Robust failure judgment method, system and equipment for low-cost indoor environment sensor group and medium

Abstract

The invention discloses a robust failure judgment method, system and equipment for a low-cost indoor environment sensor group and a medium, and the core of the method is to construct a serial five-step processing assembly line, utilize redundant information of the sensor group and combine robust statistics and statistical process control theories to judge the robustness failure of the indoor environment sensor group. According to the method, accurate decoupling of environmental events and sensor faults is achieved, and the problems that an existing method cannot distinguish environmental fluctuation and sensor degradation, and judgment lacks robustness and quantitative basis are solved. According to the method, the instantaneous deviation is judged, and misinformation caused by short-term fluctuation is avoided by setting the cumulant and the minimum duration time. The sensitivity to progressive degradation is improved.

Description

Technical Field

[0001] This invention relates to the field of sensor failure prediction technology, and in particular to a robust failure determination method, system, device and medium for low-cost indoor environmental sensor arrays. Background Technology

[0002] In scenarios such as indoor air quality monitoring, smart buildings, and industrial environmental safety, low-cost environmental sensors are widely used due to their low price, ease of deployment, and suitability for large-scale deployment. These sensors are mainly used to monitor temperature and humidity, carbon dioxide (CO2), total volatile organic compounds (TVOC), and particulate matter (PM2.5). 2.5 / PM 10 Indicators such as noise. However, due to the limitations of their sensing materials and circuit components, these sensors usually have the following problems: (1) Performance degradation: As the usage time increases, the sensitivity and accuracy gradually decrease, resulting in distortion of the output data; (2) Environmental interference: For example, events such as human activities, cooking, smoking, and opening windows for ventilation can cause the sensor output to fluctuate sharply, but these fluctuations are not equipment failures; (3) Limited service life: Low-cost sensors often experience a significant performance decline after a few months or a year of use, requiring replacement or recalibration.

[0003] Therefore, accurately assessing the health status of low-cost environmental sensors, especially determining whether they have failed, is a key scientific and engineering problem for ensuring the long-term stable operation of sensor networks.

[0004] Existing methods for determining the failure of low-cost sensors mainly rely on threshold determination, comparison with high-precision instruments, drift modeling, and manual inspection, but all of these methods have significant shortcomings.

[0005] The principle of the threshold method: Based on historical experience or data provided by the sensor manufacturer, a fixed threshold is set. When the sensor output deviates from the reference value by more than this threshold, it is considered a failure. The shortcomings of the threshold method are as follows: 1) The threshold is highly empirical and limited, and cannot adapt to different environmental scenarios; 2) It cannot identify gradual performance degradation, and can only detect sudden failures.

[0006] The principle of the comparison method with high-precision instruments is to simultaneously collect data from a low-cost sensor and a standard laboratory instrument, and determine whether the sensor has failed by comparing the differences. The shortcomings of the comparison method are as follows: 1) It requires expensive high-precision equipment as a benchmark, making it unsuitable for large-scale deployment; 2) It cannot perform real-time automated monitoring, and the data collection and comparison cycle is long.

[0007] The principle of drift-based modeling is to fit and model the changes in sensor output over time, identify its drift patterns, and predict future deviations. The shortcomings of drift-based modeling are as follows: 1) The model accuracy strongly depends on long-term, stable, and complete historical data, lacking adaptability; 2) Drift modeling easily misinterprets real environmental changes as sensor degradation.

[0008] The principle of manual maintenance is based on periodic manual sampling and comparison. If a decrease in sensor accuracy is found, the sensor is replaced or calibrated. The shortcomings of manual maintenance are as follows: 1) High cost, strong reliance on manual labor, and lack of real-time performance and large-scale applicability; 2) Blind spots are prone to occur between maintenance cycles, resulting in failure signals not being captured in time. Summary of the Invention

[0009] Objectives of the Invention: The first objective of this invention is to provide a robust failure determination method for low-cost indoor environmental sensor arrays that effectively distinguishes between environmental fluctuations and sensor degradation; the second objective of this invention is to provide a robust failure determination system for low-cost indoor environmental sensor arrays; the third objective of this invention is to provide an electronic device; and the fourth objective of this invention is to provide a computer-readable storage medium.

[0010] To achieve the first objective mentioned above, the technical solution of the robust failure determination method provided by this invention is as follows:

[0011] The robust failure determination method provided by this invention comprises an environmental sensor group consisting of N identical environmental sensors, including:

[0012] The measured values ​​of each environmental sensor in the environmental sensor group are collected at a frequency to obtain the measurement data vector set at time t. The rate of change and / or coefficient of variation of the measured values ​​of each environmental sensor within the sliding window are calculated. If the rate of change exceeds the threshold and / or the coefficient of variation increases abnormally, a mask signal is generated to suspend subsequent fault determination. If the duration of the rate of change of the measured value exceeding the threshold exceeds the preset time, the mask signal is canceled and subsequent fault determination continues.

[0013] Fault determination includes: using the median of the measured data vector set as an initial estimate, the median absolute deviation (MAD) as a measure of dispersion, and calculating the robust Z-score, z. i,t ;

[0014] The trend of the measurement data vector set is extracted by exponentially weighted moving average. The median ± 3 × MAD rule is applied to the residual series to identify sensor anomalies, or the start time of drift is determined by the cumulative sum control chart CUSUM. If the statistic C... t Sensor anomalies are identified when the time exceeds the threshold D for an extended period.

[0015] When a sensor malfunction is detected, if the malfunction is consistent with the trend of population fluctuation, it is not considered a fault; otherwise, it is considered a fault.

[0016] Finally, the sensor is scored using the following formula, and the sensor's failure is determined based on the score:

[0017]

[0018] Among them, deviation deduction points =|z i,t ×10, Trend Deduction =(C i,t / h)×100, correlation deduction Calculated from the Pearson correlation coefficient ρ, if ρ < 0.8, = (0.8-ρ)×200, if ρ≥0.8, =0, α=0~1, β=0~1, γ=0.5~1, and the values ​​of α, β, and γ satisfy: α+β+γ=1, γ ‌≥‌0.5>β≥0.25>α≥0.15; h =5, C i,t Let be the cumulative error of sensor i at time t.

[0019] Preferably, the rate of change of the measured values ​​of each environmental sensor within the sliding window. The calculation formula is as follows:

[0020]

[0021] in, Let be the measurement value of sensor i at time t. Let be the measurement value of sensor i at the beginning of the short sliding window. The length of the short sliding window. This is the sampling time interval. Preferably, =3min; That is, data is collected every 10 seconds.

[0022] Coefficients of variation of each environmental sensor within the sliding window The calculation formula is as follows:

[0023]

[0024] in, This represents the standard deviation of the data within window W. Mean is the average value of the data within window W.

[0025] Preferably, the specific steps for generating the mask signal to pause subsequent fault determination are as follows: during fault determination, when calculating the robust Z-score, set the input residual in the robust Z-score to 0, or maintain the cumulative statistic C. t constant.

[0026] Preferably, in the step of extracting the trend of the measurement data vector set through exponentially weighted moving average and applying the median ± 3 × MAD rule to identify sensor anomalies in the residual sequence, the recursive formula for calculating the exponentially weighted moving average is as follows:

[0027]

[0028] Let be the smoothing statistic of sensor i at time t, and λ be the smoothing factor. Let λ be the robust Z-score of sensor i at time t. Preferably, λ = 0.1.

[0029] Preferably, in the step of determining the start time of drift occurrence through the cumulative sum control chart (CUSUM), the tabular CUSUM calculation formula is as follows:

[0030]

[0031]

[0032] in, and Let be the positive and negative cumulative errors of sensor i at time t, respectively, and k be a reference value. Preferably, k = 0.5.

[0033] Preferably, in the step of determining that if a sensor is malfunctioning but its fluctuation trend is consistent with the population trend, it is not considered a fault when the sensor is identified as malfunctioning, the target sensor S is calculated. i With group reference sequence In window W corr The Pearson correlation coefficient ρ within the range i,t To determine whether the fluctuations of abnormal sensors are consistent with the overall fluctuation trend of the population.

[0034] As a preferred option If the value is 80-100, the sensor is considered normal. If the value is 50-80, the sensor is suspected of drifting. If the value is 0-50, the sensor is considered to be faulty.

[0035] To achieve the second objective mentioned above, the technical solution of the robust failure determination system provided by this invention is as follows:

[0036] This invention provides a robust failure determination system for a low-cost indoor environmental sensor array utilizing the above-described method, comprising:

[0037] The event masking module collects the measurement values ​​of each environmental sensor in the environmental sensor group at a frequency to obtain the measurement data vector set at time t. It calculates the rate of change and / or coefficient of variation of the measurement values ​​of each environmental sensor within a sliding window. If the rate of change exceeds a threshold and / or the coefficient of variation increases abnormally, a masking signal is generated to suspend subsequent fault determination. If the duration of the rate of change exceeding the threshold exceeds a preset time, the masking signal is canceled and subsequent fault determination continues.

[0038] The fault determination module includes a robust statistics module and a trend accumulation module. In the robust statistics module, the median of the measurement data vector set is used as an initial estimate, and the median absolute difference (MAD) is used as a measure of dispersion to calculate the robust Z-score. In the trend accumulation module, the trend of the measurement data vector set is extracted using an exponentially weighted moving average. Sensor anomalies are identified by applying the median ± 3 × MAD rule to the residual sequence, or the start time of drift is determined using the cumulative sum control chart (CUSUM). If the statistic C... t Sensor anomalies are identified when the time exceeds the threshold D for an extended period.

[0039] An independence verification module is used to exclude group environmental changes through correlation analysis;

[0040] A confidence scoring module is used to quantify the failure probability.

[0041] To achieve the third objective mentioned above, the technical solution for robust failure determination of the electronic device provided by the present invention is as follows:

[0042] This invention provides an electronic device for robust failure determination of a low-cost indoor environmental sensor array. The electronic device includes a memory and a processor. The memory stores a computer program. When the computer program is executed by the processor, the processor performs the steps of the robust failure determination method described above.

[0043] To achieve the fourth objective mentioned above, the technical solution for a robust failure determination computer-readable storage medium provided by this invention is as follows:

[0044] The present invention provides a computer-readable storage medium for robust failure determination of a low-cost indoor environmental sensor array, wherein a computer program is stored thereon, and when the computer program is executed, the steps of the robust failure determination method described above are implemented.

[0045] Beneficial effects: Compared with the prior art, the present invention has the following significant advantages:

[0046] (1) Robust event recognition mechanism: This invention introduces an "event recognition" mechanism before failure detection, which can effectively distinguish between real environmental changes (such as personnel activities, cooking, and industrial process adjustments) and sensor degradation, avoiding misjudgment. This mechanism solves the core pain point of "environmental changes being misjudged as failures" in traditional methods.

[0047] (2) Combining population residual with robust Z-score: By calculating the residual between the sensor and the population median, and combining it with robust Z-score to determine anomalies, the detection's tolerance to noise and extreme values ​​is improved. Compared with single threshold determination, it is more adaptable and stable.

[0048] (3) Introduction of Time Accumulation Statistics (EWMA / CUSUM): This invention not only judges instantaneous deviations, but also avoids false alarms caused by short-term fluctuations by setting the cumulative amount and minimum duration. It improves the sensitivity to gradual degradation.

[0049] (4) Correlation check ensures independence: Before the final judgment, correlation analysis between sensors is used to confirm whether the anomaly is unique to an individual sensor, thus avoiding the misjudgment of a group increase as a single sensor failure. This significantly reduces the system's false alarm rate.

[0050] (5) Probability-based failure determination and confidence output: The output is not limited to the binary result "failure / normal", but provides "possible failure" and quantifies the confidence level. This makes it easier for maintenance personnel to formulate differentiated maintenance strategies based on probability scores and achieve refined management.

[0051] (6) Wide range of applications and strong scalability: It can be applied to various scenarios such as indoor air quality monitoring, industrial Internet of Things, smart cities, and smart buildings; the framework design is scalable and can connect more sensor types (such as formaldehyde, ozone, and particulate matter sensors) as needed. Attached Figure Description

[0052] Figure 1 A flowchart of a robust failure determination method provided in an embodiment of this application.

[0053] Figure 2 A flowchart of another robust failure determination method provided in this application embodiment.

[0054] Figure 3 This is a schematic diagram of a robust failure determination system provided in an embodiment of this application. Detailed Implementation

[0055] The technical solution of the present invention will be further described below with reference to the accompanying drawings.

[0056] This application proposes a robust failure determination method based on multi-level statistical signal processing. Its core lies in constructing a serial five-step processing pipeline, utilizing the redundant information of a sensor cluster, and combining robust statistics and statistical process control (SPC) theory to achieve precise decoupling between "environmental events" and "sensor failures." This addresses the problems of existing methods failing to distinguish between environmental fluctuations and sensor degradation, and lacking robustness and quantitative basis in their determinations.

[0057] like Figure 1 As shown in the figure, this application provides a robust failure determination method for a low-cost indoor environmental sensor array, including the following steps S1 to S3. Wherein:

[0058] S1. Collect the measured values ​​of each environmental sensor in the environmental sensor group according to the frequency to obtain the measurement data vector set at time t. Calculate the rate of change and / or coefficient of variation of the measured values ​​of each environmental sensor within the sliding window. If the rate of change exceeds the threshold and / or the coefficient of variation increases abnormally, generate a mask signal to pause subsequent fault determination. If the duration of the rate of change exceeding the threshold exceeds the preset time, cancel the mask signal and continue subsequent fault determination.

[0059] Among them, the rate of change of the measured values ​​of each environmental sensor within the sliding window The calculation formula is as follows:

[0060]

[0061] in, Let be the measurement value of sensor i at time t. Let be the measurement value of sensor i at the beginning of the short sliding window. The length of the short sliding window (e.g., 3 minutes). The sampling time interval (e.g., sampling once every 10 seconds).

[0062] Coefficients of variation of each environmental sensor within the sliding window The calculation formula is as follows:

[0063]

[0064] in, This represents the standard deviation of the data within window W. Mean is the average value of the data within window W.

[0065] The specific steps for generating the mask signal to pause subsequent fault determination are as follows: during fault determination, when calculating the robust Z-score, set the input residual in the robust Z-score to 0, or maintain the cumulative statistic C. t constant.

[0066] S2. Fault determination includes: using the median of the measurement data vector set as an initial estimate, the median absolute deviation (MAD) as a measure of dispersion, and calculating the robust Z-score, i.e., z. i,t .

[0067] The trend of the measurement data vector set is extracted by exponentially weighted moving average. Sensor anomalies are identified by applying the median ± 3 × MAD rule to the residual sequence, or the start time of drift is determined by the cumulative sum control chart (CUSUM). If the statistic C... t Sensor anomalies are identified when the time exceeds the threshold D for an extended period.

[0068] When a sensor malfunction is detected, if the malfunction is consistent with the trend of population fluctuation, it is not considered a fault; otherwise, it is considered a fault.

[0069] The recursive formula for calculating the exponentially weighted moving average is as follows:

[0070]

[0071] Let λ be the smoothing statistic of sensor i at time t, and let λ be the smoothing factor (e.g., 0.1). Let Z be the robust Z-score of sensor i at time t.

[0072] Preferably, in the step of determining the start time of drift occurrence through the cumulative sum control chart (CUSUM), the tabular CUSUM calculation formula is as follows:

[0073]

[0074]

[0075] In the formula, and These are the positive and negative cumulative errors of sensor i at time t, respectively, and k is a reference value (e.g., 0.5).

[0076] By calculating the target sensor S i With group reference sequence In window W corr The Pearson correlation coefficient ρ within the range i,t To determine whether the fluctuations of abnormal sensors are consistent with the overall fluctuation trend of the population.

[0077] S3. Finally, the sensor is scored using the following formula, and the sensor failure is determined based on the score result:

[0078]

[0079] Among them, deviation deduction points =|z i,t ×10, Trend Deduction =(C i,t / h)×100, correlation deduction Calculated from the Pearson correlation coefficient ρ, if ρ < 0.8, = (0.8-ρ)×200, if ρ≥0.8, =0, α=0~1, β=0~1, γ=0.5~1, and the values ​​of α, β, and γ satisfy: α+β+γ=1, γ ‌≥‌0.5>β≥0.25>α≥0.15; h =5, C i,t Let be the cumulative error of sensor i at time t.

[0080] The following detailed description of this application will be based on specific application examples.

[0081] Application Examples

[0082] like Figure 2 As shown, the system architecture, data preprocessing, and core steps provided in this embodiment are as follows:

[0083] (I) System Architecture and Data Preprocessing

[0084] 1. Sensor Cluster Topology and Data Definition

[0085] Suppose that N identical environmental sensors (denoted as S1, S2, …, S) are deployed in a physically connected area (such as an open-plan office). N ).

[0086] Sampling frequency (f) s ): It is recommended to set it to 10 to 60 seconds.

[0087] Data vector: The original measurement vector at time t is .

[0088] 2. Data Stream Processing Pipeline

[0089] The data passes through the following five modules in sequence:

[0090] 1. Event Masking Module: Identifies high-frequency sudden events and freezes accumulated faults.

[0091] 2. Robust Profiling module: Calculates the Z-score for robustness to interference based on the population median and MAD.

[0092] 3. Trend Accumulation Module: Amplifies minute drift signals using EWMA / CUSUM.

[0093] 4. Independence Check Module: Eliminates group environmental changes through correlation analysis.

[0094] 5. Confidence Scoring Module: Quantifies the probability of failure.

[0095] (II) Core Steps

[0096] 1. Step One: Event Detection

[0097] 1.1 Technical Pain Points and Goals

[0098] Taking "cooking" as an example, cooking can make PM within minutes. 2.5 Concentration surged to >1000 µg / m 3 Therefore, traditional methods often misjudge changes in indoor environmental parameters caused by "cooking" as sensor "failures." The goal of this step is to identify environmental events through feature engineering and generate a "masking signal" to pause subsequent fault determination.

[0099] 1.2 Feature Extraction Algorithm

[0100] Using sliding window analysis, a short window W is defined. short (Approximately 3-5 minutes) and long window W long (Approximately 30-60 minutes).

[0101] 1.2.1 Calculation of the rate of change (Slope)

[0102] Calculate the slope m of sensor i at time t. i,t :

[0103]

[0104] Parameter setting reference (example):

[0105] PM 2.5 Cooking incidents: PM caused by cooking 2.5 The ascent rate is extremely rapid, typically 20–100 µg / m 3 / min. It is recommended to set a threshold θ. slope_PM ≈10 µg / m3 / min.

[0106] CO2 gathering of people: Set the rise rate threshold >10~20 ppm / min.

[0107] 1.2.2 Volatility Characteristics

[0108] Calculate the coefficient of variation (CV) within the window:

[0109]

[0110] The CV value of a real diffusion event is usually in a stable range, while circuit faults often lead to an abnormally high CV value.

[0111] 1.3 Event Judgment Logic and Masking Mechanism

[0112] Decision rules are established based on extracted features. A boolean variable, Event_Flag, is defined. i,t :

[0113]

[0114] Duration constraint (θ) time Real-world events typically last 15 to 60 minutes. If a high value persists for more than several hours, the Event flag is removed, and the event is considered a failure.

[0115] 1.4 Supplement to Core Protection Points: Masking Action

[0116] This is the key technical action in this step. When Event_Flag i,t When the result is TRUE, the system must actively pause the cumulative calculation in step 3. Specifically, in the cumulative formula of step 3, either set the input residual to 0, or maintain the cumulative statistic C. t Unchanged (C) t =C t-1 This allows sensors to be exempted from health assessments during "incidents," completely eliminating the risk of false alarms.

[0117] 2. Step Two: Population Residual and Robust Z-score Calculation

[0118] 2.1 Deficiencies of Traditional Methods

[0119] The traditional Z-score (z=(x - µ) / σ) has a serious flaw in sensor arrays. If even one sensor fails (such as outputting full scale), the mean µ and standard deviation σ will be severely skewed, resulting in masking and swamping effects.

[0120] 2.2 Introduction of Robust Statistics

[0121] Robust statistics with a 50% collapse point are used: median and absolute median difference (MAD).

[0122] 2.3 Detailed Calculation Steps

[0123] 2.3.1 Calculating the Group Baseline

[0124]

[0125] 2.3.2 Calculate the original residuals

[0126]

[0127] 2.3.3 Calculate the absolute median deviation (MAD)

[0128]

[0129] 2.3.4 Key Parameter: Consistency Correction Factor

[0130] To ensure that the MAD-based Z-score aligns with the standard normal distribution, a consistency correction factor of 1.4826 is introduced. The robust standard deviation estimator is:

[0131]

[0132] 2.3.5 Calculate the robust Z-score

[0133]

[0134] Where ε is a local minimum (e.g., 1 µg / m³). 3 ), to prevent division by zero errors.

[0135] 3. Step Three: Residual Cumulative Analysis (EWMA / CUSUM)

[0136] This step is used to capture small but persistent progressive drifts, which are easily overlooked in single-point detection.

[0137] 3.1 Option A: Exponentially Weighted Moving Average (EWMA)

[0138] EWMA: Exponentially Weighted Moving Average.

[0139] Recursive formula:

[0140]

[0141] Smoothing factor λ: It is recommended to set λ = 0.1~0.2. Smaller λ is more sensitive to small drifts.

[0142] Judgment threshold: usually set to |E i,t |>3 triggers an alert.

[0143] 3.2 Option B: Cumulative Sum Control Chart (CUSUM)

[0144] CUSUM: Cumulative Sum Control Chart. It is used to calculate the cumulative sum of deviations (divided into positive cumulative C...). t + and negative cumulative C t - ).

[0145] CUSUM is suitable for determining the start time of drift. A tabular version of CUSUM is used:

[0146]

[0147]

[0148] Reference value k: Set to half of the target detection offset. When detecting a 1σ offset, k = 0.5.

[0149] Threshold h: h is usually set to 5.

[0150] 3.3 Duration Confirmation

[0151] "Possible failure" is only confirmed when the statistic continues to exceed the threshold time D.

[0152] Drift failure: D = 24~48 hours.

[0153] Stuck failure: D = 1~6 hours.

[0154] 4. Step Four: Correlation Check

[0155] 4.1 Purpose

[0156] Distinguish between "sensor malfunction" and "localized real contamination." If a sensor reading is abnormal but consistent with the overall trend (high correlation), it may be a localized real event; if the correlation is lost, it is determined to be a malfunction.

[0157] 4.2 Algorithm Implementation

[0158] Calculate the target sensor S i With group reference sequence In window W corr The Pearson correlation coefficient ρ within the range i,t .

[0159] Parameter settings:

[0160] Window size W corr We recommend 12-24 hours to cover the entire event period.

[0161] Determination threshold ρ thresh :

[0162] ρ i,t >0.7: Normal or requires only calibration (Offset).

[0163] ρ i,t <0.5: Determined to be completely ineffective.

[0164] 5. Step Five: Confidence Scoring and Output (Scoring & Output)

[0165] Construct a health confidence score (Score), ranging from 0 to 100.

[0166] 5.1 Scoring Function

[0167]

[0168] Deductions:

[0169] Deviation deduction :|z i,t |×10.

[0170] Trend deduction :(C i,t / h)×100.

[0171] Relevance deduction If ρ < 0.8, deduct (0.8 - ρ) × 200 points.

[0172] Recommended weights: α=0.2, β=0.3, γ=0.5. Relevance has the highest weight because it is the strongest evidence of hardware damage.

[0173] 5.2 Output Decision Table

[0174] Table 1 Decision Table

[0175]

[0176] 6. Summary and Key Parameter Table

[0177] This scheme addresses the high false alarm rate of low-cost sensors by introducing a MAD consistency correction factor (valued at 1.4826) and an event masking mechanism, and improves the detection sensitivity for minute drifts by utilizing CUSUM and correlation checks.

[0178] Table 2 Summary of Implementation Parameters

[0179]

[0180] like Figure 3 As shown, in some embodiments, a robust failure determination system for a low-cost indoor environmental sensor array utilizing the above-described method is provided, including...

[0181] Event masking module 11: The event masking module collects the measurement values ​​of each environmental sensor in the environmental sensor group at a frequency to obtain the measurement data vector set at time t, calculates the rate of change and / or coefficient of variation of the measurement values ​​of each environmental sensor within the sliding window, and generates a masking signal to suspend subsequent fault determination if the rate of change exceeds the threshold and / or the coefficient of variation increases abnormally. If the duration of the rate of change exceeding the threshold exceeds a preset time, the masking signal is canceled and subsequent fault determination continues.

[0182] Fault determination module 12 includes a robust statistics module 121 and a trend accumulation module 122. In the robust statistics module, the median of the measurement data vector set is used as an initial estimate, and the median absolute difference (MAD) is used as a measure of dispersion to calculate the robust Z-score. In the trend accumulation module, the trend of the measurement data vector set is extracted using an exponentially weighted moving average. Sensor anomalies are identified by applying the median ± 3 × MAD rule to the residual sequence, or the start time of drift is determined using the cumulative sum control chart (CUSUM). If the statistic C... t Sensor anomalies are identified when the time exceeds the threshold D for an extended period.

[0183] Independence verification module 13, which is used to exclude group environmental changes through correlation analysis;

[0184] Confidence scoring module 14, which is used to quantify the failure probability.

[0185] In some embodiments, an electronic device is provided for robust failure determination of a low-cost indoor environmental sensor array. The electronic device includes a memory and a processor. The memory stores a computer program that, when executed by the processor, causes the processor to perform the steps of a robust failure determination method.

[0186] In some embodiments, a computer-readable storage medium is provided for robust failure determination of a low-cost indoor environmental sensor array, having stored thereon a computer program that, when executed, implements the steps of a robust failure determination method.

Claims

1. A robust failure determination method for a low-cost indoor environmental sensor array, wherein the environmental sensor array consists of N identical environmental sensors, characterized in that, include: The measured values ​​of each environmental sensor in the environmental sensor group are collected at a frequency to obtain the measurement data vector set at time t. The rate of change and / or coefficient of variation of the measured values ​​of each environmental sensor within the sliding window are calculated. If the rate of change exceeds the threshold and / or the coefficient of variation increases abnormally, a mask signal is generated to suspend subsequent fault determination. If the duration of the measured value change rate exceeding the threshold exceeds the preset time, the mask signal is canceled and subsequent fault determination continues; Fault determination includes: using the median of the measured data vector set as an initial estimate, the median absolute deviation (MAD) as a measure of dispersion, and calculating the robust Z-score, z. i,t ; The trend of the measurement data vector set is extracted by exponentially weighted moving average. The median ± 3 × MAD rule is applied to the residual series to identify sensor anomalies, or the start time of drift is determined by the cumulative sum control chart CUSUM; if the statistic C t Sensor anomalies are identified when the time exceeds the threshold D for an extended period. When a sensor malfunction is detected, if the malfunction is consistent with the trend of population fluctuation, it is not considered a fault; otherwise, it is considered a fault. Finally, the sensor is scored using the following formula, and the sensor's failure is determined based on the score: Among them, deviation deduction points =|z i,t ×10, Trend Deduction =(C i,t / h)×100, correlation deduction Calculated from the Pearson correlation coefficient ρ, if ρ < 0.8, = (0.8-ρ)×200, if ρ≥0.8, =0, α=0~1, β=0~1, γ=0.5~1, and the values ​​of α, β, and γ satisfy: α+β+γ=1, γ ‌≥‌0.5>β≥0.25>α≥0.15; h =5, C i,t Let be the cumulative error of sensor i at time t.

2. The robust failure determination method according to claim 1, characterized in that, Rate of change of measured values ​​of each environmental sensor within the sliding window The calculation formula is as follows: in, Let be the measurement value of sensor i at time t. Let be the measurement value of sensor i at the beginning of the short sliding window. The length of the short sliding window. The sampling time interval; the coefficient of variation of each environmental sensor within the sliding window. The calculation formula is as follows: in, Let W be the standard deviation of the data within window W. This represents the mean of the data within window W.

3. The robust failure determination method according to claim 1, characterized in that, The specific steps for generating a mask signal to pause subsequent fault determination are as follows: During fault determination, when calculating the robust Z-score, set the input residual in the robust Z-score to 0, or maintain the cumulative statistic C. t constant.

4. The robust failure determination method according to claim 1, characterized in that, In the step of extracting the trend of the measurement data vector set through exponentially weighted moving average and applying the median ± 3 × MAD rule to identify sensor anomalies in the residual sequence, the recursive formula for calculating the exponentially weighted moving average is as follows: in, Let be the smoothing statistic of sensor i at time t, and λ be the smoothing factor. Let Z be the robust Z-score of sensor i at time t.

5. The robust failure determination method according to claim 1, characterized in that, In the step of determining the start time of drift through the cumulative sum control chart (CUSUM), the tabular CUSUM calculation formula is as follows: in, and denoted as the positive and negative cumulative errors of sensor i at time t, respectively, and k is the reference value.

6. The robust failure determination method according to claim 1, characterized in that, In the step of determining that if a sensor malfunction is detected but its trend aligns with the overall population fluctuation, it is not considered a fault, the target sensor S is calculated. i With group reference sequence In window W corr The Pearson correlation coefficient ρ within the range i,t To determine whether the fluctuations of abnormal sensors are consistent with the overall fluctuation trend of the population.

7. The robust failure determination method according to claim 1, characterized in that, If the value is 80-100, the sensor is considered normal. If the value is 50-80, the sensor is suspected of drifting. If the value is 0-50, the sensor is considered to be faulty.

8. A robust failure determination system for a low-cost indoor environmental sensor array utilizing the method of claim 1, comprising: The event masking module collects the measurement values ​​of each environmental sensor in the environmental sensor group at a frequency to obtain the measurement data vector set at time t. It calculates the rate of change and / or coefficient of variation of the measurement values ​​of each environmental sensor within a sliding window. If the rate of change exceeds a threshold and / or the coefficient of variation increases abnormally, a masking signal is generated to suspend subsequent fault determination. If the duration of the rate of change exceeding the threshold exceeds a preset time, the masking signal is canceled and subsequent fault determination continues. The fault determination module includes a robust statistics module and a trend accumulation module. In the robust statistics module, the median of the measurement data vector set is used as the initial estimate, and the median absolute difference (MAD) is used as the dispersion measure to calculate the robust Z-score. In the trend accumulation module, the trend of the measurement data vector set is extracted by exponentially weighted moving average. Sensor anomalies are identified by applying the median ± 3 × MAD rule to the residual sequence, or the start time of drift is determined by the cumulative sum control chart CUSUM. If the statistic C... t Sensor anomalies are identified when the time exceeds the threshold D for an extended period. An independence verification module is used to exclude group environmental changes through correlation analysis; A confidence scoring module is used to quantify the failure probability.

9. An electronic device for robust failure detection of a low-cost indoor environmental sensor array, characterized in that, The electronic device includes a memory and a processor. The memory stores a computer program, which, when executed by the processor, causes the processor to perform the steps of the robust failure determination method as described in any one of claims 1-7.

10. A computer-readable storage medium for robust failure determination of a low-cost indoor environmental sensor array, having stored thereon a computer program, characterized in that, When the computer program is executed, it implements the steps of the robust failure determination method as described in any one of claims 1-7.

Images