Distributed Internet of Things operation and maintenance terminal fault analysis method and system based on big data

Abstract

The invention relates to the technical field of data analysis, in particular to a distributed Internet of Things operation and maintenance terminal fault analysis method and system based on big data, and the method comprises the steps: obtaining an operation and maintenance terminal monitoring data encryption time sequence and an equipment working condition parameter encryption time sequence, and extracting a feature mark point set through encryption domain differential operation; and constructing a feature vector and generating a fitting curve. Time migration matching is carried out by adopting an information entropy weighted genetic algorithm, high-precision time alignment is realized through fine tuning, and an abnormal feature mode is extracted and fault judgment is carried out by applying a sliding window correlation coefficient method. According to the method, after the characteristics are extracted through the encryption sequence of the operation data of the operation and maintenance terminal of the Internet of Things and the working condition parameter data of the tested equipment, the correlation abnormal characteristic mode is analyzed through sliding matching, so that the fault mode is matched.

Description

Technical Field

[0001] The present invention relates to the field of data analysis technology, and in particular to a method and system for analyzing faults of distributed Internet of Things operation and maintenance terminals based on big data. Background Art

[0002] Distributed IoT operation and maintenance terminals address the large distribution space across various application scenarios. Essentially, these terminals transmit information and perform corresponding functions through the IoT. The number of IoT operation and maintenance terminals can range from hundreds to thousands, and the orderly transmission and analysis of their data requires normal operation. If a malfunction is detected in an operation and maintenance terminal, data may be lost, distorted, or interfered with, requiring prompt diagnosis and analysis. The challenge lies in certain key industrial scenarios, where the service provider of the IoT operation and maintenance terminal and the service provider of the equipment they measure may be different. To protect technical confidentiality, the operating parameters of industrial equipment are kept confidential. However, the parameters of a single operation and maintenance terminal alone cannot determine whether a fault exists. The determination of a fault in an IoT operation and maintenance terminal depends on the operating parameters and operating conditions of the equipment it measures. Equipment operating parameters are often stored in encrypted, time-series format on a network physically isolated from the external network. The relevant detection parameters of the IoT operation and maintenance terminal are also stored in encrypted, time-series format. Summary of the Invention

[0003] (1) Technical problems to be solved

[0004] The purpose of the present invention is to provide a distributed Internet of Things operation and maintenance terminal fault analysis method and system based on big data, so as to solve the difficulty of data analysis of the encrypted time series of the operating parameters of the equipment in the diagnosis of Internet of Things operation and maintenance terminal faults.

[0005] (2) Technical solution

[0006] To achieve the above objectives, on the one hand, the present invention provides a distributed IoT operation and maintenance terminal fault analysis method based on big data, the method comprising:

[0007] Obtain the encrypted time series of monitoring data of the distributed IoT operation and maintenance terminal as the first series, and the encrypted time series of the operating parameters of the corresponding monitored equipment as the second series. Extract encryption features from the first series to obtain a first series feature marker point set. Extract encryption features from the second series to obtain a second series feature marker point set.

[0008] The time coordinate value, amplitude coordinate value, and marker type code of each marker point in the first sequence feature marker point set are constructed as a first sequence marker point feature vector, and the time coordinate value, amplitude coordinate value, and marker point type code of each marker point in the second sequence feature marker point set are constructed as a second sequence marker point feature vector; the first sequence marker point feature vectors are mapped into a coordinate system to form a first sequence marker point coordinate set, and the second sequence marker point feature vectors are mapped into a coordinate system to form a second sequence marker point coordinate set, and the first sequence marker point coordinate set is fitted by a cubic spline interpolation fitting algorithm to obtain a first sequence fitting curve, and the second sequence marker point coordinate set is fitted to obtain a second sequence fitting curve;

[0009] The first sequence fitting curve and the second sequence fitting curve are randomly time-shifted and matched, and the optimal time offset value is found through the genetic algorithm. The second sequence is time-corrected according to the optimal time offset value to achieve time alignment between the first and second sequences. The correlation anomaly feature pattern is extracted from the first and second sequences after time alignment through the sliding window correlation coefficient calculation method. The correlation anomaly feature pattern is compared with the preset fault discrimination threshold to obtain the fault analysis result of the distributed Internet of Things operation and maintenance terminal.

[0010] Furthermore, the method of performing random time offset matching on the first sequence fitting curve and the second sequence fitting curve and finding the optimal time offset value by a genetic algorithm includes:

[0011] A plurality of time offset candidate values ​​are generated within a preset time offset interval. For each time offset candidate value, the second sequence fitting curve is time-shifted. Sampling coordinate points are selected at equal intervals on the first sequence fitting curve. The sum of the squares of the Euclidean distances between each sampling coordinate point and the corresponding coordinate point of the second sequence fitting curve after the shift is calculated and recorded as the curve matching error value. The time offset candidate values ​​and the corresponding curve matching error values ​​are constructed into a time offset candidate scheme. The time offset candidate scheme is used as the initial population individual of the genetic algorithm. The population individuals with smaller curve matching error values ​​are screened through the genetic algorithm selection operation. New time offset candidate values ​​are generated through crossover and mutation operations. The optimal time offset value is obtained by iterative optimization with the minimization of the curve matching error value as the fitness goal.

[0012] Furthermore, the method of calculating the sum of squares of the Euclidean distances between each sampling coordinate point and the corresponding coordinate point of the translated second sequence fitting curve as a curve matching error value, and constructing the time offset candidate value and the corresponding curve matching error value as a time offset candidate solution includes:

[0013] The total number of sampling points is set to be equal intervals within the time range of the first series fitting curve. , No. The time coordinate of each sampling point is marked as , the corresponding amplitude coordinates are marked as , is the sampling point number and its value ranges from 1 to Integer variable for time offset candidate value , calculate the second sequence fitting curve after translation in the time coordinate The corresponding amplitude coordinates at ; Construct a matching error calculation formula based on information entropy weighting, and the curve matching error value for:

[0014] ;

[0015] in is the derivative weight coefficient, and They are the first sequence fitting curve and the second sequence fitting curve after translation. The first-order derivative at ; Calculate the comprehensive fitness value of the time offset candidate value for:

[0016] ;

[0017] in For the preset shape similarity control parameter, the time offset candidate value And the corresponding comprehensive fitness value Constructed as a candidate for time offset.

[0018] Furthermore, the method further comprises:

[0019] Set the neighborhood range of the optimal time offset value as the fine-tuning interval, and set the fine-tuning step size within the fine-tuning interval to ,generate Fine-tuning candidate offset values , is the fine-tuning sequence number and its value ranges from 1 to Integer variable; for each fine-tuning candidate offset value Recalculate the fine matching degree of the first series fitting curve and the second series fitting curve for:

[0020] ;

[0021] in is the preset matching accuracy control parameter, To fine-tune the candidate offset value The second sequence fitting curve after translation is in the time coordinate The amplitude coordinates at ; find the fine matching degree by traversal search The fine-tuned candidate offset value corresponding to the maximum value of .

[0022] Furthermore, the method of extracting encryption features from the first sequence to obtain a first sequence feature marker point set, and extracting encryption features from the second sequence to obtain a second sequence feature marker point set includes:

[0023] Performing a first-order difference operation on the first sequence to obtain first-order difference data of the first sequence, performing a second-order difference operation on the first-order difference data of the first sequence to obtain second-order difference data of the first sequence, identifying the sign change position in the first-order difference data of the first sequence as the amplitude marking point of the first sequence through an encryption domain sign judgment algorithm, identifying the sign change position in the second-order difference data of the first sequence as the inflection point marking point of the first sequence, and merging the amplitude marking point of the first sequence and the inflection point marking point of the first sequence to form a first sequence characteristic marking point set;

[0024] A first-order difference operation is performed on the second sequence to obtain first-order difference data of the second sequence, and a second-order difference operation is performed on the first-order difference data of the second sequence to obtain second-order difference data of the second sequence. The sign change position in the first-order difference data of the second sequence is identified as the amplitude marking point of the second sequence through the encryption domain sign judgment algorithm, and the sign change position in the second-order difference data of the second sequence is identified as the inflection point marking point of the second sequence. The second sequence amplitude marking point and the second sequence inflection point marking point are merged to form a second sequence characteristic marking point set.

[0025] Furthermore, the method of performing a first-order difference operation on the first sequence to obtain first-order difference data of the first sequence, performing a second-order difference operation on the first-order difference data of the first sequence to obtain second-order difference data of the first sequence, identifying a sign change position in the first-order difference data of the first sequence as a first-sequence amplitude marking point through an encryption domain sign judgment algorithm, identifying a sign change position in the second-order difference data of the first sequence as a first-sequence inflection point marking point, and merging the first-sequence amplitude marking point and the first-sequence inflection point marking point to form a first-sequence characteristic marking point set includes:

[0026] The total number of encrypted time series data points of the first sequence is recorded as , in the first sequence The encrypted data at a time point is recorded as , The value ranges from 1 to Integer variable; calculate the first order difference data of the first sequence, The first-order difference encrypted value of the position for:

[0027] ;

[0028] in is the encryption domain subtraction operator, The value ranges from 1 to Integer variable; calculate the first series of second-order difference data, the The second-order differential encrypted value of the position for:

[0029] ;

[0030] in The value ranges from 1 to Integer variable; detect the first-order difference encryption value through the encryption domain symbol judgment algorithm and The sign changes when When the conditions are met, the time position Record as the candidate amplitude position, and filter the encrypted domain amplitude change exceeding the preset fluctuation threshold The amplitude candidate positions form the first sequence of amplitude mark points; the second-order differential encryption value is detected by the encryption domain symbol judgment algorithm and The sign changes when When the conditions are met, the time position Record as the candidate inflection point position, and filter the encrypted domain where the curvature change exceeds the preset curvature threshold The candidate positions of the first sequence of inflection point marking points are formed; the first sequence of amplitude marking points and the first sequence of inflection point marking points are arranged and merged in chronological order to form a first sequence of feature marking point set.

[0031] Furthermore, the method of extracting a correlation anomaly feature pattern from the time-aligned first and second sequences by calculating a sliding window correlation coefficient, and comparing the correlation anomaly feature pattern with a preset fault discrimination threshold to obtain a fault analysis result of a distributed IoT operation and maintenance terminal includes:

[0032] The first sequence data after time alignment is obtained and recorded as , the second sequence data after time alignment is recorded as ,in The value ranges from 1 to the length of the sequence after time alignment integer variable, is the total number of data points in the time-aligned sequence; set the sliding window size to , the sliding step length is ; in the Extract the first sequence data in the window position and the second sequence data within the window , where g is the window position variable and its value ranges from 0 to The integer range of is the position variable in the window and has a value of 0 to The integer range of ;

[0033] Calculate the The local correlation coefficient of the window position for:

[0034] ;

[0035] in For the The mean of the first series of data in the window, For the The mean of the second series data in the window; the local correlation coefficient of all window positions Composed of correlation coefficient sequence, the standard deviation of the correlation coefficient sequence is calculated as , the absolute value The window position of is marked as abnormal window position, where is the mean of the correlation coefficient sequence, is the preset anomaly detection coefficient; the time interval corresponding to the abnormal window position and the local correlation coefficient value are constructed as a correlation anomaly feature pattern. When the number of abnormal window positions in the correlation anomaly feature pattern exceeds the preset fault discrimination threshold When the distributed IoT operation and maintenance terminal is faulty, it is determined that there is a fault.

[0036] Based on the same inventive concept, on the other hand, the present invention also provides a distributed Internet of Things operation and maintenance terminal fault analysis system based on big data, which is used to execute the distributed Internet of Things operation and maintenance terminal fault analysis method based on big data.

[0037] (3) Beneficial effects

[0038] Compared with the existing technology, the beneficial effect of the present invention is to extract features from the encrypted sequence of the operation data of the Internet of Things operation and maintenance terminal and the operating parameter data of the measured equipment, and then perform sliding matching to analyze the correlation abnormal feature pattern and thus match the fault mode, so as to solve the specific confidentiality requirements and be able to mine and match the Internet of Things operation and maintenance terminal as accurately as possible. BRIEF DESCRIPTION OF THE DRAWINGS

[0039] Figure 1 This is a flowchart of a method for analyzing faults in a distributed IoT operation and maintenance terminal based on big data according to Example 1 of the present invention. DETAILED DESCRIPTION

[0040] The following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative efforts are within the scope of protection of the present invention.

[0041] Before giving examples, it is necessary to explain the application scenarios of the present invention. For example, there are more than 800 IoT operation and maintenance terminals for a transnational pipeline. The distance between adjacent IoT terminals is 10 kilometers. All IoT operation and maintenance terminals need to analyze and process the collected data in real time. Naturally, they need to be tested regularly to see if there are any faults. However, the existing faults cannot be determined by the pipeline operating parameters due to changes in the pipeline's operating parameters. On the other hand, the environmental parameters of the pipeline's IoT operation and maintenance terminals vary greatly in their geographical locations. It is difficult to obtain the difference value through statistical analysis of a single IoT operation and maintenance terminal to troubleshoot the fault. The service provider of the IoT terminal needs to analyze the encrypted time series data with the pipeline operation and maintenance service provider. However, the analysis of the two encrypted sequences may not be aligned in time, and there is a problem of how to extract the periodic data in segments, because it is impossible to compare very long data segments in encrypted form.

[0042] Example 1: Figure 1 As shown, this embodiment provides a distributed IoT operation and maintenance terminal fault analysis method based on big data, the method comprising:

[0043] S1. Obtain an encrypted time series of monitoring data from a distributed IoT operation and maintenance terminal, record it as a first series, and an encrypted time series of operating parameters of the corresponding monitored equipment, record it as a second series. Extract encryption features from the first series to obtain a first series feature marker point set. Extract encryption features from the second series to obtain a second series feature marker point set.

[0044] S2. Constructing the time coordinate value, amplitude coordinate value, and marker type code of each marker point in the first sequence feature marker point set into a first sequence marker point feature vector, and constructing the time coordinate value, amplitude coordinate value, and marker type code of each marker point in the second sequence feature marker point set into a second sequence marker point feature vector; mapping the first sequence marker point feature vectors into a coordinate system to form a first sequence marker point coordinate set, mapping the second sequence marker point feature vectors into a coordinate system to form a second sequence marker point coordinate set, fitting the first sequence marker point coordinate set to obtain a first sequence fitting curve using a cubic spline interpolation fitting algorithm, and fitting the second sequence marker point coordinate set to obtain a second sequence fitting curve;

[0045] S3. Perform random time offset matching on the first sequence fitting curve and the second sequence fitting curve, and find the optimal time offset value through genetic algorithm; perform time correction on the second sequence according to the optimal time offset value to achieve time alignment between the first sequence and the second sequence, extract the correlation anomaly feature pattern from the first sequence and the second sequence after time alignment through the sliding window correlation coefficient calculation method, and compare the correlation anomaly feature pattern with the preset fault judgment threshold to obtain the fault analysis result of the distributed Internet of Things operation and maintenance terminal.

[0046] For example, the IoT operation and maintenance terminal GT-1246 at the 1246th kilometer of Line C of a natural gas pipeline was selected as the test object. This terminal is responsible for monitoring parameters such as pipeline wall thickness, internal pressure, and fluid temperature. The encrypted time series of monitoring data of this terminal between 2:00 a.m. and 6:00 a.m. on August 20, 2024, was obtained as the first series, with a collection frequency of once every 30 seconds, totaling 480 encrypted data points. Because the pipeline operator and the terminal supplier belong to different units, based on the requirements of the confidentiality agreement, both terminal data and pipeline operating data are stored and transmitted in encrypted form. The encrypted time series of natural gas operating parameters of the corresponding pipeline section was synchronously obtained as the second series, including operating information such as the pressure set value of the upstream pressure regulating station, flow control instructions, and safety valve status, also containing 480 encrypted data points.

[0047] Using encrypted domain differencing, 68 characteristic markers were identified, including 42 amplitude extreme value markers and 26 curve inflection point markers. These markers can reflect key changes in the terminal monitoring data without decrypting the original data. The same feature extraction process was performed on the second sequence, identifying 54 characteristic markers, including 31 amplitude extreme value markers and 23 inflection point markers. The time coordinate value, amplitude coordinate value, and marker type code of each marker in the first sequence were constructed into a feature vector. The 68 markers in the first sequence were mapped into a three-dimensional coordinate system to form the first sequence marker point coordinate set. The 54 markers in the second sequence were similarly mapped to form the second sequence marker point coordinate set. A cubic spline interpolation fitting algorithm was used to fit the first sequence marker point coordinate set, generating a smooth curve containing 68 control points. The second sequence marker point coordinate set was then fitted to obtain a second sequence fitted curve containing 54 control points.

[0048] Due to the geographic span of the pipeline and network transmission delays, there is an uncertain offset between the timestamps of terminal data collection and the timestamps of pipeline condition records. A genetic algorithm is required to find the optimal time offset value for accurate time alignment. The time offset search range is set to -300 seconds to +300 seconds. After 256 genetic algorithm iterations, the optimal time offset value is determined to be -127 seconds, indicating a 127-second lag between the terminal data collection and the condition data records. Based on this offset value, the second sequence is time-corrected to achieve precise time alignment between the first and second sequences. For the time-aligned data, a sliding window size of 60 data points (30-minute time windows) and a sliding step of 6 data points (3 minutes) are set. Local correlation coefficients are calculated for 70 window positions. Analysis reveals that between 3:45 AM and 4:30 AM, 15 consecutive windows exhibit abnormal correlation coefficients, with an average correlation coefficient of only 0.18, significantly lower than the normal operating range of 0.82 to 0.95. The correlation abnormal characteristic pattern composed of these 15 abnormal windows is compared with the preset fault judgment threshold of 10 windows. The number of abnormal windows exceeding the threshold indicates that the GT-1246 terminal does have a fault condition during this time period.

[0049] Furthermore, the method of performing random time offset matching on the first sequence fitting curve and the second sequence fitting curve and finding the optimal time offset value by a genetic algorithm includes:

[0050] A plurality of time offset candidate values ​​are generated within a preset time offset interval. For each time offset candidate value, the second sequence fitting curve is time-shifted. Sampling coordinate points are selected at equal intervals on the first sequence fitting curve. The sum of the squares of the Euclidean distances between each sampling coordinate point and the corresponding coordinate point of the second sequence fitting curve after the shift is calculated and recorded as the curve matching error value. The time offset candidate values ​​and the corresponding curve matching error values ​​are constructed into a time offset candidate scheme. The time offset candidate scheme is used as the initial population individual of the genetic algorithm. The population individuals with smaller curve matching error values ​​are screened through the genetic algorithm selection operation. New time offset candidate values ​​are generated through crossover and mutation operations. The optimal time offset value is obtained by iterative optimization with the minimization of the curve matching error value as the fitness goal.

[0051] For example, after determining that a genetic algorithm is needed to find the optimal time offset value, 120 candidate time offset values ​​are generated within a preset time offset range of -300 seconds to +300 seconds. These 120 candidate values ​​are evenly distributed at 5-second intervals, ranging from -300 seconds, -295 seconds, -290 seconds, to +295 seconds and +300 seconds. For each candidate time offset value, the second sequence fitting curve is subjected to a corresponding time shift operation. For example, when the candidate time offset value is -127 seconds, the entire second sequence fitting curve is shifted to the left by 127 seconds.

[0052] 200 sampling coordinate points were equally spaced on the first sequence fitting curve, with a time interval of 72 seconds, covering the entire 4-hour observation time window. The Euclidean distance between each sampling coordinate point and the coordinate point on the corresponding time position of the second sequence fitting curve after translation was calculated. The Euclidean distances of the 200 sampling points were squared and summed to obtain the curve matching error value corresponding to the time offset candidate value. After calculation, the curve matching error value corresponding to the time offset candidate value of -127 seconds was 1847.32, the curve matching error value corresponding to the time offset candidate value of -132 seconds was 1923.78, and the curve matching error value corresponding to the time offset candidate value of -122 seconds was 1889.45. Each time offset candidate value was paired with the corresponding curve matching error value to construct 120 time offset candidate solutions.

[0053] These 120 time-shift candidates were used as the initial population input for the genetic algorithm. Through the genetic algorithm selection operation, individuals with smaller curve-matching errors were prioritized for the next generation. The selection operation retained the top 60 individuals with the highest curve-matching errors and eliminated the 60 individuals with larger errors. The selected individuals were paired and combined through a crossover operation to generate new time-shift candidate values. For example, an individual with an error of -127 seconds (1847.32) was crossed with an individual with an error of -122 seconds (1889.45) to generate new candidate values ​​of -124 seconds and -125 seconds. A mutation operation was used to randomly adjust the time-shift values ​​of some individuals within a range of plus or minus 3 seconds to increase population diversity and prevent the algorithm from falling into local optima. With minimizing the curve-matching error as the fitness objective, after 256 iterations of optimization calculations, the optimal time-shift value was determined to be -127 seconds, corresponding to a minimum curve-matching error of 1847.32. This result indicates that the second sequence needs to be adjusted forward by 127 seconds to achieve optimal time alignment with the first sequence.

[0054] Furthermore, the method of calculating the sum of squares of the Euclidean distances between each sampling coordinate point and the corresponding coordinate point of the translated second sequence fitting curve as a curve matching error value, and constructing the time offset candidate value and the corresponding curve matching error value as a time offset candidate solution includes:

[0055] The total number of sampling points is set to be equal intervals within the time range of the first series fitting curve. , No. The time coordinate of each sampling point is marked as , the corresponding amplitude coordinates are marked as , is the sampling point number and its value ranges from 1 to Integer variable for time offset candidate value , calculate the second sequence fitting curve after translation in the time coordinate The corresponding amplitude coordinates at ; Construct a matching error calculation formula based on information entropy weighting, and the curve matching error value for:

[0056] ;

[0057] in is the derivative weight coefficient, and They are the first sequence fitting curve and the second sequence fitting curve after translation. The first-order derivative at ; Calculate the comprehensive fitness value of the time offset candidate value for:

[0058] ;

[0059] in For the preset shape similarity control parameter, the time offset candidate value And the corresponding comprehensive fitness value Constructed as a candidate for time offset.

[0060] For example, based on the 200 sampling coordinate points determined above, the curve matching error value corresponding to the candidate time offset value of -127 seconds is calculated in detail. A total of 200 sampling points are set at equal intervals within the 4-hour time range of the first sequence fitting curve, with a time interval of 72 seconds between each sampling point. The time coordinate of the first sampling coordinate point is 2:00:00 a.m., and the corresponding amplitude coordinate is 145.8. The time coordinate of the second sampling coordinate point is 2:01:12 a.m., and the corresponding amplitude coordinate is 148.2. This continues until the time coordinate of the 200th sampling coordinate point is 5:59:48 a.m., and the corresponding amplitude coordinate is 142.6.

[0061] For the candidate time offset value of -127 seconds, the second sequence fitting curve is shifted forward by 127 seconds, and the corresponding amplitude coordinates of the shifted curve at each sampling time coordinate are calculated. At the first sampling coordinate point at 2:00:00 AM, the corresponding amplitude coordinate of the shifted second sequence fitting curve is 143.4. At the second sampling coordinate point at 2:01:12 AM, the corresponding amplitude coordinate is 146.7. At the 200th sampling coordinate point at 5:59:48 AM, the corresponding amplitude coordinate is 140.9. Due to the complexity of data transmission and processing in pipeline operation and maintenance in this scenario, a simple Euclidean distance calculation cannot fully reflect the degree of matching between the two curves. An information entropy weighting mechanism is needed to improve matching accuracy.

[0062] First, calculate the amplitude difference for each of the 200 sample points. The amplitude difference for the first sample point is 145.8 minus 143.4, which equals 2.4. The amplitude difference for the second sample point is 148.2 minus 146.7, which equals 1.5. The amplitude difference for the 200th sample point is 142.6 minus 140.9, which equals 1.7. Squaring the absolute values ​​of all 200 amplitude differences yields a squared difference of 5.76 for the first sample point, 2.25 for the second sample point, and 2.89 for the 200th sample point. The sum of these 200 squared differences is 748.32. The proportion of the squared difference of each sample point to the total is 5.76 divided by 748.32, which equals 0.0077; 2.25 divided by 748.32, which equals 0.0030; and 2.89 divided by 748.32, which equals 0.0039.

[0063] The information entropy weighting coefficient was calculated. The information entropy weight for the first sampling point was negative 0.0077 multiplied by the logarithm of 0.0077 to the base 2, which equals negative 0.0385. The information entropy weight for the second sampling point was negative 0.0030 multiplied by the logarithm of 0.0030 to the base 2, which equals negative 0.0249. The information entropy weight for the 200th sampling point was negative 0.0039 multiplied by the logarithm of 0.0039 to the base 2, which equals negative 0.0307. The derivative weighting coefficient was set to 0.3. The first-order derivative difference between the first sequence fitting curve and the shifted second sequence fitting curve at each sampling point was calculated. The derivative difference at the first sampling point was 0.8, the derivative difference at the second sampling point was 0.6, and the derivative difference at the 200th sampling point was 0.4. The information entropy weight, the squared difference, and the derivative term were multiplied and summed to obtain a curve matching error of 1847.32.

[0064] Calculate the comprehensive fitness value for the candidate time offset of -127 seconds. The comprehensive fitness value is equal to 1 divided by 1 plus the curve matching error value of 1847.32, multiplied by the shape similarity index. The shape similarity index is calculated by calculating the difference in amplitude change between adjacent sampling points. Setting the shape similarity control parameter to 12.5, the calculated shape similarity index is 0.627. The comprehensive fitness value is equal to 1 divided by 1848.32 multiplied by 0.627, which equals 0.000338. The candidate time offset of -127 seconds and the corresponding comprehensive fitness value of 0.000338 are constructed as the candidate solution for use in subsequent selection, crossover, and mutation operations of the genetic algorithm.

[0065] Furthermore, the method further comprises:

[0066] Set the neighborhood range of the optimal time offset value as the fine-tuning interval, and set the fine-tuning step size within the fine-tuning interval to ,generate Fine-tuning candidate offset values , is the fine-tuning sequence number and its value ranges from 1 to Integer variable; for each fine-tuning candidate offset value Recalculate the fine matching degree of the first series fitting curve and the second series fitting curve for:

[0067] ;

[0068] in is the preset matching accuracy control parameter, To fine-tune the candidate offset value The second sequence fitting curve after translation is in the time coordinate The amplitude coordinates at ; find the fine matching degree by traversal search The fine-tuned candidate offset value corresponding to the maximum value of .

[0069] For example, based on the optimal time offset value of -127 seconds determined by the genetic algorithm, a fine-tuning operation is further performed to improve the accuracy of time alignment. The neighborhood range of the optimal time offset value of -127 seconds is set as the fine-tuning interval, specifically a range of 15 seconds before and after -127 seconds, that is, a 30-second interval from -142 seconds to -112 seconds. Within this fine-tuning interval, the fine-tuning step size is set to 0.5 seconds. Such a fine step size can capture the sub-second time deviation that may exist in ultra-long-distance pipeline data transmission. Based on the 30-second interval and 0.5-second step size, 61 fine-tuning candidate offset values ​​are generated, namely -142.0 seconds, -141.5 seconds, -141.0 seconds, -140.5 seconds, -112.5 seconds, and -112.0 seconds.

[0070] For each candidate fine-tuning offset value, the fine-tuning match between the first and second sequence fitting curves is recalculated. A detailed calculation is performed using a candidate fine-tuning offset value of -125.5 seconds. The second sequence fitting curve is time-shifted by -125.5 seconds, and the amplitude coordinates of the shifted curve are calculated at each of the 200 sampling points. The amplitude coordinate of the second sequence fitting curve after the shift at the first sampling point is 144.1, which differs by 1.7 from the amplitude coordinate of 145.8 at the same point in the first sequence fitting curve. The amplitude coordinate of the second sequence fitting curve after the shift at the second sampling point is 147.3, which differs by 0.9 from the amplitude coordinate of 148.2 at the corresponding point in the first sequence. Amplitude differences are calculated in this manner for all 200 sampling points.

[0071] The matching accuracy control parameter was set to 2.8, which is determined based on the actual accuracy requirements of natural gas pipeline operations. The numerator of the fine matching degree was calculated by dividing the absolute value of the amplitude difference at each sampling point by the matching accuracy control parameter 2.8 and then applying the exponential function. The contribution of the first sampling point was the natural constant (-1.7) divided by the power of 2.8, and the contribution of the second sampling point was the natural constant (-0.9) divided by the power of 2.8. The sum of the exponential function values ​​for the 200 sampling points yielded a numerator of 156.73. To calculate the denominator of the fine matching degree, the first derivative difference between the first sequence fitting curve and the shifted second sequence fitting curve at each sampling point was calculated. The derivative difference at the first sampling point was 0.6, and the derivative difference at the second sampling point was 0.4. The absolute values ​​of the derivative differences between the 199 adjacent sampling points were summed to obtain the second term in the denominator, which was 98.47. The fine matching degree was calculated as the result of dividing the numerator (156.73) by the numerator (156.73) plus the second term in the denominator (98.47).

[0072] By traversing and calculating the fine-matching scores corresponding to 61 fine-tuned candidate offset values, the fine-matching scores for a fine-tuned candidate offset value of -127.5 seconds were 0.6847, -127.0 seconds were 0.6821, -126.5 seconds were 0.6859, and -126.0 seconds were 0.6834. After a complete traversal search, the fine-matching score of 0.6859, corresponding to the fine-tuned candidate offset value of -126.5 seconds, was found to be the maximum among the 61 candidate values. The results show that compared to the roughly optimal time offset value of -127 seconds obtained by the genetic algorithm, the fine-tuned optimal offset value of -126.5 seconds achieves more precise time alignment.

[0073] Furthermore, the method of extracting encryption features from the first sequence to obtain a first sequence feature marker point set, and extracting encryption features from the second sequence to obtain a second sequence feature marker point set includes:

[0074] Performing a first-order difference operation on the first sequence to obtain first-order difference data of the first sequence, performing a second-order difference operation on the first-order difference data of the first sequence to obtain second-order difference data of the first sequence, identifying the sign change position in the first-order difference data of the first sequence as the amplitude marking point of the first sequence through an encryption domain sign judgment algorithm, identifying the sign change position in the second-order difference data of the first sequence as the inflection point marking point of the first sequence, and merging the amplitude marking point of the first sequence and the inflection point marking point of the first sequence to form a first sequence characteristic marking point set;

[0075] A first-order difference operation is performed on the second sequence to obtain first-order difference data of the second sequence, and a second-order difference operation is performed on the first-order difference data of the second sequence to obtain second-order difference data of the second sequence. The sign change position in the first-order difference data of the second sequence is identified as the amplitude marking point of the second sequence through the encryption domain sign judgment algorithm, and the sign change position in the second-order difference data of the second sequence is identified as the inflection point marking point of the second sequence. The second sequence amplitude marking point and the second sequence inflection point marking point are merged to form a second sequence characteristic marking point set.

[0076] For example, due to the data confidentiality requirements of the pipeline operator and the terminal supplier, the above process stores the original monitoring data and operating parameters in encrypted form, making it impossible to directly compare and analyze the plaintext values. A first-order difference operation is performed on the 480 encrypted data points of the first sequence, and the difference between the adjacent encrypted data points from the 1st position to the 479th position is calculated to obtain 479 first-order difference encrypted values. For example, the first-order difference encrypted value of the first position is equal to the encrypted data at the second time point minus the encrypted data at the first time point, and the first-order difference encrypted value of the second position is equal to the encrypted data at the third time point minus the encrypted data at the second time point. These 479 first-order difference values ​​constitute the first sequence of first-order difference data, which reflects the changing trend of the terminal monitoring parameters between adjacent time points. A second-order difference operation is performed on the first-order difference data of the first sequence, and the difference between the adjacent first-order difference values ​​from the 1st position to the 478th position is calculated to obtain 478 second-order difference encrypted values, which constitute the first sequence of second-order difference data.

[0077] The encrypted domain sign determination algorithm is used to identify the sign change locations in the first sequence of first-order differential data. This algorithm can determine the positive or negative nature of encrypted values ​​without decrypting the data, which is achieved due to the characteristics of the encryption algorithm. The sign changes of the first-order differential encrypted values ​​from the 1st to the 478th position and their subsequent values ​​are detected. When the first-order differential encrypted value at the 73rd position is positive and the first-order differential encrypted value at the 74th position is negative, the sign change condition is met, and time position 74 is recorded as the amplitude candidate position. When the first-order differential encrypted value at the 156th position is negative and the first-order differential encrypted value at the 157th position is positive, the sign change condition is also met, and time position 157 is recorded as the amplitude candidate position. By fully traversing the 479 first-order differential values, a total of 89 amplitude candidate positions are identified.

[0078] Candidate amplitude locations whose encrypted domain amplitude changes exceeded a preset fluctuation threshold were screened. The fluctuation threshold was set as a relative quantitative standard for encrypted domain value changes, determined based on the actual fluctuation characteristics of natural gas pipeline operations. Of the 89 candidate amplitude locations, 42 locations had encrypted domain amplitude changes exceeding the preset fluctuation threshold, and these 42 locations constituted the first sequence of amplitude markers. An encrypted domain sign determination algorithm was used to identify sign changes in the first sequence of second-order difference data. The sign changes between the encrypted second-order difference values ​​from the 1st to the 477th position and their subsequent values ​​were detected. When the encrypted second-order difference value at the 98th position was positive and the encrypted second-order difference value at the 99th position was negative, time position 99 was recorded as a candidate inflection point. After a complete traversal of the 478 second-order difference values, a total of 67 candidate inflection point locations were identified. Candidate inflection point locations whose encrypted domain curvature changes exceeded the preset curvature threshold were screened. Of the 67 candidate inflection point locations, 26 exceeded the threshold and constituted the first sequence of inflection point markers. The 42 first-sequence amplitude markers and the 26 first-sequence inflection point markers are arranged and merged in chronological order to form a first-sequence feature marker set containing 68 feature points.

[0079] Using encrypted domain differencing, 479 first-order differential encrypted values ​​and 478 second-order differential encrypted values ​​were obtained. Sign change locations in the first-order differential data were identified, yielding 76 candidate amplitude locations. After screening, 31 second-sequence amplitude markers were determined. Sign change locations in the second-order differential data were identified, yielding 58 candidate inflection point locations. After screening, 23 second-sequence inflection point markers were determined. The 31 amplitude markers and 23 inflection point markers were combined to form a second-sequence feature marker set containing 54 feature points.

[0080] Furthermore, the method of performing a first-order difference operation on the first sequence to obtain first-order difference data of the first sequence, performing a second-order difference operation on the first-order difference data of the first sequence to obtain second-order difference data of the first sequence, identifying a sign change position in the first-order difference data of the first sequence as a first-sequence amplitude marking point through an encryption domain sign judgment algorithm, identifying a sign change position in the second-order difference data of the first sequence as a first-sequence inflection point marking point, and merging the first-sequence amplitude marking point and the first-sequence inflection point marking point to form a first-sequence characteristic marking point set includes:

[0081] The total number of encrypted time series data points of the first sequence is recorded as , in the first sequence The encrypted data at a time point is recorded as , The value ranges from 1 to Integer variable; calculate the first order difference data of the first sequence, The first-order difference encrypted value of the position for:

[0082] ;

[0083] in is the encryption domain subtraction operator, The value ranges from 1 to Integer variable; calculate the first series of second-order difference data, the The second-order differential encrypted value of the position for:

[0084] ;

[0085] in The value ranges from 1 to Integer variable; detect the first-order difference encryption value through the encryption domain symbol judgment algorithm and The sign changes when When the conditions are met, the time position Record as the candidate amplitude position, and filter the encrypted domain amplitude change exceeding the preset fluctuation threshold The amplitude candidate positions form the first sequence of amplitude mark points; the second-order differential encryption value is detected by the encryption domain symbol judgment algorithm and The sign changes when When the conditions are met, the time position Record as the candidate inflection point position, and filter the encrypted domain where the curvature change exceeds the preset curvature threshold The candidate positions of the first sequence of inflection point marking points are formed; the first sequence of amplitude marking points and the first sequence of inflection point marking points are arranged and merged in chronological order to form a first sequence of feature marking point set.

[0086] For example, a total of 480 encrypted time series data points are obtained for the first sequence, each representing an encrypted value of a terminal monitoring parameter within a 30-second collection interval. The encrypted data at the first time point is recorded as the value at 2:00:00 AM, the encrypted data at the second time point is recorded as the value at 2:00:30 AM, and so on, up to the encrypted data at the 480th time point, which is recorded as the value at 5:59:30 AM.

[0087] The encrypted first-order differential value at the first position is calculated using the encryption domain subtraction operator by subtracting the encrypted data at the second time point from the encrypted data at the first time point. The encrypted first-order differential value at the second position is obtained by subtracting the encrypted data at the third time point from the encrypted data at the second time point. This encryption domain subtraction operation can obtain change information between adjacent data points while maintaining the data encryption state, avoiding the security risks of traditional methods that require decryption before calculation. The encrypted first-order differential values ​​from the third to the 479th positions are calculated in the same manner, ultimately obtaining 479 first-order differential data points. These data reflect the changing trend characteristics of the terminal monitoring parameters during the 4-hour observation period.

[0088] The encrypted second-order differential value at the first position is calculated by subtracting the first-order differential value at the second position from the first-order differential value at the second position. The encrypted second-order differential value at the second position is calculated by subtracting the first-order differential value at the third position from the first-order differential value at the third position. The encrypted second-order differential values ​​are calculated from the third to the 478th position, resulting in 478 second-order differential data points. Second-order differential data can identify changes in the rate of change of monitored parameters, which is important for detecting sudden changes and anomalies in pipeline operating conditions.

[0089] The encryption domain sign judgment algorithm detects sign changes in the first-order difference encrypted value. Based on the homomorphic properties of encryption mathematics, the positive and negative properties of the encrypted value can be determined without decryption. The first-order difference encrypted value at the 73rd position is detected to be greater than or equal to 0 and the first-order difference encrypted value at the 74th position is less than 0. If this condition is met, time position 74 is recorded as the amplitude candidate position. The first-order difference encrypted value at the 156th position is detected to be less than 0 and the first-order difference encrypted value at the 157th position is greater than or equal to 0. If this condition is met, time position 157 is recorded as the amplitude candidate position. By traversing all first-order difference values ​​from the 1st position to the 478th position, a total of 89 amplitude candidate positions that meet the sign change condition are identified.

[0090] A preset fluctuation threshold is set as the quantitative standard for the amplitude change in the encrypted domain. The threshold is determined based on the parameter fluctuation range during normal operation of the natural gas pipeline. The 89 candidate amplitude locations are screened for locations where the amplitude change in the encrypted domain exceeds the preset fluctuation threshold. The amplitude change in the encrypted domain at location 74 exceeds the threshold, and the amplitude change in the encrypted domain at location 157 also exceeds the threshold. The amplitude change in the encrypted domain at location 203 does not meet the threshold requirement and is excluded. After a complete screening, the amplitude change at 42 of the 89 candidate amplitude locations exceeds the preset fluctuation threshold. These 42 locations constitute the first sequence of amplitude markers.

[0091] The condition that the encrypted second-order difference value of the 98th position is greater than or equal to 0 and the encrypted second-order difference value of the 99th position is less than 0 is detected. When this condition is met, the time position 99 is recorded as the inflection point candidate position. The condition that the encrypted second-order difference value of the 187th position is less than 0 and the encrypted second-order difference value of the 188th position is greater than or equal to 0 is detected. When this condition is met, the time position 188 is recorded as the inflection point candidate position. By traversing all the second-order difference values ​​from the 1st position to the 477th position, a total of 67 inflection point candidate positions that meet the sign change condition are identified. The preset curvature threshold is set as the quantitative standard for the curvature change of the encrypted domain, and the positions of the 67 inflection point candidate positions whose encrypted domain curvature change exceeds the preset curvature threshold are screened. After screening, the curvature change of 26 inflection point candidate positions exceeds the threshold, forming the first sequence of inflection point markers. The 42 first-sequence amplitude markers and 26 first-sequence inflection point markers were arranged and merged in chronological order to form a first-sequence feature marker set containing 68 feature points, providing basic data for subsequent curve fitting and time alignment analysis.

[0092] Furthermore, the method of extracting a correlation anomaly feature pattern from the time-aligned first and second sequences by calculating a sliding window correlation coefficient, and comparing the correlation anomaly feature pattern with a preset fault discrimination threshold to obtain a fault analysis result of a distributed IoT operation and maintenance terminal includes:

[0093] The first sequence data after time alignment is obtained and recorded as , the second sequence data after time alignment is recorded as ,in The value ranges from 1 to the length of the sequence after time alignment integer variable, is the total number of data points in the time-aligned sequence; set the sliding window size to , the sliding step length is ; in the Extract the first sequence data in the window position and the second sequence data within the window , where g is the window position variable and its value ranges from 0 to The integer range of is the position variable in the window and has a value of 0 to The integer range of ;

[0094] Calculate the The local correlation coefficient of the window position for:

[0095] ;

[0096] in For the The mean of the first series of data in the window, For the The mean of the second series data in the window; the local correlation coefficient of all window positions Composed of correlation coefficient sequence, the standard deviation of the correlation coefficient sequence is calculated as , the absolute value The window position of is marked as abnormal window position, where is the mean of the correlation coefficient sequence, is the preset anomaly detection coefficient; the time interval corresponding to the abnormal window position and the local correlation coefficient value are constructed as a correlation anomaly feature pattern. When the number of abnormal window positions in the correlation anomaly feature pattern exceeds the preset fault discrimination threshold When the distributed IoT operation and maintenance terminal is faulty, it is determined that there is a fault.

[0097] Exemplarily, based on the optimal time offset value of -126.5 seconds determined after fine-tuning, the first sequence data and the second sequence data after time alignment are obtained for correlation anomaly analysis. After the second sequence is time-corrected according to -126.5 seconds, the two sequences are accurately time-aligned. The length of the sequence after time alignment is 480 data points, covering the complete observation time window from 2 a.m. to 6 a.m. In the first sequence data after time alignment, the first data point corresponds to the terminal monitoring parameters at 2:00:00 a.m., the second data point corresponds to the parameters at 2:00:30 a.m., and so on to the 480th data point corresponding to the parameters at 5:59:30 a.m. The second sequence data after time alignment undergoes a forward adjustment of 126.5 seconds to ensure accurate matching with the first sequence in the time dimension.

[0098] The sliding window size is set to 60 data points, corresponding to a 30-minute time window. This window size is sufficient to capture short- and medium-term variations in natural gas pipeline operating parameters. The sliding step size is set to 6 data points, corresponding to a 3-minute time interval. This step size setting ensures analytical accuracy while avoiding computational redundancy. With a sliding window size of 60 and a sliding step size of 6, 70 window positions can be set in a sequence of 480 data points, numbered from 0 to 69. The 0th window position extracts data points 1 to 60, the 1st window position extracts data points 7 to 66, the 2nd window position extracts data points 13 to 72, and so on, up to the 69th window position, which extracts data points 415 to 474.

[0099] At the zeroth window position, the first and second series of data within the window are extracted. The first series of data within the window contains 60 monitoring parameter values ​​from 2:00:00 AM to 2:29:30 AM, and the second series of data within the window contains 60 time-corrected operating condition parameter values ​​for the corresponding time period. At the first window position, 60 data points are extracted from 2:03:00 AM to 2:32:30 AM, and at the second window position, 60 data points are extracted from 2:06:00 AM to 2:35:30 AM. This sliding window approach allows for a gradual analysis of the correlation changes across different time periods throughout the four-hour observation period.

[0100] Calculate the mean of the first series of data in the zeroth window. Sum the 60 data points and divide by 60 to obtain a mean of 147.3. Calculate the mean of the second series of data in the zeroth window. Similarly, sum the 60 data points and divide by 60 to obtain a mean of 149.1. Calculate the deviation of each data point in the window from the corresponding mean. The deviation of the first data point in the first series from the mean is 145.8 minus 147.3, which equals negative 1.5. The deviation of the first data point in the second series from the corresponding mean is 143.4 minus 149.1, which equals negative 5.7. Calculate the sum of the products of the 60 deviations as the covariance numerator. The sum of the squared deviations of the first series is 108.47, and the sum of the squared deviations of the second series is 156.82. The local correlation coefficient is the covariance numerator divided by the square root of the product of the sum of squared deviations of the first and second series. The local correlation coefficient for the zeroth window position is 0.847.

[0101] The local correlation coefficient for the first window position is 0.823, the second window position is 0.856, and the third window position is 0.831. The local correlation coefficients for the 15th to 29th window positions, corresponding to the time period from 3:45 to 4:30 a.m., decrease significantly. The correlation coefficients for the 15th window position are 0.247, the 16th window position is 0.189, and the 17th window position is 0.156. The lowest correlation coefficient for the 23rd window position is only 0.092. The local correlation coefficients for the 70 window positions are combined into a correlation coefficient sequence with a mean of 0.734 and a standard deviation of 0.248.

[0102] The anomaly detection coefficient was set to 2.5, based on experience with anomaly detection in natural gas pipeline operations and maintenance. Window positions with absolute values ​​exceeding 2.5 standard deviations were marked as anomalous. The absolute difference between the correlation coefficient of 0.247 for the 15th window position and the mean of 0.734 was 0.487, exceeding the threshold of 0.620 (2.5 times the standard deviation of 0.248). Therefore, the 15th window position was marked as an anomaly. The correlation coefficients for the 16th through 29th window positions all deviated significantly from the mean and exceeded the anomaly detection threshold, resulting in 15 consecutive window positions being marked as anomalous. The time interval corresponding to these 15 anomalous window positions, from 3:45 AM to 4:30 AM, and the corresponding local correlation coefficient values ​​were constructed as a correlation anomaly feature pattern. The preset fault discrimination threshold is set to 10 windows. The number of abnormal window positions in the correlation abnormality feature pattern, 15, exceeds the threshold. Therefore, it is determined that the GT-1246 terminal has a fault condition between 3:45 and 4:30 in the morning, requiring further maintenance inspection and troubleshooting.

[0103] Example 2: Based on the same inventive concept, this embodiment also provides a distributed Internet of Things operation and maintenance terminal fault analysis system based on big data, which is used to execute the distributed Internet of Things operation and maintenance terminal fault analysis method based on big data.

[0104] It should be noted that, regarding the system in the above embodiment, the specific manner in which each module performs operations has been described in detail in the embodiment of the method, and will not be elaborated on here.

[0105] Finally, it should be noted that although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art can still modify the technical solutions described in the aforementioned embodiments, or make equivalent substitutions for some of the technical features therein. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principles of the present invention should be included in the scope of protection of the present invention.

Claims

1. A distributed IoT operation and maintenance terminal fault analysis method based on big data, characterized in that: The method comprises: Obtain the encrypted time series of monitoring data of the distributed IoT operation and maintenance terminal as the first series, and the encrypted time series of the operating parameters of the corresponding monitored equipment as the second series. Extract encryption features from the first series to obtain a first series feature marker point set. Extract encryption features from the second series to obtain a second series feature marker point set. The time coordinate value, amplitude coordinate value, and marker type code of each marker point in the first sequence feature marker point set are constructed as a first sequence marker point feature vector, and the time coordinate value, amplitude coordinate value, and marker point type code of each marker point in the second sequence feature marker point set are constructed as a second sequence marker point feature vector; the first sequence marker point feature vectors are mapped into a coordinate system to form a first sequence marker point coordinate set, and the second sequence marker point feature vectors are mapped into a coordinate system to form a second sequence marker point coordinate set, and the first sequence marker point coordinate set is fitted by a cubic spline interpolation fitting algorithm to obtain a first sequence fitting curve, and the second sequence marker point coordinate set is fitted to obtain a second sequence fitting curve; The first sequence fitting curve and the second sequence fitting curve are randomly time-shifted and matched, and the optimal time offset value is found through the genetic algorithm. The second sequence is time-corrected according to the optimal time offset value to achieve time alignment between the first and second sequences. The correlation anomaly feature pattern is extracted from the first and second sequences after time alignment through the sliding window correlation coefficient calculation method. The correlation anomaly feature pattern is compared with the preset fault discrimination threshold to obtain the fault analysis result of the distributed Internet of Things operation and maintenance terminal.

2. The method for analyzing faults of distributed IoT operation and maintenance terminals based on big data according to claim 1, characterized in that: The method of performing random time offset matching on the first sequence fitting curve and the second sequence fitting curve and finding the optimal time offset value by using a genetic algorithm includes: A plurality of time offset candidate values ​​are generated within a preset time offset interval. For each time offset candidate value, the second sequence fitting curve is time-shifted. Sampling coordinate points are selected at equal intervals on the first sequence fitting curve. The sum of the squares of the Euclidean distances between each sampling coordinate point and the corresponding coordinate point of the second sequence fitting curve after the shift is calculated and recorded as the curve matching error value. The time offset candidate values ​​and the corresponding curve matching error values ​​are constructed into a time offset candidate scheme. The time offset candidate scheme is used as the initial population individual of the genetic algorithm. The population individuals with smaller curve matching error values ​​are screened through the genetic algorithm selection operation. New time offset candidate values ​​are generated through crossover and mutation operations. The optimal time offset value is obtained by iterative optimization with the minimization of the curve matching error value as the fitness goal.

3. The method for analyzing faults of distributed IoT operation and maintenance terminals based on big data according to claim 2 is characterized in that: The method of calculating the sum of squares of the Euclidean distances between each sampling coordinate point and the corresponding coordinate point of the translated second sequence fitting curve as a curve matching error value, and constructing the time offset candidate value and the corresponding curve matching error value as a time offset candidate solution includes: The total number of sampling points is set to be equal intervals within the time range of the first series fitting curve. , No. The time coordinate of each sampling point is marked as , the corresponding amplitude coordinates are marked as , is the sampling point number and its value ranges from 1 to Integer variable for time offset candidate value , calculate the second sequence fitting curve after translation in the time coordinate The corresponding amplitude coordinates at ; Construct a matching error calculation formula based on information entropy weighting, and the curve matching error value for: ； in is the derivative weight coefficient, and They are the first sequence fitting curve and the second sequence fitting curve after translation. The first-order derivative at ; Calculate the comprehensive fitness value of the time offset candidate value for: ； in For the preset shape similarity control parameter, the time offset candidate value And the corresponding comprehensive fitness value Constructed as a candidate for time shifting.

4. The method for analyzing faults of distributed IoT operation and maintenance terminals based on big data according to claim 2, characterized in that: The method further comprises: Set the neighborhood range of the optimal time offset value as the fine-tuning interval, and set the fine-tuning step size within the fine-tuning interval to ,generate Fine-tuning candidate offset values , is the fine-tuning sequence number and its value ranges from 1 to Integer variable; for each fine-tuning candidate offset value Recalculate the fine matching degree of the first series fitting curve and the second series fitting curve for: ； in is the preset matching accuracy control parameter, To fine-tune the candidate offset value The second sequence fitting curve after translation is in the time coordinate The amplitude coordinates at ; find the fine matching degree by traversal search The fine-tuned candidate offset value corresponding to the maximum value of .

5. The method for analyzing faults of distributed IoT operation and maintenance terminals based on big data according to claim 1, characterized in that: The method of extracting encryption features from the first sequence to obtain a first sequence feature marker point set, and extracting encryption features from the second sequence to obtain a second sequence feature marker point set includes: Performing a first-order difference operation on the first sequence to obtain first-order difference data of the first sequence, performing a second-order difference operation on the first-order difference data of the first sequence to obtain second-order difference data of the first sequence, identifying the sign change position in the first-order difference data of the first sequence as the amplitude marking point of the first sequence through an encryption domain sign judgment algorithm, identifying the sign change position in the second-order difference data of the first sequence as the inflection point marking point of the first sequence, and merging the amplitude marking point of the first sequence and the inflection point marking point of the first sequence to form a first sequence characteristic marking point set; A first-order difference operation is performed on the second sequence to obtain first-order difference data of the second sequence, and a second-order difference operation is performed on the first-order difference data of the second sequence to obtain second-order difference data of the second sequence. The sign change position in the first-order difference data of the second sequence is identified as the amplitude marking point of the second sequence through the encryption domain sign judgment algorithm, and the sign change position in the second-order difference data of the second sequence is identified as the inflection point marking point of the second sequence. The second sequence amplitude marking point and the second sequence inflection point marking point are merged to form a second sequence characteristic marking point set.

6. The method for analyzing faults of distributed IoT operation and maintenance terminals based on big data according to claim 5 is characterized in that: The method of performing a first-order difference operation on a first sequence to obtain first-order difference data of the first sequence, performing a second-order difference operation on the first-order difference data of the first sequence to obtain second-order difference data of the first sequence, identifying a sign change position in the first-order difference data of the first sequence as a first-sequence amplitude marking point through an encryption domain sign judgment algorithm, identifying a sign change position in the second-order difference data of the first sequence as a first-sequence inflection point marking point, and merging the first-sequence amplitude marking point and the first-sequence inflection point marking point to form a first-sequence characteristic marking point set includes: The total number of encrypted time series data points of the first sequence is recorded as , in the first sequence The encrypted data at a time point is recorded as , The value ranges from 1 to Integer variable; calculate the first order difference data of the first sequence, The first-order difference encrypted value of the position for: ； in is the encryption domain subtraction operator, The value ranges from 1 to Integer variable; calculate the first series of second-order difference data, the The second-order differential encrypted value of the position for: ； in The value ranges from 1 to Integer variable; detect the first-order difference encryption value through the encryption domain symbol judgment algorithm and The sign changes when When the conditions are met, the time position Record as the candidate amplitude position, and filter the encrypted domain amplitude change exceeding the preset fluctuation threshold The amplitude candidate positions form the first sequence of amplitude mark points; the second-order differential encryption value is detected by the encryption domain symbol judgment algorithm and The sign changes when When the conditions are met, the time position Record as the candidate inflection point position, and filter the encrypted domain where the curvature change exceeds the preset curvature threshold The candidate positions of the first sequence of inflection point marking points are formed; the first sequence of amplitude marking points and the first sequence of inflection point marking points are arranged and merged in chronological order to form a first sequence of feature marking point set.

7. The method for analyzing faults of distributed IoT operation and maintenance terminals based on big data according to claim 6, characterized in that: The method of extracting a correlation anomaly feature pattern from the time-aligned first and second sequences by using a sliding window correlation coefficient calculation method, and comparing the correlation anomaly feature pattern with a preset fault discrimination threshold to obtain a distributed Internet of Things operation and maintenance terminal fault analysis result includes: The first sequence data after time alignment is recorded as , the second sequence data after time alignment is recorded as ,in The value ranges from 1 to the length of the sequence after time alignment integer variable, is the total number of data points in the time-aligned sequence; set the sliding window size to , the sliding step length is ; in the Extract the first sequence data in the window position and the second sequence data within the window , where g is the window position variable and its value ranges from 0 to The integer range of is the position variable in the window and has a value of 0 to The integer range of ; Calculate the The local correlation coefficient of the window position for: ； in For the The mean of the first series of data in the window, For the The mean of the second series data in the window; the local correlation coefficient of all window positions Composed of correlation coefficient sequence, the standard deviation of the correlation coefficient sequence is calculated as , the absolute value The window position of is marked as abnormal window position, where is the mean of the correlation coefficient sequence, is the preset anomaly detection coefficient; the time interval corresponding to the abnormal window position and the local correlation coefficient value are constructed as a correlation anomaly feature pattern. When the number of abnormal window positions in the correlation anomaly feature pattern exceeds the preset fault discrimination threshold When the distributed IoT operation and maintenance terminal is faulty, it is determined that there is a fault.

8. A distributed IoT operation and maintenance terminal fault analysis system based on big data, characterized by: The system is used to execute the distributed Internet of Things operation and maintenance terminal fault analysis method based on big data as described in any one of claims 1-7.

Images