Perforation oil stealing branch pipe data identification system and method based on neural algorithm

Abstract

The invention relates to the technical field of pipeline monitoring, and discloses a punching oil stealing branch pipe data identification system and method based on a neural algorithm, and the system comprises a collection unit which collects sensor data in a to-be-monitored area, and judges whether a continuous monitoring mode is started or not; when it is judged that the continuous monitoring mode is started, the processing unit inputs a preset fuzzy algorithm to obtain continuous monitoring duration according to the obtained vibration exceeding value and the magnetic field gradient descent value; the judgment unit continuously records sensor data in continuous monitoring duration, establishes a sensor data set, establishes a current feature set, compares the current feature set with a historical early warning data set, and determines an early warning level in combination with a neural network model; the adjusting unit establishes a vibration data coordinate system and a magnetic field gradient data coordinate system to obtain data inflection points in the coordinate systems, and judges whether the early warning level is adjusted or not; and the storage unit stores the current feature set and the early warning level. According to the invention, the accuracy of anomaly detection is improved, and the limitation of a static threshold and external interference is avoided.

Description

Technical Field

[0001] The present invention relates to the technical field of pipeline monitoring, and in particular to a neural algorithm-based perforated oil theft branch pipe data recognition system and method. Background Art

[0002] During the operation of oil and gas pipelines and branch pipes, oil theft seriously threatens the safety and economic benefits of pipelines. Especially when oil theft occurs in pipelines through drilling, it will not only cause oil and gas leakage, but also may cause environmental pollution and economic losses. Therefore, how to efficiently and accurately detect and identify oil theft through drilling has become a key issue in the current field of oil and gas pipeline monitoring.

[0003] Traditional oil and gas pipeline detection methods mostly rely on manual inspections, regular inspections, and single sensor monitoring. There are some problems. For example, traditional detection methods are usually unable to detect abnormal behavior in pipelines in real time, especially the more concealed oil theft method such as drilling and stealing oil. It is easy to miss or misdetect when oil theft occurs. Manual inspections and regular inspections are usually not able to be carried out 24 hours a day, and are greatly affected by factors such as weather and environment, with a high probability of missed detection and delayed response. Even if sensors are used for detection, current detection systems mostly rely on static alarm thresholds for judgment, but this method lacks adaptability to complex scenarios and is easily interfered by external environmental factors, resulting in large alarm errors. Moreover, it is impossible to obtain more accurate detection results through multi-dimensional data fusion.

[0004] Therefore, it is necessary to design a neural algorithm-based perforated oil theft branch pipe data recognition system and method to solve the problems existing in the current technology. Summary of the invention

[0005] In view of this, the present invention proposes a perforated oil theft branch pipe data identification system and method based on neural algorithm, aiming to solve the current problems of reliance on manual inspection, insufficient monitoring accuracy and delayed data analysis.

[0006] In one aspect, the present invention proposes a neural algorithm-based perforated oil theft branch pipe data recognition system, comprising: A collection unit is configured to collect sensor data in the area to be monitored, wherein the sensor data includes vibration data and magnetic field gradient data, and determine whether to start the continuous monitoring mode according to the sensor data; a processing unit configured to, when the acquisition unit determines to start the continuous monitoring mode, obtain a vibration excess value according to the vibration data and the vibration data threshold, obtain a magnetic field gradient decrease value according to the magnetic field gradient data and the magnetic field gradient threshold, input the vibration excess value and the magnetic field gradient decrease value into a preset fuzzy algorithm, and obtain a continuous monitoring duration of the continuous monitoring mode according to an output result; A judgment unit is configured to continuously record sensor data within the continuous monitoring period, establish a sensor data set, extract maximum vibration data and minimum magnetic field gradient data and establish a current feature set, compare the current feature set with a historical warning data set, and determine a warning level in combination with a neural network model; the historical warning data set includes a plurality of historical feature sets and a plurality of historical warning levels, each of the historical feature sets corresponds to a historical warning level, and the historical feature sets include historical maximum vibration data and historical minimum magnetic field gradient data; an adjustment unit configured to establish a vibration data coordinate system and a magnetic field gradient data coordinate system, mark the vibration data in the sensor data set into the vibration data coordinate system in time sequence, mark the magnetic field gradient data in the sensor data set into the magnetic field gradient data coordinate system in time sequence, obtain a data inflection point in the coordinate system according to the vibration data coordinate system and the magnetic field gradient data coordinate system after the marked data, determine whether to adjust the warning level according to the data inflection point, and when it is determined that the warning level is to be adjusted, issue a warning at the adjusted warning level; The storage unit is configured to store the current feature set and warning level.

[0007] Furthermore, when the acquisition unit determines whether to start the continuous monitoring mode according to the sensor data, it includes: The acquisition unit compares the vibration data with the vibration data threshold, and compares the magnetic field gradient data with the magnetic field gradient data threshold, and determines whether to start the continuous monitoring mode according to the comparison results; When the vibration data is greater than a vibration data threshold and the magnetic field gradient data is less than a magnetic field gradient data threshold, determining to start the continuous monitoring mode; When the vibration data is less than or equal to the vibration data threshold or the magnetic field gradient data is greater than or equal to the magnetic field gradient data threshold, it is determined that the continuous monitoring mode is not turned on, and a Gaussian mixture distribution is used to smooth the signal.

[0008] Further, when the acquisition unit determines not to start the continuous monitoring mode and uses Gaussian mixture distribution to smooth the signal, it includes: When the vibration data is greater than the vibration data threshold and the magnetic field gradient data is greater than or equal to the magnetic field gradient data threshold, the vibration data is recorded as abnormal data; When the vibration data is less than or equal to the vibration data threshold and the magnetic field gradient data is less than the magnetic field gradient data threshold, recording the magnetic field gradient data as abnormal data; Retrieving data of a previous time period of a data type corresponding to the abnormal data, processing the retrieved data using a Gaussian mixture distribution to determine smoothed data, and changing the abnormal data at the current time period to the smoothed data; When Gaussian mixture distribution is used to process the retrieved data to determine the smoothed data, the kernel density function expression is:

[0009] Where n represents the number of data in the retrieved data, h represents the smoothing bandwidth, Indicates retrieving the i-th data in the data, and x indicates the value of a certain frequency point to be estimated; The data value corresponding to the highest frequency of the kernel density is used as the smoothed data.

[0010] Furthermore, the processing unit inputs the vibration excess value and the magnetic field gradient decrease value into a preset fuzzy algorithm, and obtains the continuous monitoring time of the continuous monitoring mode according to the output result, including: Define the vibration excess value and the magnetic field gradient drop value as input variables, and define the continuous monitoring duration as the output variable; Dividing the vibration excess value into a plurality of vibration excess value fuzzy sets according to the magnitude of the vibration excess value, and dividing the magnetic field gradient decrease value into a plurality of magnetic field gradient decrease value fuzzy sets according to the magnitude of the magnetic field gradient decrease value, and determining a fuzzy rule base according to the vibration excess value fuzzy sets and the magnetic field gradient decrease value fuzzy sets; The current vibration excess value and magnetic field gradient decrease value are compared with the vibration excess value fuzzy set and the magnetic field gradient decrease value fuzzy set respectively, and the vibration excess value membership and the magnetic field gradient decrease value membership are determined based on the triangular membership function; Calculating the output value of each rule in the fuzzy rule base according to the vibration excess value membership and the magnetic field gradient decrease value membership; Defuzzify the output value of each rule by weighted summing the following formula, and obtain a total output value, where the total output value is the continuous monitoring duration;

[0011] Among them, E represents the continuous monitoring duration, E0 represents the continuous monitoring duration reference value, Qi represents the output weight of the i-th rule, and Si represents the output value of the i-th rule.

[0012] Furthermore, when the judgment unit compares the current feature set with the historical warning data set and determines the warning level in combination with the neural network model, it includes: The judgment unit calculates the similarity between the current feature set and each historical feature set respectively; When there is data in all the historical feature sets whose similarity with the current feature set is greater than a similarity threshold, determining the warning level according to the historical warning levels corresponding to the historical feature sets whose similarity is greater than the similarity threshold; When the similarities between all the historical feature sets and the current feature set are less than a similarity threshold, the neural network model is established according to the historical warning data group, and the current feature set is input into the neural network model to obtain the warning level.

[0013] Furthermore, when the judgment unit determines the warning level according to the historical warning level corresponding to the historical feature set whose similarity is greater than the similarity threshold, it includes: When there is only one data in the historical feature set whose similarity with the current feature set is greater than the similarity threshold, the historical warning level corresponding to the historical feature set is used as the warning level; When there is not unique data in the historical feature set whose similarity with the current feature set is greater than a similarity threshold, the highest historical warning level corresponding to the historical feature set is used as the warning level.

[0014] Furthermore, the judgment unit establishes the neural network model according to the historical warning data group, inputs the current feature set into the neural network model, and obtains the warning level, including: Sampling the historical warning data group according to a preset ratio to obtain a training subset and a test subset; Acquire a pre-selected convolutional neural model, iteratively train the convolutional neural model according to the training subset, and evaluate the iteratively trained convolutional neural model according to the test subset to obtain the neural network model; The current feature set is input into the neural network model to obtain the warning level.

[0015] Further, when the adjustment unit determines whether to adjust the warning level according to the data inflection point, it includes: When there is an inflection point in the vibration data coordinate system or the magnetic field gradient data coordinate system, the adjustment unit determines to adjust the warning level; When there is no inflection point in either the vibration data coordinate system or the magnetic field gradient data coordinate system, the adjustment unit determines not to adjust the warning level.

[0016] Further, when the adjustment unit determines to adjust the warning level, it includes: When there is an inflection point in one of the vibration data coordinate system and the magnetic field gradient data coordinate system, the warning level is raised by one level; When both the vibration data coordinate system and the magnetic field gradient data coordinate system have inflection points, the warning level is increased by two levels.

[0017] Compared with the prior art, the beneficial effects of the present invention are as follows: by integrating vibration data and magnetic field gradient data, combining neural network models and fuzzy algorithms, it is possible to collect and analyze multi-dimensional sensor data in real time, dynamically start the continuous monitoring mode when an abnormal signal is detected, and adaptively calculate the continuous monitoring duration through the fuzzy algorithm. It effectively reduces manual intervention and improves the degree of automation of monitoring. By establishing a comparison between historical data and the current feature set, and combining neural networks to make warning level judgments, it is possible to accurately identify oil theft behaviors based on different detection results. The use of coordinate system analysis methods, combined with time series data and data inflection point judgments, further improves the accuracy and flexibility of anomaly detection, avoids the limitations of static thresholds and external interference in traditional detection methods, and can be adjusted according to changes in complex scenarios to ensure stable and reliable detection results under various environmental conditions. It effectively improves the accuracy and response speed of oil and gas pipeline monitoring.

[0018] On the other hand, the present application also provides a method for identifying perforated oil theft branch pipe data based on a neural algorithm, which is applied to the above-mentioned perforated oil theft branch pipe data identification system based on a neural algorithm, comprising: Collecting sensor data in the area to be monitored, the sensor data including vibration data and magnetic field gradient data, and determining whether to start the continuous monitoring mode according to the sensor data; When it is determined that the continuous monitoring mode is turned on, a vibration excess value is obtained according to the vibration data and the vibration data threshold, a magnetic field gradient decrease value is obtained according to the magnetic field gradient data and the magnetic field gradient threshold, the vibration excess value and the magnetic field gradient decrease value are input into a preset fuzzy algorithm, and the continuous monitoring duration of the continuous monitoring mode is obtained according to the output result; Continuously record the sensor data within the continuous monitoring period, establish a sensor data set, extract the maximum vibration data and the minimum magnetic field gradient data and establish a current feature set, compare the current feature set with the historical warning data group, and determine the warning level in combination with the neural network model; the historical warning data group includes a plurality of historical feature sets and a plurality of historical warning levels, each of the historical feature sets corresponds to a historical warning level, and the historical feature sets include historical maximum vibration data and historical minimum magnetic field gradient data; Establishing a vibration data coordinate system and a magnetic field gradient data coordinate system, marking the vibration data in the sensor data set into the vibration data coordinate system in time sequence, marking the magnetic field gradient data in the sensor data set into the magnetic field gradient data coordinate system in time sequence, obtaining the data inflection point in the coordinate system according to the vibration data coordinate system and the magnetic field gradient data coordinate system after the marked data, judging whether to adjust the warning level according to the data inflection point, and issuing a warning at the adjusted warning level when it is judged that the warning level is to be adjusted; The current feature set and warning level are stored.

[0019] It can be understood that the above-mentioned neural algorithm-based perforated oil theft branch pipe data identification system and method have the same beneficial effects, which will not be repeated here. BRIEF DESCRIPTION OF THE DRAWINGS

[0020] Various other advantages and benefits will become apparent to those of ordinary skill in the art by reading the detailed description of the preferred embodiments below. The accompanying drawings are only for the purpose of illustrating the preferred embodiments and are not to be considered as limiting the present invention. Moreover, the same reference symbols are used throughout the accompanying drawings to represent the same components. In the accompanying drawings: Figure 1 A structural block diagram of a neural algorithm-based perforated oil theft branch pipe data recognition system provided in an embodiment of the present invention; Figure 2 A flowchart of a method for identifying branch pipe data for oil theft based on a neural algorithm provided in an embodiment of the present invention. DETAILED DESCRIPTION

[0021] Exemplary embodiments of the present disclosure will be described in more detail below with reference to the accompanying drawings. Although exemplary embodiments of the present disclosure are shown in the accompanying drawings, it should be understood that the present disclosure can be implemented in various forms and should not be limited by the embodiments described herein. On the contrary, these embodiments are provided in order to enable a more thorough understanding of the present disclosure and to fully convey the scope of the present disclosure to those skilled in the art. It should be noted that, in the absence of conflict, the embodiments of the present invention and the features described in the embodiments can be combined with each other. The present invention will be described in detail below with reference to the accompanying drawings and in combination with the embodiments.

[0022] In some embodiments of the present application, see Figure 1 As shown, a data identification system for drilling and stealing oil branch pipes based on neural algorithm includes: a collection unit, a processing unit, a judgment unit, an adjustment unit and a storage unit, wherein: The acquisition unit is configured to acquire sensor data in the area to be monitored, the sensor data including vibration data and magnetic field gradient data, and determine whether to start the continuous monitoring mode based on the sensor data.

[0023] The processing unit is configured such that when the acquisition unit determines to start the continuous monitoring mode, the processing unit obtains the vibration excess value based on the vibration data and the vibration data threshold, obtains the magnetic field gradient decrease value based on the magnetic field gradient data and the magnetic field gradient threshold, inputs the vibration excess value and the magnetic field gradient decrease value into a preset fuzzy algorithm, and obtains the continuous monitoring time of the continuous monitoring mode based on the output result.

[0024] The judgment unit is configured to continuously record sensor data within the continuous monitoring period, establish a sensor data set, extract maximum vibration data and minimum magnetic field gradient data and establish a current feature set, compare the current feature set with the historical warning data set, and determine the warning level in combination with the neural network model. The historical warning data set includes a number of historical feature sets and a number of historical warning levels, each historical feature set corresponds to a historical warning level, and the historical feature set includes historical maximum vibration data and historical minimum magnetic field gradient data.

[0025] The adjustment unit is configured to establish a vibration data coordinate system and a magnetic field gradient data coordinate system, mark the vibration data in the sensor data set into the vibration data coordinate system in time sequence, mark the magnetic field gradient data in the sensor data set into the magnetic field gradient data coordinate system in time sequence, obtain the data inflection point in the coordinate system according to the vibration data coordinate system and the magnetic field gradient data coordinate system after the marked data, determine whether to adjust the warning level according to the data inflection point, and when it is determined that the warning level is to be adjusted, issue a warning at the adjusted warning level.

[0026] The storage unit is configured to store the current feature set and the warning level.

[0027] Specifically, the acquisition unit collects sensor data inside and outside the pipeline in real time, mainly including vibration data and magnetic field gradient data. Vibration data reflects the vibration of the pipeline caused by pressure changes or external factors, while magnetic field gradient data can monitor the changes in the magnetic field around the pipeline, especially in the case of drilling and oil theft, the magnetic field of the pipeline will change significantly. Based on these sensor data, the acquisition unit determines whether it is necessary to turn on the continuous monitoring mode.

[0028] When the acquisition unit detects that the continuous monitoring mode needs to be turned on, the processing unit compares the collected vibration data and magnetic field gradient data with their respective thresholds. The specific process is as follows: Calculation of vibration excess value: If the collected vibration data exceeds the set vibration threshold, the processing unit will calculate the vibration excess value, that is, the difference between the vibration data and the vibration threshold. If the magnetic field gradient data is lower than the preset magnetic field gradient threshold, the processing unit will calculate the magnetic field gradient drop value, that is, the difference between the magnetic field gradient threshold and the actual magnetic field gradient. These values ​​(vibration excess value and magnetic field gradient drop value) will be input into the preset fuzzy algorithm, and the fuzzy algorithm will obtain the duration of continuous monitoring based on these data. The continuous monitoring duration can adapt to different abnormal situations and dynamically adjust the monitoring cycle according to the changes in vibration and magnetic field.

[0029] The judgment unit records the sensor data during the continuous monitoring period, and performs feature extraction and comparison based on the recorded data: the judgment unit extracts the maximum vibration data and the minimum magnetic field gradient data from the collected sensor data set, and forms a current feature set with these data. The current feature set will be compared with the historical feature set in the historical warning data group, which contains the historical feature set and the corresponding warning level. Each historical feature set contains the historical maximum vibration data and the historical minimum magnetic field gradient data. Through this comparison, it is identified whether the current monitoring data is consistent with the historical warning data, thereby determining the current warning level.

[0030] The adjustment unit adjusts the warning level through the following steps: The adjustment unit will establish coordinate systems for vibration data and magnetic field gradient data respectively, and mark the data in the sensor data set into these two coordinate systems in time series. In the vibration data coordinate system and the magnetic field gradient data coordinate system, the adjustment unit will look for inflection points in the data. The data inflection point refers to the place where the data change trend changes significantly, which usually means that the pipeline status is abnormal. By analyzing these inflection points, it is determined whether the warning level needs to be adjusted. If the current warning level needs to be adjusted based on the data inflection point analysis, the adjustment unit will readjust the warning level and trigger a new warning.

[0031] The storage unit stores the current feature set and warning level for comparison and reference in subsequent monitoring. The stored data can help the system make more accurate judgments and responses when facing similar abnormal situations using historical data and warning levels.

[0032] It is understandable that by combining vibration data and magnetic field gradient data, the abnormal state of the pipeline can be fully and accurately reflected, especially in the detection of highly concealed abnormal behaviors such as drilling and oil theft, which can effectively avoid the blind spots of single sensor detection. The fuzzy algorithm can dynamically adjust the duration of continuous monitoring according to the data collected in real time, ensure the flexibility and accuracy of monitoring in different scenarios, and reduce invalid monitoring and delayed response. By comparing the current feature set with the historical warning data group and combining it with the neural network model for judgment, it can accurately identify anomalies and give a reasonable warning level. The intelligent analysis of historical and real-time data improves the accuracy of detection and reduces the risk of false alarms and missed alarms. The adjustment unit can adjust the warning level in time according to the changes in the pipeline status through coordinate system analysis and data inflection point judgment, further improving the real-time and accuracy of the alarm mechanism.

[0033] In some embodiments of the present application, when the acquisition unit determines whether to turn on the continuous monitoring mode based on the sensor data, it includes: the acquisition unit compares the vibration data with the vibration data threshold, and compares the magnetic field gradient data with the magnetic field gradient data threshold, and determines whether to turn on the continuous monitoring mode based on the comparison results.

[0034] When the vibration data is greater than the vibration data threshold and the magnetic field gradient data is less than the magnetic field gradient data threshold, it is determined that the continuous monitoring mode is turned on.

[0035] When the vibration data is less than or equal to the vibration data threshold or the magnetic field gradient data is greater than or equal to the magnetic field gradient data threshold, it is determined that the continuous monitoring mode is not turned on, and the Gaussian mixture distribution is used to smooth the signal.

[0036] In some embodiments of the present application, when the acquisition unit determines not to start the continuous monitoring mode and uses Gaussian mixture distribution to smooth the signal, it includes: When the vibration data is greater than the vibration data threshold and the magnetic field gradient data is greater than or equal to the magnetic field gradient data threshold, the vibration data is recorded as abnormal data.

[0037] When the vibration data is less than or equal to the vibration data threshold and the magnetic field gradient data is less than the magnetic field gradient data threshold, the magnetic field gradient data is recorded as abnormal data.

[0038] The data of the previous time period of the data type corresponding to the abnormal data is retrieved, and the retrieved data is processed using Gaussian mixture distribution to determine smooth data, and the abnormal data at the current time is changed into smooth data.

[0039] When Gaussian mixture distribution is used to process the retrieved data to determine the smoothed data, the kernel density function expression is:

[0040] Where n represents the number of data in the retrieved data, h represents the smoothing bandwidth, It means to retrieve the i-th data in the data, and x represents the value of a certain frequency point to be estimated.

[0041] The data value corresponding to the highest frequency of the kernel density is used as the smoothed data.

[0042] Specifically, the acquisition unit compares the vibration data collected in real time with the preset vibration data threshold, and compares the magnetic field gradient data with the preset magnetic field gradient data threshold. When the vibration data is greater than the vibration threshold and the magnetic field gradient data is less than the magnetic field gradient threshold, it is considered that there is an abnormal situation (drilling and oil theft occurs). When drilling is performed, the pipeline vibrates abnormally, and when the pipeline operates normally, a stable electromagnetic gradient signal is formed between the fluid and the pipeline. When drilling is performed, the original flow state is disturbed, causing the electromagnetic gradient to decrease, thereby determining to turn on the continuous monitoring mode; if the vibration data is lower than or equal to the vibration threshold, or the magnetic field gradient data is higher than or equal to the magnetic field gradient threshold, it is determined that the continuous monitoring mode is not turned on. When it is determined that the continuous monitoring mode is not turned on, short-term abnormal fluctuations may be caused by environmental noise or other factors. At this time, Gaussian mixture distribution is used to smooth the signal to eliminate noise interference. If the vibration data is greater than the vibration threshold and the magnetic field gradient data is greater than or equal to the magnetic field gradient threshold, the vibration data is regarded as abnormal data; if the vibration data is less than or equal to the vibration threshold and the magnetic field gradient data is less than the magnetic field gradient threshold, the magnetic field gradient data is regarded as abnormal data; for data marked as abnormal, the historical data of the corresponding data type in the previous period of time is retrieved, and then the Gaussian mixture distribution method is used to perform signal smoothing.

[0043] It is understandable that the use of threshold comparison to determine the start-up conditions of the monitoring mode and the use of Gaussian mixture distribution for signal smoothing in the non-continuous monitoring state effectively reduce the risk of false alarms caused by short-term noise and abnormal data fluctuations. By estimating the kernel density of historical data, representative data after smoothing can be obtained to ensure the accuracy of subsequent feature extraction and neural network warning, thereby improving the robustness and reliability of the overall perforation oil theft detection system.

[0044] In some embodiments of the present application, the processing unit inputs the vibration excess value and the magnetic field gradient decrease value into a preset fuzzy algorithm, and obtains the continuous monitoring time of the continuous monitoring mode according to the output result, including: defining the vibration excess value and the magnetic field gradient decrease value as input variables, and defining the continuous monitoring time as the output variable.

[0045] The vibration excess value is divided into several vibration excess value fuzzy sets, and the magnetic field gradient descent value is divided into several magnetic field gradient descent value fuzzy sets. The fuzzy rule base is determined according to the vibration excess value fuzzy sets and the magnetic field gradient descent value fuzzy sets.

[0046] The current vibration excess value and magnetic field gradient descent value are compared with the vibration excess value fuzzy set and the magnetic field gradient descent value fuzzy set respectively, and the vibration excess value membership and the magnetic field gradient descent value membership are determined based on the triangle membership function.

[0047] The output value of each rule in the fuzzy rule base is calculated according to the membership of the vibration excess value and the membership of the magnetic field gradient decrease value.

[0048] The output value of each rule is defuzzified by weighted summation using the following formula, and the total output value is obtained. The total output value is the continuous monitoring duration.

[0049]

[0050] Among them, E represents the continuous monitoring duration, E0 represents the continuous monitoring duration reference value, Qi represents the output weight of the i-th rule, and Si represents the output value of the i-th rule.

[0051] Specifically, according to the size of the vibration exceedance value, it is divided into several fuzzy sets, such as "small vibration exceedance", "medium vibration exceedance" and "large vibration exceedance". Similarly, according to the size of the magnetic field gradient decrease value, it is divided into several fuzzy sets, such as "slight decrease", "medium decrease" and "significant decrease". By fuzzifying the input variables and converting them into processable fuzzy sets, it helps to adapt and flexibly deal with different input data. The fuzzy rule base contains a series of fuzzy rules for calculating the output results based on the membership values ​​of the input variables. The fuzzy rule base is usually established by historical determinations. For example, if the vibration exceedance value is "large vibration exceedance" and the magnetic field gradient decrease value is "significant decrease", the continuous monitoring time needs to be longer; if the vibration exceedance value is "small vibration exceedance" and the magnetic field gradient decrease value is "slight decrease", the continuous monitoring time is shorter; the continuous monitoring time is dynamically adjusted according to the actual situation. The reference value of the continuous monitoring time can be set according to the complex situation around the specific pipeline, which can be 0.5h or 1h.

[0052] It is understandable that through fuzzy algorithms and membership calculations, the duration of continuous monitoring can be dynamically adjusted according to the vibration data and magnetic field gradient data collected in real time. It can effectively monitor the pipeline status under different environments and operating conditions, avoiding the errors caused by the traditional static threshold method. Through multi-dimensional fuzzy reasoning and weighted summation, multiple factors can be combined to determine the duration of continuous monitoring, thereby improving the accuracy of anomaly detection, especially in complex scenarios. It can better identify highly concealed abnormal behaviors such as drilling and oil theft. The combination of Gaussian mixture distribution and fuzzy reasoning helps to smooth noise data and abnormal signals, reduce the risk of false alarms caused by environmental interference, and ensure that key abnormal signals are effectively detected. By establishing and adjusting the fuzzy rule base, the rule base can be optimized on the basis of continuous accumulation of data and experience, thereby gradually improving the accuracy of early warning and the intelligence level of the system.

[0053] In some embodiments of the present application, when the judgment unit compares the current feature set with the historical warning data group and determines the warning level in combination with the neural network model, it includes: The judgment unit calculates the similarity between the current feature set and each historical feature set respectively.

[0054] When there is data in all historical feature sets whose similarity with the current feature set is greater than the similarity threshold, the warning level is determined according to the historical warning levels corresponding to the historical feature sets whose similarity is greater than the similarity threshold.

[0055] When the similarity between all historical feature sets and the current feature set is less than the similarity threshold, a neural network model is established based on the historical warning data group, and the current feature value is input into the neural network model to obtain the warning level.

[0056] In some embodiments of the present application, when the judgment unit determines the warning level according to the historical warning level corresponding to the historical feature set whose similarity is greater than the similarity threshold, it includes: when the data in the historical feature set whose similarity is greater than the similarity threshold with the current feature set is unique, the historical warning level corresponding to the historical feature set is used as the warning level. When the data in the historical feature set whose similarity is greater than the similarity threshold with the current feature set is not unique, the highest historical warning level corresponding to the historical feature set is used as the warning level.

[0057] In some embodiments of the present application, the judgment unit establishes a neural network model based on the historical warning data group, inputs the current feature value into the neural network model, and obtains the warning level, including: sampling the historical warning data group according to a preset ratio to obtain a training subset and a test subset. Obtain a pre-selected convolutional neural model, and iteratively train the convolutional neural model according to the training subset, and evaluate the iteratively trained convolutional neural model according to the test subset to obtain a neural network model. Input the current feature set into the neural network model to obtain the warning level.

[0058] Specifically, the similarity between historical data and current data is used to determine whether the same type of abnormal behavior exists. If similar historical data exists, the current warning level is quickly given based on the historical warning level. When the similarity is low, the abnormal patterns in the historical data are learned through the neural network model to identify abnormal behaviors in the current data and give warning levels. Neural networks are particularly suitable for identifying complex abnormal patterns that are difficult to clearly determine using rules or thresholds.

[0059] It is understandable that by combining historical data similarity calculation and neural network model, the most appropriate warning method is automatically selected in different scenarios, ensuring the flexibility and accuracy of anomaly detection. When the historical data is similar to the current feature set, the warning result is given based on the historical warning level, which reduces the dependence on complex algorithms and improves the response speed. If the current data is less similar to the historical data, the neural network model can accurately identify potential drilling and oil theft behaviors based on the abnormal patterns learned from the historical data. The detection method based on deep learning can handle complex anomalies that are difficult to identify through traditional rules. It does not rely on human intervention and automatically adjusts the warning level according to the similarity of the data and the learning results of the neural network, which improves the intelligence level of the system and reduces the possibility of human errors and missed detections.

[0060] In some embodiments of the present application, when the adjustment unit determines whether to adjust the warning level according to the data inflection point, it includes: when there is an inflection point in the vibration data coordinate system or the magnetic field gradient data coordinate system, the adjustment unit determines to adjust the warning level. When there is no inflection point in both the vibration data coordinate system and the magnetic field gradient data coordinate system, the adjustment unit determines not to adjust the warning level.

[0061] In some embodiments of the present application, when the adjustment unit determines to adjust the warning level, it includes: when there is an inflection point in one of the vibration data coordinate system and the magnetic field gradient data coordinate system, the warning level is increased by one level. When there are inflection points in both the vibration data coordinate system and the magnetic field gradient data coordinate system, the warning level is increased by two levels.

[0062] Specifically, when the vibration data changes over time, if the rate of change of the data changes dramatically at certain moments (such as a sudden increase or decrease in vibration), an inflection point will be formed. This change is due to an abnormality in the pipeline, such as a sudden increase at the beginning of the drilling operation, and a sudden decrease in vibration at the completion of the drilling operation. The magnetic field gradient data usually shows a stable change under normal conditions. If a sudden change occurs (such as a sudden drop in the magnetic field due to drilling), an inflection point in the magnetic field gradient data will be formed.

[0063] It is understandable that by combining the inflection point detection of vibration data and magnetic field gradient data, it is possible to respond promptly at the early stage of abnormal events. If only one coordinate system has an inflection point, the warning level rises moderately; when inflection points appear in both coordinate systems at the same time, the warning level rises rapidly, indicating the severity of the abnormality. Grading adjustment can better reflect the degree of abnormality and ensure that the corresponding response measures are in place in a timely manner. Using real-time data and multi-dimensional analysis, it has a high dynamic adaptability to the response to pipeline abnormalities. The adjustment of warning levels in different situations makes monitoring more flexible. Through the dynamic adjustment of inflection point detection and warning levels, false alarms caused by environmental changes or data noise can be reduced, and real abnormal events can be accurately identified, reducing the probability of missed reports.

[0064] In the above embodiment, by integrating vibration data and magnetic field gradient data, combining neural network models and fuzzy algorithms, multi-dimensional sensor data can be collected and analyzed in real time. When an abnormal signal is detected, the continuous monitoring mode is dynamically turned on, and the continuous monitoring duration is adaptively calculated through the fuzzy algorithm. It effectively reduces manual intervention and improves the degree of automation of monitoring. By establishing a comparison between historical data and the current feature set, combined with neural networks for early warning level judgment, oil theft behavior can be accurately identified according to different detection results. The use of coordinate system analysis methods, combined with time series data and data inflection point judgment, further improves the accuracy and flexibility of anomaly detection, avoids the limitations of static thresholds and external interference in traditional detection methods, and can be adjusted according to changes in complex scenarios to ensure stable and reliable detection results under various environmental conditions. Effectively improve the accuracy and response speed of oil and gas pipeline monitoring.

[0065] In another preferred embodiment based on the above embodiment, refer to Figure 2 As shown, this embodiment provides a method for identifying perforated oil stealing branch pipe data based on a neural algorithm, which is applied to the above-mentioned perforated oil stealing branch pipe data identification system based on a neural algorithm, including: S100: Collecting sensor data in the area to be monitored, where the sensor data includes vibration data and magnetic field gradient data, and determining whether to start the continuous monitoring mode based on the sensor data; S200: when it is determined that the continuous monitoring mode is turned on, a vibration excess value is obtained according to the vibration data and the vibration data threshold, a magnetic field gradient decrease value is obtained according to the magnetic field gradient data and the magnetic field gradient threshold, the vibration excess value and the magnetic field gradient decrease value are input into a preset fuzzy algorithm, and the continuous monitoring time of the continuous monitoring mode is obtained according to the output result; S300: continuously record sensor data within the continuous monitoring period, establish a sensor data set, extract maximum vibration data and minimum magnetic field gradient data and establish a current feature set, compare the current feature set with the historical warning data set, and determine the warning level in combination with the neural network model; the historical warning data set includes several historical feature sets and several historical warning levels, each historical feature set corresponds to a historical warning level, and the historical feature set includes historical maximum vibration data and historical minimum magnetic field gradient data; S400: Establish a vibration data coordinate system and a magnetic field gradient data coordinate system, mark the vibration data in the sensor data set into the vibration data coordinate system in time sequence, mark the magnetic field gradient data in the sensor data set into the magnetic field gradient data coordinate system in time sequence, obtain the data inflection point in the coordinate system according to the vibration data coordinate system and the magnetic field gradient data coordinate system after the marked data, determine whether to adjust the warning level according to the data inflection point, and when it is determined that the warning level is to be adjusted, issue a warning at the adjusted warning level; S500: Storing the current feature set and warning level.

[0066] It is understandable that by integrating vibration data and magnetic field gradient data, combining neural network models and fuzzy algorithms, it is possible to collect and analyze multi-dimensional sensor data in real time. When an abnormal signal is detected, the continuous monitoring mode is dynamically turned on, and the continuous monitoring duration is adaptively calculated through the fuzzy algorithm. It effectively reduces manual intervention and improves the degree of automation of monitoring. By establishing a comparison between historical data and the current feature set, combined with neural networks for early warning level judgment, it is possible to accurately identify oil theft behavior based on different test results. The use of coordinate system analysis methods, combined with time series data and data inflection point judgment, further improves the accuracy and flexibility of anomaly detection, avoids the limitations of static thresholds and external interference in traditional detection methods, and can be adjusted according to changes in complex scenarios to ensure stable and reliable detection results under various environmental conditions. It effectively improves the accuracy and response speed of oil and gas pipeline monitoring.

[0067] Those skilled in the art will appreciate that the embodiments of the present application may be provided as methods, systems or computer program products. Therefore, the present application may take the form of a complete hardware embodiment, a complete software embodiment, or an embodiment combining software and hardware. Moreover, the present application may take the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer-usable program codes.

[0068] The present application is described with reference to the flowcharts and / or block diagrams of the methods, devices (systems) and computer program products according to the embodiments of the present application. It should be understood that each process and / or box in the flowchart and / or block diagram, as well as the combination of processes and / or boxes in the flowchart and / or block diagram, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded processor or other programmable data processing device to produce a machine, so that the instructions executed by the processor of the computer or other programmable data processing device generate instructions for implementing the processes in the flowchart and / or block diagram. Figure 1 A process or multiple processes and / or boxes Figure 1 A device that provides the functions specified in a block or multiple blocks.

[0069] These computer program instructions may also be stored in a computer-readable memory capable of directing a computer or other programmable data processing device to operate in a specific manner, so that the instructions stored in the computer-readable memory produce an article of manufacture comprising an instruction device, which implements the process Figure 1 A process or multiple processes and / or boxes Figure 1 A function specified in one or more boxes.

[0070] These computer program instructions can also be loaded onto a computer or other programmable data processing device so that a series of operating steps are executed on the computer or other programmable device to produce a computer-implemented process, thereby providing instructions for implementing the process. Figure 1 A process or multiple processes and / or boxes Figure 1 The steps for the functions specified in one or more boxes.

[0071] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention rather than to limit it. Although the present invention has been described in detail with reference to the above embodiments, ordinary technicians in the relevant field should understand that the specific implementation methods of the present invention can still be modified or replaced by equivalents. Any modification or equivalent replacement that does not depart from the spirit and scope of the present invention should be covered within the scope of protection of the claims of the present invention.

Claims

1. A data identification system for drilling and stealing oil branch pipes based on neural algorithm, characterized in that: include: A collection unit is configured to collect sensor data in the area to be monitored, wherein the sensor data includes vibration data and magnetic field gradient data, and determine whether to start the continuous monitoring mode according to the sensor data; a processing unit configured to, when the acquisition unit determines to start the continuous monitoring mode, obtain a vibration excess value according to the vibration data and the vibration data threshold, obtain a magnetic field gradient decrease value according to the magnetic field gradient data and the magnetic field gradient threshold, input the vibration excess value and the magnetic field gradient decrease value into a preset fuzzy algorithm, and obtain a continuous monitoring duration of the continuous monitoring mode according to an output result; A judgment unit is configured to continuously record sensor data within the continuous monitoring period, establish a sensor data set, extract maximum vibration data and minimum magnetic field gradient data and establish a current feature set, compare the current feature set with a historical warning data set, and determine a warning level in combination with a neural network model; the historical warning data set includes a plurality of historical feature sets and a plurality of historical warning levels, each of the historical feature sets corresponds to a historical warning level, and the historical feature sets include historical maximum vibration data and historical minimum magnetic field gradient data; an adjustment unit configured to establish a vibration data coordinate system and a magnetic field gradient data coordinate system, mark the vibration data in the sensor data set into the vibration data coordinate system in time sequence, mark the magnetic field gradient data in the sensor data set into the magnetic field gradient data coordinate system in time sequence, obtain a data inflection point in the coordinate system according to the vibration data coordinate system and the magnetic field gradient data coordinate system after the marked data, determine whether to adjust the warning level according to the data inflection point, and when it is determined that the warning level is to be adjusted, issue a warning at the adjusted warning level; The storage unit is configured to store the current feature set and warning level.

2. The neural algorithm-based perforated oil theft branch pipe data identification system according to claim 1 is characterized in that: When the acquisition unit determines whether to start the continuous monitoring mode according to the sensor data, it includes: The acquisition unit compares the vibration data with the vibration data threshold, and compares the magnetic field gradient data with the magnetic field gradient data threshold, and determines whether to start the continuous monitoring mode according to the comparison results; When the vibration data is greater than a vibration data threshold and the magnetic field gradient data is less than a magnetic field gradient data threshold, determining to start the continuous monitoring mode; When the vibration data is less than or equal to the vibration data threshold or the magnetic field gradient data is greater than or equal to the magnetic field gradient data threshold, it is determined that the continuous monitoring mode is not turned on, and a Gaussian mixture distribution is used to smooth the signal.

3. The neural algorithm-based perforated oil theft branch pipe data identification system according to claim 2 is characterized in that: When the acquisition unit determines not to start the continuous monitoring mode and uses Gaussian mixture distribution to smooth the signal, it includes: When the vibration data is greater than the vibration data threshold and the magnetic field gradient data is greater than or equal to the magnetic field gradient data threshold, the vibration data is recorded as abnormal data; When the vibration data is less than or equal to the vibration data threshold and the magnetic field gradient data is less than the magnetic field gradient data threshold, recording the magnetic field gradient data as abnormal data; Retrieving data of a previous time period of a data type corresponding to the abnormal data, processing the retrieved data using a Gaussian mixture distribution to determine smoothed data, and changing the abnormal data at the current time period to the smoothed data; When Gaussian mixture distribution is used to process the retrieved data to determine the smoothed data, the kernel density function expression is: ； Where n represents the number of data in the retrieved data, h represents the smoothing bandwidth, Indicates retrieving the i-th data in the data, and x indicates the value of a certain frequency point to be estimated; The data value corresponding to the highest frequency of the kernel density is used as the smoothed data.

4. The neural algorithm-based perforated oil theft branch pipe data identification system according to claim 1 is characterized in that: When the processing unit inputs the vibration excess value and the magnetic field gradient decrease value into a preset fuzzy algorithm and obtains the continuous monitoring time of the continuous monitoring mode according to the output result, it includes: Define the vibration excess value and the magnetic field gradient drop value as input variables, and define the continuous monitoring duration as the output variable; Dividing the vibration excess value into a plurality of vibration excess value fuzzy sets according to the magnitude of the vibration excess value, and dividing the magnetic field gradient decrease value into a plurality of magnetic field gradient decrease value fuzzy sets according to the magnitude of the magnetic field gradient decrease value, and determining a fuzzy rule base according to the vibration excess value fuzzy sets and the magnetic field gradient decrease value fuzzy sets; The current vibration excess value and magnetic field gradient decrease value are compared with the vibration excess value fuzzy set and the magnetic field gradient decrease value fuzzy set respectively, and the vibration excess value membership and the magnetic field gradient decrease value membership are determined based on the triangular membership function; Calculating the output value of each rule in the fuzzy rule base according to the vibration excess value membership and the magnetic field gradient decrease value membership; Defuzzify the output value of each rule by weighted summing the following formula, and obtain a total output value, where the total output value is the continuous monitoring duration; ； Among them, E represents the continuous monitoring duration, E0 represents the continuous monitoring duration reference value, Qi represents the output weight of the i-th rule, and Si represents the output value of the i-th rule.

5. The neural algorithm-based perforated oil theft branch pipe data identification system according to claim 1 is characterized in that: When the judgment unit compares the current feature set with the historical warning data set and determines the warning level in combination with the neural network model, it includes: The judgment unit calculates the similarity between the current feature set and each historical feature set respectively; When there is data in all the historical feature sets whose similarity with the current feature set is greater than a similarity threshold, determining the warning level according to the historical warning levels corresponding to the historical feature sets whose similarity is greater than the similarity threshold; When the similarities between all the historical feature sets and the current feature set are less than a similarity threshold, the neural network model is established according to the historical warning data group, and the current feature set is input into the neural network model to obtain the warning level.

6. The neural algorithm-based perforated oil theft branch pipe data identification system according to claim 5 is characterized in that: When the judgment unit determines the warning level according to the historical warning level corresponding to the historical feature set whose similarity is greater than the similarity threshold, it includes: When there is only one data in the historical feature set whose similarity with the current feature set is greater than the similarity threshold, the historical warning level corresponding to the historical feature set is used as the warning level; When there is not unique data in the historical feature set whose similarity with the current feature set is greater than a similarity threshold, the highest historical warning level corresponding to the historical feature set is used as the warning level.

7. The neural algorithm-based perforated oil theft branch pipe data identification system according to claim 5 is characterized in that: The judgment unit establishes the neural network model according to the historical warning data group, inputs the current feature set into the neural network model, and obtains the warning level, including: Sampling the historical warning data group according to a preset ratio to obtain a training subset and a test subset; Acquire a pre-selected convolutional neural model, iteratively train the convolutional neural model according to the training subset, and evaluate the iteratively trained convolutional neural model according to the test subset to obtain the neural network model; The current feature set is input into the neural network model to obtain the warning level.

8. The neural algorithm-based perforated oil theft branch pipe data identification system according to claim 7 is characterized in that: When the adjustment unit determines whether to adjust the warning level according to the data inflection point, it includes: When there is an inflection point in the vibration data coordinate system or the magnetic field gradient data coordinate system, the adjustment unit determines to adjust the warning level; When there is no inflection point in either the vibration data coordinate system or the magnetic field gradient data coordinate system, the adjustment unit determines not to adjust the warning level.

9. The neural algorithm-based perforated oil theft branch pipe data identification system according to claim 8 is characterized in that: When the adjustment unit determines to adjust the warning level, it includes: When there is an inflection point in one of the vibration data coordinate system and the magnetic field gradient data coordinate system, the warning level is raised by one level; When both the vibration data coordinate system and the magnetic field gradient data coordinate system have inflection points, the warning level is increased by two levels.

10. A method for identifying oil-stealing branch pipe data based on a neural algorithm, applied to the system for identifying oil-stealing branch pipe data based on a neural algorithm as claimed in any one of claims 1 to 9, characterized in that: include: Collecting sensor data in the area to be monitored, the sensor data including vibration data and magnetic field gradient data, and determining whether to start the continuous monitoring mode according to the sensor data; When it is determined that the continuous monitoring mode is turned on, a vibration excess value is obtained according to the vibration data and the vibration data threshold, a magnetic field gradient decrease value is obtained according to the magnetic field gradient data and the magnetic field gradient threshold, the vibration excess value and the magnetic field gradient decrease value are input into a preset fuzzy algorithm, and the continuous monitoring duration of the continuous monitoring mode is obtained according to the output result; Continuously record the sensor data within the continuous monitoring period, establish a sensor data set, extract the maximum vibration data and the minimum magnetic field gradient data and establish a current feature set, compare the current feature set with the historical warning data group, and determine the warning level in combination with the neural network model; the historical warning data group includes a plurality of historical feature sets and a plurality of historical warning levels, each of the historical feature sets corresponds to a historical warning level, and the historical feature sets include historical maximum vibration data and historical minimum magnetic field gradient data; Establishing a vibration data coordinate system and a magnetic field gradient data coordinate system, marking the vibration data in the sensor data set into the vibration data coordinate system in time sequence, marking the magnetic field gradient data in the sensor data set into the magnetic field gradient data coordinate system in time sequence, obtaining the data inflection point in the coordinate system according to the vibration data coordinate system and the magnetic field gradient data coordinate system after the marked data, judging whether to adjust the warning level according to the data inflection point, and issuing a warning at the adjusted warning level when it is judged that the warning level is to be adjusted; The current feature set and warning level are stored.

Images