Multi-source data fusion metering method and system of skid-mounted three-phase separation equipment

Abstract

The invention discloses a multi-source data fusion metering method and system for skid-mounted three-phase separation equipment, and particularly relates to the technical field of data analysis, and the method is realized through the following steps: 1, based on a preset data source type, employing a corresponding sensor to collect equipment operation data, and forming a collection data set; 2, dividing each data source data set into sub-data sets according to a fixed time length; 3, performing feature extraction on the sub-data sets, analyzing quality change and fluctuation conditions, and generating feature data; 4, importing the feature data into a convolutional neural network model for classification, and judging whether the sub-data set is high-quality data or not; step 5, according to a classification result, determining whether to send out an early warning signal to guarantee the accuracy of data fusion; according to the method, division and feature extraction are carried out on the multiple data sources collected by the equipment, high-quality and low-quality data can be effectively recognized and classified, credible data sources can be screened out, and the accuracy of multi-source data fusion is improved.

Description

Technical Field

[0001] The present invention relates to the technical field of data analysis, and more specifically, to a multi-source data fusion metering method and system for skid-mounted three-phase separation equipment. Background Art

[0002] With the rapid development of the oil and gas industry, skid-mounted three-phase separation equipment is widely used in the oil, natural gas and other industries to separate the oil, water and gas in the mixed fluid. In order to accurately measure the operation and separation effect of the three-phase separation equipment, a multi-source data fusion measurement method is adopted. This method monitors the working status of the separation equipment in real time by integrating sensor data from different sources, such as flow sensors, temperature sensors, pressure sensors, etc. Through data fusion technology, the measurement data from multiple sensors are weighted to improve the measurement accuracy and reduce the influence of single sensor errors.

[0003] Multi-source data fusion measurement can not only optimize the coordinated use of various sensors, but also automatically adjust measurement parameters under different operating conditions to improve the adaptability of the system. Data quality is the prerequisite for ensuring data fusion quality, but the existing technology has defects in multi-source data quality control, lacks effective monitoring and early warning mechanisms, and is difficult to actively manage data quality. When the data quality is poor, low-quality data sets are not conducive to subsequent multi-source data fusion measurement, thereby affecting the accuracy and stability of the overall system. Therefore, a multi-source data fusion measurement method and system for skid-mounted three-phase separation equipment is proposed to solve the above problems. Summary of the invention

[0004] To achieve the above object, the present invention provides the following technical solutions: The multi-source data fusion measurement method of the skid-mounted three-phase separation equipment comprises the following steps: Step 1: Based on multiple preset data source types, corresponding sensor devices are used to collect the operation data of the skid-mounted three-phase separation device to obtain a collection data set corresponding to each data source type; Step 2: Divide the collected data set corresponding to each data source type according to a fixed time length, and record each divided part as a sub-data set for separate analysis; Step 3: Extract features from each sub-dataset, and then analyze the changes in the sub-dataset to provide feature data for subsequent classification; Step 4: For each sub-dataset, the feature data after change analysis is imported into the pre-trained convolutional neural network model to obtain different classification results and decide whether to participate in multi-source data fusion measurement; Step 5: Based on the classification results of the sub-datasets output by all data source types in the convolutional neural network model, decide whether to issue a warning signal.

[0005] In a preferred embodiment, the collected data set corresponding to each data source type is divided according to a fixed time length, and each of the divided parts is recorded as a sub-data set, which means: According to a fixed time length, the entire collected data sequence of each data source type is divided into multiple continuous data segments, and a unique identifier is assigned to each divided data segment, which is recorded as a sub-dataset.

[0006] In a preferred embodiment, performing feature extraction on each sub-data set first refers to: Extract the quality change characteristics of the data source corresponding to the sub-dataset and the data change direction characteristics corresponding to the sub-dataset.

[0007] In a preferred embodiment, analyzing the changes in the sub-datasets refers to: Based on the quality change characteristics of the data source, the source quality changes of the sub-dataset are analyzed to generate a quality change index. Based on the data change direction characteristics of the data source, the fluctuation amplitude and fluctuation direction changes of the sub-dataset are analyzed to generate a fluctuation change index.

[0008] In a preferred embodiment, the logic for obtaining the quality change index is: There are multiple quality dimensions i by default. For each quality dimension i, the following normalization method is used: ; is the standardized quality score of quality dimension i at time t, is the original quality score of quality dimension i at time t, is the minimum score of quality dimension i, is the maximum score of quality dimension i, is the average of the original quality scores of quality dimension i, , All are preset non-zero adjustment coefficients; Introducing sliding time window W and time decay factor , using the weighted average method to combine the quality scores of all dimensions and calculate the overall quality score of the data source: ; is the preset proportional coefficient corresponding to mass dimension i, n is the total number of dimensions of mass dimension i, is the standardized quality score of quality dimension i at time k, is the overall quality score of the data source at time t; Introducing a volatility penalty : ; is the average quality score after standardization of quality dimension i; The formula for the total quality score is modified by combining the penalty term: ; is the overall quality score of the corrected data source at time t, is the preset control coefficient corresponding to the volatility penalty term; Finally, the average of the overall quality scores of all corrected data sources within a fixed time length of the sub-dataset is calculated to obtain the quality change index.

[0009] In a preferred embodiment, the logic for obtaining the fluctuation change index is: Within a fixed time length of the sub-dataset, the absolute difference between the data of every two adjacent time points is calculated and the average value is taken to represent the fluctuation range of the sub-dataset within the fixed time length; Within a fixed time length of the sub-dataset, the data change rate is further calculated, that is, the difference between the data changes at adjacent time points, and the average of the absolute values ​​is taken to indicate the severity of the fluctuation; When the direction of change at adjacent time points reverses, it is counted as one direction change, and the total number of direction changes is divided by the fixed time length to obtain the direction change frequency; By calculating the standard deviation of the data changes at adjacent time points within a fixed time length, the discrete degree of data changes is measured, and the fluctuation stability used to evaluate the stability of data fluctuations is obtained; Multiply the fluctuation amplitude, fluctuation severity, direction change frequency, and fluctuation smoothness by the preset weight coefficients respectively, and then sum them up to obtain the fluctuation change index.

[0010] In a preferred embodiment, importing the feature data after the change analysis into a pre-trained convolutional neural network model to obtain different classification results and determine whether to participate in multi-source data fusion measurement refers to: The quality change index and fluctuation change index corresponding to each sub-dataset are obtained and used as the input data of the convolutional neural network model together to obtain a classification result of 1 or 0. When the classification result is 1, it indicates that the sub-dataset is a high-quality data set, and when the classification result is 0, it indicates that the sub-dataset is a low-quality data set; When the classification results of each sub-data set corresponding to multiple data source types within the same fixed time length are all 1, the fixed time length is marked as the multi-source data fusion time period and participates in the multi-source data fusion measurement.

[0011] In a preferred embodiment, the logic of issuing the warning signal is: The number of time periods marked as multi-source data fusion time periods is divided by the total number of all time periods to obtain a ratio value 1, and the ratio value 1 is compared with a preset warning threshold 1. If the ratio value 1 is less than or equal to the preset warning threshold 1, a type 1 warning signal is generated; Get the total number of sub-datasets corresponding to a data source type with a classification result of 0, and then divide it by the total number of time periods to get proportion value 2, compare proportion value 2 with the preset warning threshold 2, and if proportion value 2 is greater than or equal to the preset warning threshold 2, a second-class warning signal is generated.

[0012] In a preferred embodiment, a multi-source data fusion metering system for a skid-mounted three-phase separation device includes: The data acquisition module uses corresponding sensor devices to collect the operation data of the skid-mounted three-phase separation device based on multiple preset data source types, and obtains a collection data set corresponding to each data source type; The time division module divides the collected data set corresponding to each data source type according to a fixed time length, and records each divided part as a sub-data set for separate analysis; The feature extraction module first extracts features from each sub-dataset, then analyzes the changes in the sub-dataset to provide feature data for subsequent classification; The data classification module imports the feature data after change analysis into the pre-trained convolutional neural network model for each sub-data set to obtain different classification results and decide whether to participate in multi-source data fusion measurement; The early warning module decides whether to issue an early warning signal based on the classification results output by the sub-datasets of all data source types in the convolutional neural network model.

[0013] Technical effects and advantages of the present invention: The present invention can effectively identify and classify high-quality and low-quality data by dividing and extracting features from multiple data sources collected by the device. Using convolutional neural networks (CNNs) to classify feature data can automatically screen out reliable data sources to prevent low-quality data from affecting subsequent analysis and decision-making. This ensures the complementarity of different data sources in subsequent multi-source data fusion, fills the gap in a single data source, and improves the accuracy and reliability of multi-source data fusion.

[0014] After the data is divided into sub-data sets, the present invention can analyze each time period independently to timely discover data fluctuations, trends and anomalies, so as to capture potential anomalies or equipment failures, ensure continuous monitoring during equipment operation, and avoid ignoring local short-term anomalies. By introducing the quality change index and the fluctuation change index, quantitative analysis of data changes is carried out, which helps to timely identify potential problems in equipment operation.

[0015] The present invention generates warning signals by setting thresholds. When the data quality is unstable or a data source is abnormal, the system can automatically send out a first or second type of warning signal. These warning signals can help operation and maintenance personnel to promptly identify problems in equipment operation and carry out targeted maintenance to reduce the probability of equipment failure. The first type of warning signal reflects the decline or instability of the overall equipment data quality, indicating that the equipment may have a minor fault or incomplete data collection; the second type of warning signal focuses on the quality problem of a single data source, and can locate the specific sensor equipment with abnormalities, helping operation and maintenance personnel to quickly troubleshoot faults.

[0016] The present invention can perform parallel processing by dividing a large-scale data set into multiple sub-data sets, avoiding the computational and storage burden of processing the entire data set. After the sub-data set is divided, each data segment can be analyzed independently, which significantly improves the computational efficiency. After the data is divided by time period, the storage space is also reasonably allocated, which helps to reduce the pressure of big data storage and supports archiving and backup based on time period.

[0017] The present invention can effectively improve the accuracy of multi-source data fusion through automated data classification and fusion evaluation. When multiple data sources provide high-quality data in the same time period, the system can automatically mark it as a multi-source data fusion time period, providing reliable data input for subsequent multi-source data fusion. Fusion of data collected by different sensors can overcome the noise and deviation that may exist in a single data source and enhance the stability of equipment operation.

[0018] The present invention provides intelligent data analysis and automatic early warning functions, which are helpful for continuous monitoring, performance evaluation, fault prediction and maintenance decision-making of equipment. Through real-time feedback of data quality and equipment status, potential problems can be discovered in advance, downtime and maintenance costs can be reduced, the operating efficiency of equipment can be improved, and potential failures and unstable factors of equipment can be identified in advance, which helps to optimize the service life and performance of equipment, avoid excessive maintenance or neglect of certain potential problems, and ensure that the equipment is always in the best working condition. BRIEF DESCRIPTION OF THE DRAWINGS

[0019] In order to facilitate understanding by those skilled in the art, the present invention is further described below in conjunction with the accompanying drawings; Figure 1It is a schematic diagram of the multi-source data fusion metering method of the skid-mounted three-phase separation equipment in the present invention.

[0020] Figure 2 It is a schematic diagram of the multi-source data fusion metering system of the skid-mounted three-phase separation equipment in the present invention. DETAILED DESCRIPTION

[0021] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. 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 creative work are within the scope of protection of the present invention.

[0022] Reference Figure 1 - Figure 2 The following embodiments are obtained:

[0023] Example 1: Skid-mounted three-phase separation equipment is widely used in the oil and gas industries to separate the oil, water and gas phases in the mixed fluid. In order to accurately measure the operation and separation effect of the three-phase separation equipment, a multi-source data fusion measurement method is adopted. This method monitors the working status of the separation equipment in real time by integrating sensor data from different sources, such as flow sensors, temperature sensors, pressure sensors, etc. Through data fusion technology, the measurement data from multiple sensors are weighted and calculated to improve the measurement accuracy and reduce the influence of single sensor errors.

[0024] Multi-source data fusion measurement can not only optimize the coordinated use of various sensors, but also automatically adjust measurement parameters under different operating conditions to improve the adaptability of the system. For example, when the separation effect is poor, the fusion data can be used to determine whether the equipment operating parameters need to be adjusted, and equipment failures or abnormal operations can be discovered in time to ensure separation efficiency. It can also be used to conduct predictive analysis based on historical data, evaluate the long-term operation of the equipment, provide data support for maintenance and overhaul, and reduce equipment downtime.

[0025] Sensor data refers to physical quantity data collected by various sensors, such as flow, temperature, pressure, etc., which are used to monitor equipment status and environmental changes in real time. It should be noted that data quality is the prerequisite for ensuring data fusion quality, but the existing technology has defects in multi-source data quality control and lacks effective monitoring and early warning mechanisms, making it difficult to actively manage data quality. When data quality is poor, it is difficult for existing technologies to adjust sensors or data collection strategies through rapid feedback. Low-quality data sets are not conducive to subsequent multi-source data fusion measurement, thereby affecting the accuracy and stability of the overall system.

[0026] Therefore, the present invention proposes a multi-source data fusion measurement method for a skid-mounted three-phase separation device, comprising the following steps: Step 1: Based on the preset multiple data source types, use the corresponding sensor devices to collect the operating data of the skid-mounted three-phase separation equipment to obtain the collected data set corresponding to each data source type; collect data through the preset data source types (such as temperature, pressure, flow, etc.) to lay the foundation for subsequent data analysis and fusion. Use different sensors to correspond to different data sources, which can comprehensively monitor the operating status of the three-phase separation equipment and ensure data diversity, which is helpful to accurately judge the equipment performance, operation quality and potential abnormalities after multi-source data fusion.

[0027] Step 2: Divide the collected data set corresponding to each data source type according to a fixed time length, and record each divided part as a sub-data set for separate analysis; divide the continuous data collection sequence into sub-data sets of fixed time length to facilitate segmented analysis and processing. Through the division of sub-data sets, the fluctuations, changing trends and abnormal situations of data sources in different time periods can be effectively identified, ensuring the timeliness of data analysis and the accuracy of local features, and providing favorable conditions for subsequent feature extraction and classification.

[0028] Step 3: Perform feature extraction on each sub-dataset first, and then analyze the changes in the sub-datasets to provide feature data for subsequent classification; through feature extraction, the original data is converted into feature data with practical significance, such as quality change features and data change direction features. By analyzing the changes in the sub-datasets, the degree of data fluctuation, the frequency of direction changes and quality changes can be quantified to form a quality change index and a fluctuation change index. Feature data is the core basis for subsequent classification and decision-making, and can reflect the stability and reliability of the data source.

[0029] Step 4: For each sub-dataset, the feature data after change analysis is imported into the pre-trained convolutional neural network model to obtain different classification results and decide whether to participate in multi-source data fusion measurement; the convolutional neural network (CNN) model is used to classify the feature data of the sub-dataset to identify the quality level of the data source (high quality or low quality). A classification result of 1 indicates that the sub-dataset is of high quality and has high data reliability, which is suitable for data fusion measurement; a classification result of 0 indicates that the data quality is poor and is not suitable for fusion. Through automated classification, it is ensured that the sub-datasets participating in multi-source data fusion have high quality and reduce the impact of errors and unstable factors on the fusion results.

[0030] Step 5: Based on the classification results of the sub-datasets output of all data source types in the convolutional neural network model, decide whether to issue a warning signal. Evaluate the data fusion results of different time periods. If all sub-datasets of data source types are of high quality (classification result is 1) in the same time period, it is marked as a multi-source data fusion time period. When the proportion of the marked multi-source data fusion time period is less than or equal to the preset threshold value one, the system issues a first-class warning signal, indicating that there are hidden dangers in the overall data quality of the skid-mounted three-phase separation equipment. If the proportion of all sub-datasets of a single data source with low quality (classification result is 0) is too high, a second-class warning signal is issued, indicating that the data source, i.e., the sampling equipment, may have a fault or abnormality. The warning signal provides a basis for the operation and maintenance of the equipment and fault diagnosis, which helps to discover problems in advance and ensure the safety and stability of the skid-mounted three-phase separation equipment and sampling equipment.

[0031] Dividing the collected data set corresponding to each data source type according to a fixed time length and recording each divided part as a sub-data set means: dividing the entire collected data sequence of each data source type into multiple continuous data segments according to a fixed time length, assigning a unique identifier to each divided data segment, and recording it as a sub-data set.

[0032] Large-scale datasets may contain a large amount of raw data, which may be time-consuming to process and analyze. By dividing the data into sub-datasets at fixed time lengths, the dataset can be divided into multiple smaller parts, each of which can be processed independently. This can avoid the computational and memory burden of processing the entire dataset at once and improve processing efficiency.

[0033] If the data is divided into multiple sub-datasets, each sub-dataset can be processed in parallel on different computing resources, further speeding up the analysis. For time series data, after being divided by a fixed time length, each sub-dataset has the same time span and timestamp. This can maintain time consistency and facilitate tasks such as time series analysis and trend analysis. If each sub-dataset corresponds to a time window, this division helps to perform sliding window analysis and identify patterns or outliers in the time series that change over time. Each sub-dataset is assigned a unique identifier for easy traceability and management. The identifier allows you to quickly find data for a specific time period, reducing the complexity of data management. After being divided by time period, data in different time periods can be clearly divided, which helps with subsequent data labeling and classification tasks.

[0034] The entire data set may be very large. After dividing by time period, the storage volume of each sub-data set is reduced, which makes it easier to store data in a dispersed manner and reduce the pressure on storage devices. Especially when the storage system is limited, data can be archived according to time periods to facilitate backup and recovery by time period. This can also better track and manage the life cycle of data. Some analysis tasks may only need to process data in a specific time period. After dividing by time period, data in a specific time period can be easily selected and extracted for analysis.

[0035] Performing feature extraction on each sub-dataset first means: extracting the quality change features of the data source corresponding to the sub-dataset and the data change direction features corresponding to the sub-dataset. Analyzing the changes in the sub-dataset means: analyzing the source quality changes of the sub-dataset based on the quality change features of the data source to generate a quality change index, and analyzing the fluctuation amplitude and fluctuation direction of the sub-dataset based on the data change direction features of the data source to generate a fluctuation change index.

[0036] Feature extraction is a part of data preprocessing, and its main purpose is to extract valuable information from raw data for analysis tasks. These features can usually reveal the inherent patterns of the data and help subsequent modeling and analysis.

[0037] Quality change characteristics: refers to the law of how the quality of a data source changes over time or other factors. For example, the accuracy, completeness, and reliability of a data source may change at different time points or under different conditions, affecting the effectiveness of subsequent analysis.

[0038] Data change direction characteristics: refers to the direction of fluctuation or trend change of the data itself, such as upward trend, downward trend, and stability. By capturing the direction of data change, we can help understand the dynamic mechanism behind the data and how to deal with such changes.

[0039] By extracting these features from each sub-dataset, analysts can identify anomalies or trends in the data, thereby better understanding the reasons behind data changes and providing effective input for subsequent data analysis and model building.

[0040] The analysis of changes in sub-datasets is mainly done by quantitatively evaluating the features and generating specific indexes and indicators to help understand the performance of data in time series or under different conditions. The following is the specific analysis process: Quality change index: Based on the quality change characteristics, analyze whether the quality of the data has changed and quantify the degree of this change. For example, some data sources may have quality fluctuations due to equipment failure, inconsistent collection methods, etc. By generating a quality change index, these changes can be quantified and tracked, thereby providing guidance for subsequent data cleaning or repair.

[0041] Fluctuation change index: Based on the characteristics of data change direction, the amplitude and direction of data fluctuation are analyzed to generate a fluctuation change index. The fluctuation amplitude can reflect the stability of the data, and the fluctuation direction can reveal the change of data trend. For example, if the fluctuation amplitude of a sub-data set is too large, it may mean that the data source is relatively unstable, or the data will fluctuate greatly in the short term, affecting the subsequent prediction or decision of multi-source data fusion measurement.

[0042] The logic for obtaining the quality change index is: There are multiple quality dimensions i by default. For each quality dimension i, the following normalization method is used: ; Based on the standardization of maximum and minimum values, the adjustment coefficient is introduced , To enhance the standardization effect so that it can adapt to the complex situation of different data quality distribution; is the standardized quality score of quality dimension i at time t, is the original quality score of quality dimension i at time t, is the minimum score of quality dimension i, is the maximum score of quality dimension i, is the average of the raw quality scores of quality dimension i. The average provides the overall data center trend, so that the normalization formula can take into account the relative volatility. , All are preset non-zero adjustment coefficients; the original quality scores are standardized to eliminate the dimensional differences between different quality dimensions, so that all quality dimensions can be compared and calculated within the same numerical range.

[0043] Introducing sliding time window W and time decay factor , weight the quality scores within the time window, give higher weight to the data at the most recent moment, gradually reduce the influence of the data far from the current moment, emphasize the importance of real-time data, reduce the weight of outdated data, and increase sensitivity to the current state. Use the weighted average method to combine the quality scores of all dimensions and calculate the overall quality score of the data source: ; The preset proportional coefficient corresponding to quality dimension i is used to assign weights to different quality dimensions, reflecting the importance of each dimension to the overall quality score. n is the total number of dimensions of quality dimension i. is the standardized quality score of quality dimension i at time k, is the overall quality score of the data source at time t; Introducing a volatility penalty : ; is the average quality score of quality dimension i after standardization. The standard deviation of the quality score after standardization reflects the severity of data fluctuation. In the calculation of the overall quality score, the quality dimension with large fluctuation is penalized to reduce the impact of abnormal fluctuations on the results.

[0044] The formula for the total quality score is modified by combining the penalty term: ; is the overall quality score of the corrected data source at time t; is the preset control coefficient corresponding to the volatility penalty term;;Control coefficient of volatility penalty term Control the intensity of the volatility penalty term and affect the intensity of the penalty for volatile data. Finally, calculate the average of the overall quality scores of all corrected data sources within a fixed time length of the sub-dataset to obtain the quality change index. After introducing the volatility penalty term, the original overall quality score is corrected to reduce the impact of high volatility data on the final result. The volatility penalty term is to measure the volatility of the data source and reduce the contribution of the quality dimension with large fluctuations to the overall quality score through the penalty mechanism, thereby improving the stability and reliability of the overall score.

[0045] The logic for obtaining the volatility change index is: Within a fixed time length of the sub-dataset, the absolute difference between the data of every two adjacent time points is calculated and the average value is taken to represent the fluctuation range of the sub-dataset within the fixed time length; Within a fixed time length of the sub-dataset, the data change rate is further calculated, that is, the difference between the data changes at adjacent time points, and the average of the absolute values ​​is taken to indicate the severity of the fluctuation; When the direction of change at adjacent time points reverses, it is counted as one direction change, and the total number of direction changes is divided by the fixed time length to obtain the direction change frequency; By calculating the standard deviation of the data changes at adjacent time points within a fixed time length, the discrete degree of data changes is measured, and the fluctuation stability used to evaluate the stability of data fluctuations is obtained; Multiply the fluctuation amplitude, fluctuation severity, direction change frequency, and fluctuation smoothness by the preset weight coefficients respectively, and then sum them up to obtain the fluctuation change index.

[0046] Volatility: Calculate the average absolute difference between each two adjacent time points within a fixed period of time. This indicator reflects the basic volatility of the data, that is, the "average change" of the data during that period of time. A larger volatility means that the data has changed more during that period of time, which may represent a more drastic fluctuation in the market or system. This indicator helps to understand the overall volatility of the data.

[0047] Volatility: Calculate the average absolute value of the difference between the changes in data at adjacent time points. This indicator is used to measure the "severity" of fluctuations, that is, whether the fluctuations are violent or rapid. If the volatility is high, it means that the data changes quickly and the fluctuations are more violent, which usually indicates that the market or system has strong short-term instability. For example, violent fluctuations in the stock market are often accompanied by higher risks.

[0048] Direction change frequency: When the direction of change at adjacent time points reverses, it is counted as a direction change, and then the frequency of direction changes per unit time is calculated. This reflects the number of reversals of data changes. A high frequency of direction changes means that the trend of data frequently reverses, which may indicate instability in market sentiment or system status. A higher frequency usually means that the market or data behavior is more volatile and difficult to predict or grasp.

[0049] Fluctuation stability: The standard deviation of the data change is calculated to measure the dispersion of the fluctuation. The larger the standard deviation, the more unstable the data fluctuation. Fluctuation stability can help evaluate the stability of the data. If the standard deviation is large, it means that the data fluctuation lacks consistency and regularity, which may mean that the operation of the system or market is not stable enough and the risk is high.

[0050] Multiply the fluctuation amplitude, fluctuation severity, direction change frequency and fluctuation smoothness by the preset weight coefficient respectively, and then perform weighted summation to finally get the fluctuation change index. The fluctuation change index is a comprehensive assessment of the above multiple fluctuation characteristics, and can provide an overall indicator of volatility over a certain period of time. The larger the value, the more severe the data fluctuation.

[0051] Importing the feature data after change analysis into the pre-trained convolutional neural network model to obtain different classification results and decide whether to participate in multi-source data fusion measurement means: The quality change index and fluctuation change index corresponding to each sub-dataset are obtained and used as the input data of the convolutional neural network model together to obtain a classification result of 1 or 0. When the classification result is 1, it indicates that the sub-dataset is a high-quality data set, and when the classification result is 0, it indicates that the sub-dataset is a low-quality data set; When the classification results of each sub-data set corresponding to multiple data source types within the same fixed time length are all 1, the fixed time length is marked as the multi-source data fusion time period and participates in the multi-source data fusion measurement.

[0052] The change characteristics of the data source are first extracted and quantified by calculating the "quality change index" and "fluctuation change index". These indices can reflect the degree of fluctuation and change trend of data quality, and are usually important indicators for measuring data reliability, stability and volatility. The quality change index may measure the changes in dimensions such as data accuracy and reliability. The volatility change index may measure the volatility or instability of data within a certain period of time. The pre-trained convolutional neural network model is used to classify the input data (i.e., the quality change index and the volatility change index), which can effectively identify the potential patterns of the data. After the data is processed by the CNN model, the output is the classification result. A classification result of 1 indicates that the quality of the sub-dataset is high, and a classification result of 0 indicates that the data quality is low.

[0053] Classification result is 1 (high-quality dataset): If a sub-dataset is judged as a high-quality dataset within a certain period of time (i.e., the classification result is 1), then the dataset meets the required quality standards and can be used as valid input data for further processing or analysis.

[0054] Classification result is 0 (low-quality dataset): If the classification result is 0, it means that the sub-dataset is considered to be of poor quality and cannot be used for effective data analysis or fusion due to problems such as data instability, missing values, and noise.

[0055] If the classification results of each sub-dataset corresponding to multiple data sources (which may be different types of data sources) are all 1 (high quality) within the same fixed time period, the data in this time period is considered to meet the conditions for multi-source data fusion. This means that in this time period, different data sources have provided high-quality data, which can be fused to improve the accuracy and reliability of the overall data.

[0056] When it is determined that the data quality of multiple data sources is high within a fixed time period, multi-source data fusion measurement can be performed. The purpose of data fusion is to make up for the shortcomings of a single data source by combining information from different data sources, thereby improving the accuracy, stability and comprehensiveness of the analysis. The significance of multi-source data fusion measurement: Improve the reliability of data processing: By fusing high-quality multi-source data, the errors, noise and shortcomings that may exist in a single data source can be reduced. Enhance the accuracy of analysis results: The information provided by different data sources may reveal the same phenomenon from different angles. Fusion of this information can help to obtain more comprehensive and accurate conclusions. Adapt to the complementarity of different types of data: Sometimes different types of data (such as sensor data, user feedback, environmental monitoring data, etc.) can complement each other and provide deeper insights after fusion. Multi-source data fusion measurement is based on the adopted and used data, and subsequent operations are prior arts well known to those in the field, so they will not be elaborated on here.

[0057] The logic of issuing the early warning signal is: Divide the number of time periods marked as multi-source data fusion time periods by the total number of time periods to obtain a ratio value of one. Compare the ratio value of one with the preset warning threshold value of one. If the ratio value of one is less than or equal to the preset warning threshold value of one, a type of warning signal is generated; a type of warning signal is reflected in multiple data sources. There are fewer time periods that can simultaneously meet high-quality standards, and the overall data quality shows signs of instability or decline. When the ratio of multi-source data fusion time periods is too low, it may indicate potential problems such as minor equipment failures, incomplete data collection, and sensor performance degradation, and the equipment status needs to be checked in time. The poor high-quality consistency between data sources indicates that some data sources may be interfered with, resulting in poor overall data fusion effects.

[0058] Get the total number of sub-data sets corresponding to a data source type with a classification result of 0, and then divide it by the total number of all time periods to get the ratio value 2. Compare the ratio value 2 with the preset warning threshold 2. If the ratio value 2 is greater than or equal to the preset warning threshold 2, a second-class warning signal is generated. The second-class warning signal evaluates the performance of a single data source. When the proportion of low-quality data from a certain data source is too high, it may indicate that the sensor device has a fault, data anomalies, or signal loss. A single data source has a lot of low-quality data, which will affect the accuracy of data fusion. Through the second-class warning signal, unreliable data sources can be identified in time for maintenance or replacement. Unlike the first-class warning signal, the second-class warning signal can locate the specific data source, which is convenient for operation and maintenance personnel to quickly find the root cause of the problem and shorten the fault diagnosis time.

[0059] By distinguishing between overall quality anomalies (first-class warning) and single data source anomalies (second-class warning), different levels of warning information are provided to facilitate hierarchical response and processing. Through the early warning mechanism, data anomalies can be discovered in a timely manner to avoid misjudgments or system failures caused by data quality degradation. By eliminating low-quality data sources or marking abnormal periods, the reliability and credibility of multi-source data fusion are guaranteed. Providing data support for daily monitoring, maintenance and fault diagnosis of equipment helps to realize intelligent management of equipment.

[0060] Embodiment 2: A multi-source data fusion metering system for a skid-mounted three-phase separation device, comprising: The data acquisition module uses corresponding sensor devices to collect the operation data of the skid-mounted three-phase separation device based on multiple preset data source types, and obtains a collection data set corresponding to each data source type; The time division module divides the collected data set corresponding to each data source type according to a fixed time length, and records each divided part as a sub-data set for separate analysis; The feature extraction module first extracts features from each sub-dataset, then analyzes the changes in the sub-dataset to provide feature data for subsequent classification; The data classification module imports the feature data after change analysis into the pre-trained convolutional neural network model for each sub-data set to obtain different classification results and decide whether to participate in multi-source data fusion measurement; The early warning module decides whether to issue an early warning signal based on the classification results output by the sub-datasets of all data source types in the convolutional neural network model.

[0061] The above formulas are all dimensionless and numerical calculations. The formula is a formula for the most recent real situation obtained by collecting a large amount of data and performing software simulation. The preset parameters in the formula are set by technicians in this field according to actual conditions.

[0062] It should be understood that in the various embodiments of the present application, the size of the serial numbers of the above-mentioned processes does not mean the order of execution. The execution order of each process should be determined by its function and internal logic, and should not constitute any limitation on the implementation process of the embodiments of the present application.

[0063] Those of ordinary skill in the art will appreciate that the units and algorithm steps of each example described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are performed in hardware or software depends on the specific application and design constraints of the technical solution. Professional and technical personnel can use different methods to implement the described functions for each specific application, but such implementation should not be considered to be beyond the scope of this application.

[0064] Those skilled in the art can clearly understand that, for the convenience and brevity of description, the specific working processes of the systems, devices and units described above can refer to the corresponding processes in the aforementioned method embodiments and will not be repeated here.

[0065] The above is only a specific implementation of the present application, but the protection scope of the present application is not limited thereto. Any person skilled in the art who is familiar with the present technical field can easily think of changes or substitutions within the technical scope disclosed in the present application, which should be included in the protection scope of the present application. Therefore, the protection scope of the present application should be based on the protection scope of the claims.

Claims

1. A multi-source data fusion measurement method for skid-mounted three-phase separation equipment, characterized in that: The following steps are involved: Step 1: Based on multiple preset data source types, corresponding sensor devices are used to collect the operation data of the skid-mounted three-phase separation device to obtain a collection data set corresponding to each data source type; Step 2: Divide the collected data set corresponding to each data source type according to a fixed time length, and record each divided part as a sub-data set for separate analysis; Step 3: Extract features from each sub-dataset, and then analyze the changes in the sub-dataset to provide feature data for subsequent classification; Step 4: For each sub-dataset, the feature data after change analysis is imported into the pre-trained convolutional neural network model to obtain different classification results and decide whether to participate in multi-source data fusion measurement; Step 5: Based on the classification results of the sub-datasets output by all data source types in the convolutional neural network model, decide whether to issue a warning signal.

2. The multi-source data fusion measurement method for skid-mounted three-phase separation equipment according to claim 1 is characterized in that: Divide the collected data set corresponding to each data source type according to a fixed time length, and record each divided part as a sub-data set, which means: According to a fixed time length, the entire collected data sequence of each data source type is divided into multiple continuous data segments, and a unique identifier is assigned to each divided data segment, which is recorded as a sub-dataset.

3. The multi-source data fusion measurement method for skid-mounted three-phase separation equipment according to claim 2 is characterized in that: Performing feature extraction on each sub-dataset first means: Extract the quality change characteristics of the data source corresponding to the sub-dataset and the data change direction characteristics corresponding to the sub-dataset.

4. The multi-source data fusion measurement method for skid-mounted three-phase separation equipment according to claim 3 is characterized in that: Analyzing changes in sub-datasets refers to: Based on the quality change characteristics of the data source, the source quality changes of the sub-dataset are analyzed to generate a quality change index. Based on the data change direction characteristics of the data source, the fluctuation amplitude and fluctuation direction changes of the sub-dataset are analyzed to generate a fluctuation change index.

5. The multi-source data fusion measurement method for skid-mounted three-phase separation equipment according to claim 4 is characterized in that: The logic for obtaining the quality change index is: There are multiple quality dimensions i by default. For each quality dimension i, the following normalization method is used: ； is the standardized quality score of quality dimension i at time t, is the original quality score of quality dimension i at time t, is the minimum score of quality dimension i, is the maximum score of quality dimension i, is the average of the original quality scores of quality dimension i, , All are preset non-zero adjustment coefficients; Introducing sliding time window W and time decay factor , using the weighted average method to combine the quality scores of all dimensions and calculate the overall quality score of the data source: ; is the preset proportional coefficient corresponding to mass dimension i, n is the total number of dimensions of mass dimension i, is the standardized quality score of quality dimension i at time k, is the overall quality score of the data source at time t; Introducing a volatility penalty : ; is the average quality score after standardization of quality dimension i; The formula for the total quality score is modified by combining the penalty term: ; is the overall quality score of the corrected data source at time t, is the preset control coefficient corresponding to the volatility penalty term; Finally, the average of the overall quality scores of all corrected data sources within a fixed time length of the sub-dataset is calculated to obtain the quality change index.

6. The multi-source data fusion measurement method for skid-mounted three-phase separation equipment according to claim 5 is characterized in that: The logic for obtaining the volatility change index is: Within a fixed time length of the sub-dataset, the absolute difference between the data of every two adjacent time points is calculated and the average value is taken to represent the fluctuation range of the sub-dataset within the fixed time length; Within a fixed time length of the sub-dataset, the data change rate is further calculated, that is, the difference between the data changes at adjacent time points, and the average of the absolute values ​​is taken to indicate the severity of the fluctuation; When the direction of change at adjacent time points reverses, it is counted as one direction change, and the total number of direction changes is divided by the fixed time length to obtain the direction change frequency; By calculating the standard deviation of the data changes at adjacent time points within a fixed time length, the discrete degree of data changes is measured, and the fluctuation stability used to evaluate the stability of data fluctuations is obtained; Multiply the fluctuation amplitude, fluctuation severity, direction change frequency, and fluctuation smoothness by the preset weight coefficients respectively, and then sum them up to obtain the fluctuation change index.

7. The multi-source data fusion measurement method for skid-mounted three-phase separation equipment according to claim 6 is characterized in that: Importing the feature data after change analysis into the pre-trained convolutional neural network model to obtain different classification results and decide whether to participate in multi-source data fusion measurement means: The quality change index and fluctuation change index corresponding to each sub-dataset are obtained and used as the input data of the convolutional neural network model together to obtain a classification result of 1 or 0. When the classification result is 1, it indicates that the sub-dataset is a high-quality data set, and when the classification result is 0, it indicates that the sub-dataset is a low-quality data set; When the classification results of each sub-data set corresponding to multiple data source types within the same fixed time length are all 1, the fixed time length is marked as the multi-source data fusion time period and participates in the multi-source data fusion measurement.

8. The multi-source data fusion measurement method for skid-mounted three-phase separation equipment according to claim 7 is characterized in that: The logic of issuing the early warning signal is: The number of time periods marked as multi-source data fusion time periods is divided by the total number of all time periods to obtain a ratio value 1, and the ratio value 1 is compared with a preset warning threshold 1. If the ratio value 1 is less than or equal to the preset warning threshold 1, a type 1 warning signal is generated; Get the total number of sub-datasets corresponding to a data source type with a classification result of 0, and then divide it by the total number of time periods to get proportion value 2, compare proportion value 2 with the preset warning threshold 2, and if proportion value 2 is greater than or equal to the preset warning threshold 2, a second-class warning signal is generated.

9. A multi-source data fusion metering system for a skid-mounted three-phase separation device, which is implemented based on the multi-source data fusion metering method for a skid-mounted three-phase separation device according to any one of claims 1 to 8, characterized in that: include: The data acquisition module uses corresponding sensor devices to collect the operation data of the skid-mounted three-phase separation device based on multiple preset data source types, and obtains a collection data set corresponding to each data source type; The time division module divides the collected data set corresponding to each data source type according to a fixed time length, and records each divided part as a sub-data set for separate analysis; The feature extraction module first extracts features from each sub-dataset, then analyzes the changes in the sub-dataset to provide feature data for subsequent classification; The data classification module imports the feature data after change analysis into the pre-trained convolutional neural network model for each sub-data set to obtain different classification results and decide whether to participate in multi-source data fusion measurement; The early warning module decides whether to issue an early warning signal based on the classification results output by the sub-datasets of all data source types in the convolutional neural network model.

Images