A deep-cold hydrogen storage tank airtightness detection system

Through the combination of environmental data analysis and machine learning algorithms, detection conditions are dynamically adjusted, and slight leakage in hydrogen storage tanks are identified and responded to, which solves the problem of slow response in traditional systems and improves the accuracy and safety of sealing detection.

CN118551329BActive Publication Date: 2024-11-12ZHONGGU ENERGY EQUIP (JIANGSU) CO LTD
View PDF 0 Cites 0 Cited by

Patent Information

Application Number
CN202411019109.X
Authority / Receiving Office
CN · China
Patent Type
Patents(China)
Current Assignee / Owner
Filing Date
2024-07-29
Publication Date
2024-11-12
Estimated Expiration
2044-07-29

AI Technical Summary

Technical Problem

The traditional deep-cold hydrogen storage tank sealing detection system responds slowly in terms of rapid fluctuations and micro leakage detection, and lacks adaptive adjustments, resulting in early leakage or defect inspection, increasing the safety risks and maintenance costs of hydrogen storage tanks.

Method used

The environmental data analysis module is used to collect and process temperature and pressure data through monitoring sensors, and use long and short-term memory networks and isolated forest algorithms to perform pattern recognition and abnormal pattern matching, dynamically adjust detection conditions, identify leakage signals in real time and generate detection parameter adjustment indicators to improve Accuracy and reliability of the detection.

Benefits of technology

Real-time dynamic adjustment of the airtightness detection of deep-cold hydrogen storage tanks is achieved, which reduces the false alarm rate, improves the reliability and response speed of detection, and reduces potential safety risks.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN118551329B_ABST
    Figure CN118551329B_ABST
Patent Text Reader

Abstract

The present invention relates to the field of airtightness detection technology, specifically to a deep-cold hydrogen storage tank airtightness detection system, the system includes an environmental data analysis module, a detection strategy optimization module, an abnormal behavior recognition module and a detection result feedback module, the environmental data analysis module is based on an environmental monitoring sensor, collects temperature and pressure data, performs normalization processing, performs window segmentation on time series data, and performs pattern recognition on the fluctuation of the sealed environment to obtain an environmental feature vector. In the present invention, the long short-term memory network can perform in-depth learning on the complex relationship between historical and current environmental data, realize real-time dynamic adjustment of the detection strategy, and effectively respond to environmental changes. This kind of adaptive adjustment reduces the false alarm rate and improves the reliability of airtightness detection. The isolation forest algorithm can process and locate anomalies faster in feature extraction and abnormal pattern matching, especially in identifying tiny leakage signals, and reduce potential safety risks.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The present invention relates to the technical field of airtightness detection, and in particular to a deep-cold hydrogen storage tank airtightness detection system. Background Art

[0002] The field of tightness testing technology mainly focuses on evaluating and ensuring the gas or liquid sealing performance of containers or systems under predetermined conditions. This technology covers testing methods for sealing performance from pressure vessels to complex systems, including but not limited to pressure change testing, vacuum testing, tracer gas detection, acoustic emission testing and optical detection technology. Leakage testing plays an important role in ensuring equipment safety, environmental protection and improving product quality, especially in the fields of chemical industry, biomedicine, aerospace and aviation.

[0003] Among them, the deep-cold hydrogen storage tank tightness detection system is specially used to detect and ensure the tightness of storage tanks storing liquid hydrogen or cryogenic hydrogen under extremely low temperature conditions. Its main purpose is to prevent hydrogen leakage and ensure the safe operation of the storage tank. During the storage and transportation of liquid hydrogen, any tiny sealing defects may lead to hydrogen leakage, thereby bringing safety risks. Therefore, helium mass spectrometry leak detection, ultrasonic detection or pressure decay method are used to ensure that the storage tank still maintains excellent sealing performance in a low temperature environment.

[0004] Traditional systems such as pressure change testing and vacuum testing require a stable environment to ensure data accuracy, and are slow to respond to rapid fluctuations and tiny leaks, resulting in missed detection of early leaks or defects. In addition, there is a lack of sufficient support for data processing speed and adaptive adjustment, resulting in the inability to respond to safety threats in a timely and effective manner in industrial applications with rapidly changing demands, which not only increases the risk of hydrogen storage tanks, but also affects their maintenance costs and efficiency. Summary of the invention

[0005] The purpose of the present invention is to solve the shortcomings in the prior art and to propose a deep-cold hydrogen storage tank airtightness detection system.

[0006] In order to achieve the above object, the present invention adopts the following technical solution: a deep-cold hydrogen storage tank airtightness detection system, the system comprising:

[0007] The environmental data analysis module collects temperature and pressure data based on environmental monitoring sensors, performs normalization processing, performs window segmentation on time series data, and performs pattern recognition on the fluctuation of the sealed environment to obtain environmental feature vectors;

[0008] The detection strategy optimization module is based on the environmental feature vector and combines historical detection data to screen historical detection conditions that are consistent with the current environment through a long short-term memory network, and dynamically adjusts the hydrogen storage tank detection conditions to generate detection parameter adjustment indicators;

[0009] The abnormal behavior recognition module uses the isolation forest algorithm to extract features and match abnormal patterns on the data based on the detection parameter adjustment index and real-time monitoring data, identifies atypical behaviors and leakage signals in the sealing test, and obtains a leakage signal feature set;

[0010] The detection result feedback module evaluates the accuracy and reliability of the detection result based on the leakage signal feature set, displays the detection result using visualization technology, adjusts the operation process according to the detection result, and establishes a feedback analysis log.

[0011] The present invention is improved in that the step of window segmenting the time series data is specifically as follows:

[0012] Based on environmental monitoring sensors, temperature and pressure data are collected, normalized, and time series data is segmented using the formula:

[0013]

[0014] and

[0015]

[0016] Get the temperature window segmentation of time series data and pressure window segmentation ,in, is the normalized temperature data, is the normalized pressure data, Represents the window size, is the adjustment factor, To avoid small amounts of division by zero, is the index value;

[0017] The temperature window segmentation and pressure window segmentation of the time series data are analyzed using the formula:

[0018]

[0019] Generate a weighted mean for each temperature window segment ,in, and is the weight coefficient of temperature and pressure data, is the window size, and For the window temperature and pressure data points.

[0020] The present invention is improved in that the step of obtaining the environmental feature vector is specifically as follows:

[0021] Apply pattern recognition algorithms to identify key patterns in the time series segmented data using the formula:

[0022]

[0023] Pattern parameters extracted from segmented data, where is the index or time stamp of the time series, is the corresponding normalized temperature or pressure value, 、 and are the model coefficients, is the smoothing parameter;

[0024] Integrate the key patterns and parameters identified using the formula:

[0025]

[0026] Get the environmental feature vector ,in, and is the trend coefficient in linear regression, and is a parameter used to adjust the sensitivity of the model, is the key frequency in periodic analysis.

[0027] The present invention is improved in that the steps of screening the historical detection conditions are specifically as follows:

[0028] Based on the environmental feature vector and combined with historical detection data, the formula is adopted:

[0029]

[0030] Calculate the similarity of each historical data point ,in, is the first feature vector of the current environment dimension, The historical detection data Item dimension, It is The weight factor of the dimension, is the vector dimension;

[0031] Set a similarity threshold and filter historical data with similarity scores higher than the threshold using the formula:

[0032]

[0033] Select historical data points that meet the criteria ,in, is the selected benchmark similarity threshold, for The standard deviation of The historical detection data stronghold;

[0034] Based on filtered historical data points , using the formula:

[0035]

[0036] Get historical detection status that matches the current environment ,in, is the weight assigned according to the similarity score, From the screening process The historical detection data items obtained in .

[0037] The present invention is improved in that the step of obtaining the detection parameter adjustment index is specifically as follows:

[0038] Based on the historical detection conditions that are consistent with the current environment, the detection conditions of the hydrogen storage tank are dynamically adjusted using the formula:

[0039]

[0040] Get the difference after adjusting the detection conditions ,in, For the current environment The detection parameters, is the average value of the parameter in historical data, To adjust the weight coefficient of the parameter, To avoid division by zero errors for small positive numbers, is the total number of detection parameters;

[0041] According to the adjusted hydrogen storage tank detection conditions, calculate the detection parameter adjustment index using the formula:

[0042]

[0043] in, is the difference after adjustment for each test condition, is the coefficient used to balance the detection condition, is the total number of detection parameters, It is the detection parameter adjustment indicator.

[0044] The present invention is improved in that the steps of feature extraction and abnormal pattern matching are specifically as follows:

[0045] Based on the detection parameter adjustment index, the isolation forest algorithm is used to process the real-time monitoring data and perform preliminary outlier identification using the formula:

[0046]

[0047] Get anomaly score ,in, For a single data point monitored in real time, It is a data point With the average The Euclidean distance between is the sensitivity adjustment parameter, is the number of trees affecting the parameter, is the tree depth, It is a small amount that avoids the denominator being zero;

[0048] Based on the anomaly score, data points with scores higher than the threshold are selected as key anomalies. The formula is:

[0049]

[0050] in, is the baseline abnormality threshold, is the threshold adjustment factor, is the standard deviation of the anomaly scores, is the set of filtered abnormal data points;

[0051] A collection of filtered abnormal data points Perform feature extraction and pattern matching, the formula is:

[0052]

[0053] in, is a set of abnormal data points, is a parameter to adjust the complexity of feature extraction, is the feature extraction function, is the key feature pattern set.

[0054] The present invention is improved in that the steps of identifying the atypical behavior and leakage signal are specifically as follows:

[0055] The data set after the feature extraction and abnormal pattern matching is classified to determine whether there is a leakage signal, using the formula:

[0056]

[0057] Get leak signals or atypical behavior ,in, From the feature set Extracted from Features, It is The weight coefficient of the feature, It is used to adjust the features The power index of influence, is the bias term, used to adjust the threshold, is the adjustment parameter, is the number of features used to calculate the geometric mean.

[0058] The present invention is improved in that the steps of obtaining the feedback analysis log are specifically as follows:

[0059] Based on the leakage signal feature set, the detection results are collected using the formula:

[0060]

[0061] in, Indicates Test results, is and the result The associated weight coefficient, is the weight adjustment index, is the total number of results, is the sum of the collected test results;

[0062] The collected test results are analyzed to evaluate the deviation from the expected performance using the formula:

[0063]

[0064] in, is the average value of the test results, It is The standard deviation of the results, is the nonlinear adjustment coefficient, Indicates Test results, is the variance of the outcome deviation;

[0065] Based on the analyzed test results, key information is recorded in the feedback analysis log , using the formula:

[0066]

[0067] in, It is The variance of the result deviation, Is a function that adds analysis results and descriptions to the log.

[0068] Compared with the prior art, the advantages and positive effects of the present invention are:

[0069] In the present invention, the long short-term memory network can conduct in-depth learning of the complex relationship between historical and current environmental data, realize real-time dynamic adjustment of the detection strategy, and effectively respond to environmental changes. This adaptive adjustment reduces the false alarm rate and improves the reliability of closed detection. The isolation forest algorithm can process and locate anomalies more quickly in feature extraction and abnormal pattern matching, especially in identifying tiny leakage signals, and reduce potential safety risks. BRIEF DESCRIPTION OF THE DRAWINGS

[0070] Figure 1 A module diagram of a deep-cold hydrogen storage tank airtightness detection system is proposed for the present invention;

[0071] Figure 2 The present invention proposes a flow chart for window segmentation of time series data in a deep-cold hydrogen storage tank airtightness detection system;

[0072] Figure 3 The present invention proposes a flow chart for obtaining environmental feature vectors in a deep-cold hydrogen storage tank airtightness detection system;

[0073] Figure 4 The present invention proposes a flow chart for screening historical detection conditions in a deep-cold hydrogen storage tank airtightness detection system;

[0074] Figure 5 The present invention proposes a flow chart for obtaining a detection parameter adjustment index in a deep-cold hydrogen storage tank airtightness detection system;

[0075] Figure 6 A flow chart of feature extraction and abnormal pattern matching in a deep-cold hydrogen storage tank airtightness detection system is proposed for the present invention;

[0076] Figure 7 The present invention proposes a flow chart for identifying atypical behaviors and leakage signals in a cryogenic hydrogen storage tank airtightness detection system;

[0077] Figure 8 The present invention proposes a flow chart for obtaining a feedback analysis log in a deep-cold hydrogen storage tank airtightness detection system. DETAILED DESCRIPTION

[0078] In order to make the purpose, technical solution and advantages of the present invention more clearly understood, the present invention is further described in detail below in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present invention and are not intended to limit the present invention.

[0079] In the description of the present invention, it should be understood that the terms "length", "width", "up", "down", "front", "back", "left", "right", "vertical", "horizontal", "top", "bottom", "inside", "outside", etc., indicating positions or positional relationships, are based on the positions or positional relationships shown in the accompanying drawings, and are only for the convenience of describing the present invention and simplifying the description, rather than indicating or implying that the device or element referred to must have a specific orientation, be constructed and operated in a specific orientation, and therefore cannot be understood as limiting the present invention. In addition, in the description of the present invention, the meaning of "multiple" is two or more, unless otherwise clearly and specifically defined. Example

[0080] See also Figure 1 The present invention provides a technical solution: a deep-cold hydrogen storage tank airtightness detection system comprises:

[0081] The environmental data analysis module collects temperature and pressure data based on environmental monitoring sensors, performs normalization processing, performs window segmentation on time series data, and performs pattern recognition on the fluctuation of the sealed environment to obtain environmental feature vectors;

[0082] The detection strategy optimization module is based on the environmental feature vector and combines historical detection data. Through the long short-term memory network, it screens the historical detection conditions that are consistent with the current environment, dynamically adjusts the hydrogen storage tank detection conditions, and generates detection parameter adjustment indicators;

[0083] The abnormal behavior recognition module uses the isolation forest algorithm to extract features and match abnormal patterns based on the detection parameter adjustment indicators and real-time monitoring data to identify atypical behaviors and leakage signals in the sealing test and obtain the leakage signal feature set.

[0084] The detection result feedback module evaluates the accuracy and reliability of the detection results based on the leakage signal feature set, displays the detection results using visualization technology, adjusts the operation process according to the detection results, and establishes a feedback analysis log.

[0085] The environmental feature vector includes the environmental temperature index, the pressure stability index, and the data periodicity index. The detection parameter adjustment index includes the environmental response sensitivity, the historical data adaptability, and the detection cycle adaptability. The leakage signal feature set includes the leakage intensity index, the leakage frequency response information, and the leakage area location.

[0086] See also Figure 2 , the specific steps for window segmentation of time series data are:

[0087] Based on environmental monitoring sensors, temperature and pressure data are collected, normalized, and time series data is segmented using the formula:

[0088]

[0089] and

[0090]

[0091] Get the temperature window segmentation of time series data and pressure window segmentation ,in, is the normalized temperature data, is the normalized pressure data, Represents the window size, is the adjustment factor, To avoid small amounts of division by zero, Is the index value, indicating that the First, take out the length The subsequence is processed;

[0092] The temperature window segmentation and pressure window segmentation of the time series data are analyzed using the formula:

[0093]

[0094] Generate a weighted mean for each temperature window segment ,in, and is the weight coefficient of temperature and pressure data, is the window size, and For the window temperature and pressure data points.

[0095] Assume the following data:

[0096] and The original data range is and ;

[0097] The normalization formula is: ;

[0098] Assume the value is 0.5;

[0099] Assume the value is O.01.

[0100] Calculation process:

[0101] Assume that at a certain point in time :

[0102]

[0103]

[0104]

[0105]

[0106] Assume the following data:

[0107] =0.3, reflecting that temperature data may be more important than pressure data in some cases;

[0108] The window size, assuming it is 10, means that 10 consecutive data points are considered in the calculation.

[0109] Calculation process:

[0110] Assume there are 10 and Data points, all with values ​​of 0.98:

[0111]

[0112]

[0113]

[0114] See also Figure 3 , the specific steps for obtaining the environmental feature vector are:

[0115] Apply pattern recognition algorithms to identify key patterns in time series segmented data, including periodicity and trends, using the formula:

[0116]

[0117] Pattern parameters extracted from segmented data, where is the index or time stamp of the time series, is the corresponding normalized temperature or pressure value, 、 and are model coefficients used to adjust for the sensitivity of the trend and to accommodate changes in the time scale of differencing, is a smoothing parameter used to avoid zero denominator and adjust the model sensitivity;

[0118] Integrate the key patterns and parameters identified using the formula:

[0119]

[0120] Get the environmental feature vector ,in, and is the trend coefficient in linear regression, and is a parameter used to adjust the sensitivity of the model, is the key frequency in periodic analysis.

[0121] Assume the following data:

[0122] Assume the index of the time series is for ;

[0123] =0.5, =0.3, =1.5, =2.

[0124] Calculation process:

[0125] calculate

[0126] calculate

[0127] Substituting the values ​​into the formula:

[0128]

[0129] Assume the following data:

[0130] Assume it is 0.05.

[0131] Calculation process:

[0132] calculate :

[0133]

[0134] calculate :

[0135]

[0136] calculate :

[0137]

[0138] Eigenvectors:

[0139]

[0140] See also Figure 4 , the specific steps for screening historical detection conditions are as follows:

[0141] Based on the environmental feature vector and combined with historical detection data, the formula is adopted:

[0142]

[0143] Calculate the similarity of each historical data point ,in, is the first feature vector of the current environment dimension, The historical detection data Item dimension, It is The weight factor of the dimension, is the vector dimension;

[0144] Set a similarity threshold and filter historical data with similarity scores higher than the threshold using the formula:

[0145]

[0146] Select historical data points that meet the criteria ,in, is the selected benchmark similarity threshold, for The standard deviation of is used to adjust the threshold to fit the distribution of the data, The historical detection data stronghold;

[0147] Based on filtered historical data points , using the formula:

[0148]

[0149] Get historical detection status that matches the current environment ,in, is based on From the screening process The historical detection data items obtained in .

[0150] Assume the following data:

[0151] Assumptions (normalized temperature), (normalized pressure);

[0152] No. of the historical data points dimension, assuming that for a certain historical point, ;

[0153] Assumptions The weights are equal, ;

[0154] The vector dimension is set to 2.

[0155] Calculation process:

[0156] Calculate the numerator:

[0157]

[0158] Calculate the denominator:

[0159]

[0160] Calculating Similarity :

[0161]

[0162] Assume the following data:

[0163] The baseline similarity threshold is assumed to be 0.5;

[0164] Assuming that the data distribution is relatively uniform, .

[0165] Calculation process:

[0166] Calculate the threshold:

[0167]

[0168] examine Is it higher than the threshold, 0.34>0.05.

[0169] Assume 1;

[0170] Historical data points that meet the conditions, assuming that and , the values ​​are assumed to be 0.65 and 0.45.

[0171] Calculate the weighted average:

[0172]

[0173] See also Figure 5 , the specific steps for obtaining the detection parameter adjustment index are:

[0174] Based on the historical detection conditions that match the current environment, the detection conditions of the hydrogen storage tank are dynamically adjusted using the formula:

[0175]

[0176] Get the difference after adjusting the detection conditions ,in, For the current environment The detection parameters, is the average value of the parameter in historical data, To adjust the weight coefficient of the parameter, To avoid division by zero errors for small positive numbers, is the total number of detection parameters;

[0177] According to the adjusted hydrogen storage tank detection conditions, calculate the detection parameter adjustment index using the formula:

[0178]

[0179] in, is the difference after adjustment for each test condition, is the coefficient used to balance the detection condition, is the total number of detection parameters, is the detection parameter adjustment indicator, To avoid Is a non-negative number.

[0180] Assume the following data:

[0181] There are two detection parameters , are temperature and pressure respectively, and the given parameters are as follows:

[0182] ;

[0183] ;

[0184] ;

[0185] .

[0186] Calculation process:

[0187] For temperature:

[0188]

[0189] For pressure:

[0190]

[0191] Total adjustment index:

[0192]

[0193] Assumptions and .

[0194] Calculation process:

[0195]

[0196] See also Figure 6 ,The steps of feature extraction and abnormal pattern matching are as follows:

[0197] Based on the detection parameter adjustment index, the isolation forest algorithm is used to process the real-time monitoring data and perform preliminary outlier identification using the formula:

[0198]

[0199] Get anomaly score ,in, For a single data point monitored in real time, It is a data point With the average The Euclidean distance between is the sensitivity adjustment parameter, is the number of trees affecting the parameter, is the tree depth, It is a small amount that avoids the denominator being zero;

[0200] Based on the anomaly score, data points with scores higher than the threshold are selected as key anomalies. The formula is:

[0201]

[0202] in, is the baseline abnormality threshold, is the threshold adjustment factor, is the standard deviation of the anomaly scores, is the set of filtered abnormal data points;

[0203] Perform feature extraction and pattern matching on the screened abnormal data point set. The formula is:

[0204]

[0205] in, is a set of abnormal data points, is a parameter to adjust the complexity of feature extraction, is the feature extraction function, is the key feature pattern set.

[0206] Assume the following data:

[0207] Assumptions =15.0;

[0208] Assumptions =10.0;

[0209] Assume that ;

[0210] Assumptions is 0.1;

[0211] Assumptions for 100 trees;

[0212] Assumptions is 10;

[0213] Assumptions is 0.01.

[0214] Calculation process:

[0215] Calculate the square of the distance:

[0216]

[0217] Apply the formula to calculate :

[0218]

[0219] The result is very close to 0, showing is a significant outlier.

[0220] Assume the following data:

[0221] Set to 0.05;

[0222] Set to 1.5;

[0223] Set to 0.02.

[0224] Calculation process:

[0225] Calculate the adjusted threshold:

[0226]

[0227] Comparing anomaly scores With the adjusted threshold:

[0228]

[0229] Assumptions The following data points are included:

[0230] Data Point 1: ;

[0231] Data Point 2: ;

[0232] Data Point 3: ;

[0233] Assumptions The value of is 0.5.

[0234] Data processing:

[0235] First, calculate The average of all points in :

[0236]

[0237] Next, calculate the deviation of each point from the mean and apply Make adjustments:

[0238]

[0239]

[0240]

[0241] Feature combination:

[0242] Combine the deviation values ​​to construct the feature vector:

[0243]

[0244] See also Figure 7 , the specific steps for identifying atypical behaviors and leakage signals are:

[0245] Classify the data set after feature extraction and abnormal pattern matching to determine whether there is a leakage signal, using the formula:

[0246]

[0247] Get leak signals or atypical behavior ,in, From the feature set Extracted from Features, It is The weight coefficient of the feature, It is used to adjust the features The power index of influence, is the bias term, used to adjust the threshold, is to adjust the parameters to improve the model's sensitivity to feature interactions by integrating the geometric mean of all features. is the number of features used to calculate the geometric mean.

[0248] Assume the following data:

[0249] Assume there are three features =2.0, =3.0 and =4.0;

[0250] Assume that the corresponding weight =0.5, =0.3, =0.2;

[0251] Power Index =1, , =1;

[0252] Bias =-1.5, nonlinear parameter =0.1.

[0253] Calculation process:

[0254] Calculate the adjusted weight impact of each feature:

[0255] for :

[0256]

[0257] for :

[0258]

[0259] for :

[0260]

[0261] Compute the geometric mean of the features and scale:

[0262] Geometric mean:

[0263]

[0264] Adjustment of the geometric mean:

[0265]

[0266] Combine all calculated values:

[0267] sum:

[0268]

[0269]

[0270] Apply the sigmoid function to calculate the final probability:

[0271]

[0272] The results show that a given set of features has a very high probability of indicating leakage or atypical behavior.

[0273] See also Figure 8 , the specific steps for obtaining feedback analysis logs are:

[0274] Based on the leakage signal feature set, the detection results are collected using the formula:

[0275]

[0276] in, Indicates Test results, is and the result The associated weight coefficient, is the weight adjustment index, is the total number of results, is the sum of the collected test results;

[0277] The collected test results are analyzed to evaluate the deviation from the expected performance using the formula:

[0278]

[0279] in, is the average value of the test results, It is The standard deviation of the results, is the nonlinear adjustment coefficient, Indicates Test results, is the variance of the outcome deviation;

[0280] Based on the analyzed test results, key information is recorded in the feedback analysis log , using the formula:

[0281]

[0282] in, It is The variance of the result deviation, Is a function that adds analysis results and descriptions to the log.

[0283] Assume the following data:

[0284] Assume there are three test results ;

[0285] The corresponding weights are ;

[0286] Adjustment Index .

[0287] Calculation process

[0288] Apply the formula to each result:

[0289]

[0290]

[0291]

[0292] so .

[0293] Assume the following data:

[0294] Assumptions .

[0295] Calculate the average:

[0296]

[0297] Apply the formula to calculate the variance:

[0298]

[0299]

[0300]

[0301]

[0302]

[0303] so .

[0304] Assumptions The value is 56.4 and the value will be recorded in the log.

[0305] Application formula:

[0306]

[0307] Indicates the operation of logging, recording the variance value and associated description.

[0308] The above are only preferred embodiments of the present invention and are not intended to limit the present invention in other forms. Any technician familiar with the profession may use the technical contents disclosed above to change or modify them into equivalent embodiments with equivalent changes and apply them to other fields. However, any simple modification, equivalent change and modification made to the above embodiments based on the technical essence of the present invention without departing from the technical solution of the present invention still falls within the protection scope of the technical solution of the present invention.

Claims

1. A cryogenic hydrogen storage tank tightness detection system, characterized in that: The system comprises: The environmental data analysis module collects temperature and pressure data based on environmental monitoring sensors, performs normalization processing, performs window segmentation on time series data, and performs pattern recognition on the fluctuation of the sealed environment to obtain environmental feature vectors; The detection strategy optimization module is based on the environmental feature vector and combines historical detection data to screen historical detection conditions that are consistent with the current environment through a long short-term memory network, and dynamically adjusts the hydrogen storage tank detection conditions to generate detection parameter adjustment indicators; The abnormal behavior recognition module uses the isolation forest algorithm to extract features and match abnormal patterns on the data based on the detection parameter adjustment index and real-time monitoring data, identifies atypical behaviors and leakage signals in the sealing test, and obtains a leakage signal feature set; The detection result feedback module evaluates the accuracy and reliability of the detection result based on the leakage signal feature set, displays the detection result using visualization technology, adjusts the operation process according to the detection result, and establishes a feedback analysis log; The steps of window segmenting the time series data are specifically as follows: Based on environmental monitoring sensors, temperature and pressure data are collected, normalized, and time series data is segmented using the formula: ; and ; Get the temperature window segmentation of time series data and pressure window segmentation ,in, is the normalized temperature data, is the normalized pressure data, Represents the window size, is the adjustment factor, To avoid small amounts of division by zero, is the index value; The temperature window segmentation and pressure window segmentation of the time series data are analyzed using the formula: ; Generate a weighted mean for each temperature window segment ,in, and is the weight coefficient of temperature and pressure data, is the window size, and For the window temperature and pressure data points; The steps of obtaining the environmental feature vector are specifically as follows: Apply pattern recognition algorithms to identify key patterns in the time series segmented data using the formula: ; Pattern parameters extracted from segmented data, where is the index or time stamp of the time series, is the corresponding normalized temperature or pressure value, 、 and are the model coefficients, is the smoothing parameter; Integrate the key patterns and parameters identified using the formula: ; Get the environmental feature vector ,in, and is the trend coefficient in linear regression, and is a parameter used to adjust the sensitivity of the model, is the key frequency in periodic analysis; The steps for screening the historical detection conditions are specifically as follows: Based on the environmental feature vector and combined with historical detection data, the formula is adopted: ; Calculate the similarity of each historical data point ,in, is the first feature vector of the current environment dimension, The historical detection data Item dimension, It is The weight factor of the dimension, is the vector dimension; Set a similarity threshold and filter historical data with similarity scores higher than the threshold using the formula: ; Select historical data points that meet the criteria ,in, is the selected benchmark similarity threshold, for The standard deviation of The historical detection data stronghold; Based on filtered historical data points , using the formula: ; Get historical detection status that matches the current environment ,in, is the weight assigned according to the similarity score, From the screening process The historical detection data items obtained in The steps for obtaining the detection parameter adjustment index are specifically as follows: Based on the historical detection conditions that are consistent with the current environment, the detection conditions of the hydrogen storage tank are dynamically adjusted using the formula: ; Get the difference after adjusting the detection conditions ,in, For the current environment The detection parameters, is the average value of the parameter in historical data, To adjust the weight coefficient of the parameter, To avoid division by zero errors for small positive numbers, is the total number of detection parameters; According to the adjusted hydrogen storage tank detection conditions, calculate the detection parameter adjustment index using the formula: ; in, is the difference after adjustment for each test condition, is the coefficient used to balance the detection condition, is the total number of detection parameters, is the detection parameter adjustment indicator; The steps of feature extraction and abnormal pattern matching are specifically as follows: Based on the detection parameter adjustment index, the isolation forest algorithm is used to process the real-time monitoring data and perform preliminary outlier identification using the formula: ; Get anomaly score ,in, For a single data point monitored in real time, It is a data point With the average The Euclidean distance between is the sensitivity adjustment parameter, is the number of trees affecting the parameter, is the tree depth, It is a small amount that avoids the denominator being zero; Based on the anomaly score, data points with scores higher than the threshold are selected as key anomalies. The formula is: ; in, is the baseline abnormality threshold, is the threshold adjustment factor, is the standard deviation of the anomaly scores, is the set of filtered abnormal data points; A collection of filtered abnormal data points Perform feature extraction and pattern matching, the formula is: ; in, is a set of abnormal data points, is a parameter to adjust the complexity of feature extraction, is the feature extraction function, is the key feature pattern set.

2. The cryogenic hydrogen storage tank tightness detection system according to claim 1 is characterized in that: The steps for identifying the atypical behavior and leakage signals are specifically as follows: The data set after the feature extraction and abnormal pattern matching is classified to determine whether there is a leakage signal, using the formula: ; Get leak signals or atypical behavior ,in, From the feature set Extracted from Features, It is The weight coefficient of the feature, It is used to adjust the features The power index of the impact, is a bias term used to adjust the threshold, is the adjustment parameter, is the number of features used to calculate the geometric mean.

3. The cryogenic hydrogen storage tank tightness detection system according to claim 1 is characterized in that: The steps for obtaining the feedback analysis log are specifically as follows: Based on the leakage signal feature set, the detection results are collected using the formula: ; in, Indicates Test results, is and the result The associated weight coefficient, is the weight adjustment index, is the total number of results, is the sum of the collected test results; The collected test results are analyzed to evaluate the deviation from the expected performance using the formula: ; in, is the average value of the test results, It is The standard deviation of the results, is the nonlinear adjustment coefficient, Indicates Test results, is the variance of the outcome deviation; Based on the analyzed test results, key information is recorded in the feedback analysis log , using the formula: ; in, It is The variance of the result deviation, Is a function that adds analysis results and descriptions to the log.