to which the detection device deviates from the health benchmark.

[0083] S6. Tiered Early Warning: A tiered early warning mechanism is activated based on the magnitude of the anomaly coefficient. Step (S6) requires setting alarm thresholds, for example: This is normal. This is a warning (indicating potential performance issues). This is an alarm (indicating a high probability of a fault).

[0084] Second embodiment, Figure 2 The structural composition of the fault detection system 100 for the detection device is shown. Figure 2 In this system, anomaly coefficient calculation and graded early warning for the detection device are performed based on the current sliding window, specifically including:

[0085] Data acquisition module 101 is configured to acquire multi-dimensional time-series monitoring data of each detection device in the health detection device subset in real time;

[0086] Entropy calculation module 102 is connected to data acquisition module and is configured to calculate the information entropy value of each detection device in the health detection device subset within the current sliding window.

[0087] Anomaly detection module 103 is connected to entropy calculation module and is configured to calculate mean and standard deviation based on information entropy values ​​of a subset of health detection devices. Then, based on mean and standard deviation, it calculates standard scores corresponding to the information entropy values ​​of all detection devices in the current sliding window. The absolute value of the standard scores corresponding to each detection device is taken to obtain the anomaly coefficient of each detection device.

[0088] The fault warning module 104 is connected to the anomaly detection module and is configured to perform graded warnings based on the anomaly coefficient.

[0089] In some embodiments, the output interface of the fault early warning module 104 is connected to the fire control panel or centralized monitoring system of the energy storage power station, and is used to trigger alarms in the station, generate maintenance work orders, or upload them to the remote monitoring center.

[0090] The fault detection system of this embodiment fully implements the fault detection method described above. Through the coordinated linkage of four modules, it achieves real-time and accurate fault detection of the composite fire detection device in the energy storage compartment, effectively avoiding the shortcomings of existing technologies. It can accurately identify the performance drift, minor faults and serious faults of the detection device, while supporting the recovery monitoring of the faulty device, forming a complete closed-loop management system. This provides reliable technical support for the fire safety of energy storage power stations, taking into account detection accuracy, robustness and engineering practicality.

[0091] The above are merely preferred embodiments of this application and do not limit the patent scope of this application. Any equivalent structural or procedural transformations made using the content of this application's specification and drawings, or direct or indirect applications in other related technical fields, are similarly included within the patent scope of this application.

Claims

1. A fault detection method for a detection device, which calculates the anomaly coefficient and provides graded early warning based on the current sliding window, characterized in that, include: Acquire multi-dimensional time-series monitoring data from each detection device; Calculate the information entropy value of each detection device within the current sliding window; The mean and standard deviation are calculated based on the information entropy values ​​of a subset of health detection devices. Then, based on the mean and standard deviation, the standard scores corresponding to the information entropy values ​​of all detection devices within the current sliding window are calculated. The absolute value of the standard score corresponding to each detection device is taken to obtain the anomaly coefficient of each detection device, and the detection device is given a graded early warning based on the anomaly coefficient.

2. The fault detection method for the detection device according to claim 1, characterized in that, The subset of health detection devices is generated after each sliding window is processed or before the current sliding window is processed.

3. The fault detection method for the detection device according to claim 1, characterized in that, The subset of health detection devices is generated by dynamically identifying and filtering health detection devices based on the historical anomaly coefficients of each detection device within a recent continuous sliding window.

4. The fault detection method for the detection device according to any one of claims 1 to 3, characterized in that, The steps for generating the subset of health detection devices include: Before processing the current sliding window or after processing each sliding window, check the historical anomaly coefficients of each detection device; If the abnormality coefficient of any detection device reaches the alarm threshold within a consecutive preset number of sliding windows, then the detection device will be moved from the healthy list to the fault isolation list. The health detection devices constitute a subset of the detection devices listed in the health list; In the initial state, all detection devices are included in the health list.

5. The fault detection method for the detection device according to claim 4, characterized in that, Any detection device in the fault isolation list is re-included in the health list during a periodic period or when a specific trigger condition is met.

6. The fault detection method for the detection device according to claim 1, characterized in that, The information entropy value includes single sensor information entropy, conditional information entropy, information gain, joint information entropy, and sliding information entropy change rate.

7. The fault detection method for the detection device according to claim 1, characterized in that, The length of the sliding window is 10 to 60 seconds.

8. The fault detection method for the detection device according to claim 1, characterized in that, The sliding window has a sliding step size of 1 to 5 seconds.

9. A fault detection system for a detection device, which calculates the anomaly coefficient and provides graded early warning based on the current sliding window, characterized in that, include: The data acquisition module is configured to acquire multi-dimensional time-series monitoring data from each detection device in real time. The entropy calculation module is connected to the data acquisition module and is configured to calculate the information entropy value of each detection device within the current sliding window. The anomaly detection module is connected to the entropy calculation module. It is configured to calculate the mean and standard deviation based on the information entropy values ​​of a subset of health detection devices. Then, based on the mean and standard deviation, it calculates the standard score corresponding to the information entropy value of each detection device in the current sliding window. The absolute value of the standard score corresponding to each detection device is taken to obtain the anomaly coefficient of each detection device. The fault warning module is connected to the anomaly detection module and is configured to perform graded warnings based on the anomaly coefficient.

10. The fault detection system for the detection device according to claim 9, characterized in that, The output interface of the fault early warning module is connected to the fire control panel or centralized monitoring system of the energy storage power station, and is used to trigger alarms in the station, generate maintenance work orders, or upload them to the remote monitoring center.

Images