Gathering and transportation pipeline effusion state identification method and system based on time-frequency characteristic screening

Abstract

The invention relates to the technical field of pipeline effusion detection, in particular to a gathering and transportation pipeline effusion state recognition method and system based on time-frequency feature screening, and the method comprises the following steps: S1, obtaining modeling data of a plurality of effusion states of a gathering and transportation pipeline, and calibrating a state label of the modeling data; s2, extracting time-frequency features and marking index digits; s3, screening time-frequency features by using a Lasso regression method, calculating feature coefficients of the time-frequency features, retaining the time-frequency features of which the feature coefficients are not 0, and recording corresponding index digits as sample index digits; s4, constructing a random forest model; and S5, acquiring real-time data of the gathering and transportation pipeline, extracting time-frequency characteristics of the real-time data, extracting the time-frequency characteristics according to the sample index bits, and inputting the extracted time-frequency characteristics into the random forest model to obtain the effusion state of the gathering and transportation pipeline. According to the invention, the identification accuracy of the hydrops state is improved.

Description

Technical Field

[0001] The present invention relates to the technical field of pipeline liquid accumulation detection, and particularly to a method and system for identifying the liquid accumulation state of gathering and transportation pipelines based on time-frequency feature screening. Background Art

[0002] The existing pipeline liquid accumulation identification methods mainly focus on the research of liquid accumulation formation laws based on simulation and the research of monitoring and early warning methods based on experience. These methods lack online identification research based on data-driven, and do not deeply analyze the sensor data of gathering and transportation pipelines, so insufficient information is obtained.

[0003] The data-driven identification method involves three key steps: data acquisition, feature extraction, and classification model construction. During the online identification process, by inputting real-time data into the model, the prediction result can be quickly obtained. In terms of feature extraction, since a limited number of sensors are required in gathering and transportation pipelines to accurately identify the liquid accumulation state, the features extracted solely relying on time-domain signals or frequency-domain signals are often insufficient and incomplete in terms of information volume. Therefore, the method of time-frequency feature fusion can obtain more comprehensive information. Although high-dimensional time-frequency features can be extracted in the existing technology, they are not effectively screened, which results in a large amount of useless information, thus reducing the accuracy of classification and identification. Summary of the Invention

[0004] The present invention aims to solve at least one of the technical problems existing in the related art. For this purpose, the present invention provides a method and system for identifying the liquid accumulation state of gathering and transportation pipelines based on time-frequency feature screening, which solves the technical problem of low accuracy of liquid accumulation state identification in the existing technology for data processing methods, obtains relatively comprehensive time-frequency features, retains the key information highly relevant to the label in the time-frequency features, and improves the accuracy of liquid accumulation state identification.

[0005] The present invention provides a method for identifying the liquid accumulation state of gathering and transportation pipelines based on time-frequency feature screening, including the following steps: S1, obtaining the modeling data of several liquid accumulation states of the gathering and transportation pipeline, and calibrating the state labels of the modeling data; S2, extracting the time-frequency features of the modeling data and marking the index bits of each time-frequency feature; S3, using the Lasso regression method to screen the time-frequency features and calculate the feature coefficients of the time-frequency features, retaining the time-frequency features with non-zero feature coefficients and recording the corresponding index bits as sample index bits; S4, using the screened time-frequency features as modeling samples, and using the state labels corresponding to the time-frequency features in the modeling samples as supervision to construct a random forest model; S5, obtaining the real-time data of the gathering and transportation pipeline, and extracting the time-frequency features of the real-time data; S6. Extract part of the data from the time-frequency features of the real-time data according to the number of bits of the sample index, and input the extracted time-frequency features into the random forest model to obtain the liquid accumulation state of the gathering and transportation pipeline.

[0006] A further improvement of the method for identifying the liquid accumulation state of a gathering and transportation pipeline based on time-frequency feature screening according to the present invention is that step S1 includes the following steps: S11: Collect the detection signals of the sensors of the gathering and transportation pipeline to construct a data set, and determine the liquid accumulation state corresponding to each data point in the data set; S12: Use the sliding window method to select some data points in the data set as modeling data and construct a two-dimensional matrix; It is assumed that the modeling data consists of sliding windows, and the length of each sliding window is , then each sliding window signal is . Combine each sliding window signal to form modeling data, and the modeling data is a two-dimensional matrix , is the real number field; S13: Calibrate the state labels of the modeling data according to the liquid accumulation state. The values of the state labels are integers and are in a proportional relationship with the liquid level height corresponding to the liquid accumulation state. The modeling data corresponding dimensional label vector can be expressed as: where represents the state label corresponding to the first sliding window signal, represents the state label corresponding to the second sliding window signal, represents the th sliding window signal corresponding state label.

[0007] A further improvement of the method for identifying the liquid accumulation state of a gathering and transportation pipeline based on time-frequency feature screening according to the present invention is that in step S2, extracting the time-frequency features of the modeling data includes the following steps: S21. Determine the time-domain signal of each sliding window signal in the modeling data, and perform Fourier transform on the time-domain signal to obtain the corresponding frequency-domain signal; S22. Extract time-domain features from the time-domain signal and frequency-domain features from the frequency-domain signal; S23. Fuse the time-domain features and frequency-domain features to obtain the fused time-frequency features; S24. Perform normalization processing on the fused time-frequency features to obtain the time-frequency features.

[0008] A further improvement of the method for identifying the liquid accumulation state of gathering and transportation pipelines based on time-frequency feature screening in the present invention lies in that the extraction method of time-domain features is as follows: Wherein, represents the value of the time-domain signal at the th moment, represents the total number of points of the time-domain signal, represents the first feature of the time-domain signal, represents the second feature of the time-domain signal, represents the third feature of the time-domain signal, represents the fourth feature of the time-domain signal, represents the fifth feature of the time-domain signal, represents the sixth feature of the time-domain signal, represents the seventh feature of the time-domain signal, represents the eighth feature of the time-domain signal, represents the ninth feature of the time-domain signal, represents the tenth feature of the time-domain signal, represents the eleventh feature of the time-domain signal, represents the twelfth feature of the time-domain signal, represents the thirteenth feature of the time-domain signal, represents the fourteenth feature of the time-domain signal, represents the fifteenth feature of the time-domain signal.

[0009] A further improvement of the method for identifying the liquid accumulation state of gathering and transportation pipelines based on time-frequency feature screening in the present invention lies in that the extraction method of frequency-domain features is as follows: Wherein, represents the total number of points of the frequency-domain signal, represents the th amplitude or energy value of the frequency-domain signal, represents the th frequency corresponding to the frequency-domain signal, represents the first feature of the frequency-domain signal, represents the second feature of the frequency-domain signal, represents the third feature of the frequency-domain signal, represents the fourth feature of the frequency-domain signal, represents the fifth feature of the frequency-domain signal, represents the sixth feature of the frequency-domain signal, represents the seventh feature of the frequency-domain signal, represents the eighth feature of the frequency-domain signal, represents the ninth feature of the frequency-domain signal, Represents the tenth feature of the frequency-domain signal, Represents the eleventh feature of the frequency-domain signal, Represents the twelfth feature of the frequency-domain signal, Represents the thirteenth feature of the frequency-domain signal.

[0010] A further improvement of the method for identifying the liquid accumulation state of gathering and transportation pipelines based on time-frequency feature screening in the present invention lies in that the time-frequency features are fused into , and the two-dimensional matrix composed of the fused time-frequency features of several sliding window signals is , The method of standardization is: Among them, Represents the data in the th row and the th column of the two-dimensional matrix , Represents the average value of the data in the th column, Represents the standard deviation of the data in the th column, Represents the data after standardization; After each sliding window signal is standardized, the standardized time-frequency feature obtained is , and the two-dimensional matrix composed of the standardized time-frequency features of each sliding window signal is .

[0011] A further improvement of the method for identifying the liquid accumulation state of gathering and transportation pipelines based on time-frequency feature screening in the present invention lies in that step S3 specifically includes: Using the Lasso regression method to set the regression model as, Among them, Represents the feature coefficient vector, Represents the random error, Represents the feature coefficient corresponding to the first time-frequency feature of each sliding window signal in the modeling data, Represents the feature coefficient corresponding to the second time-frequency feature of each sliding window signal in the modeling data, Represents the feature coefficient corresponding to the 28th time-frequency feature of each sliding window signal in the modeling data; The expression of the regression model in the least squares regularization estimation is: Among them, Represents the penalty coefficient, Represents the standardized feature corresponding to the th sliding window signal, The value range is , represents the status label corresponding to the th sliding window signal; The eigen coefficient vector is calculated by the method of least squares regularization estimation ; Record that the th eigen coefficient is not 0, and retain the normalized time-frequency feature in the normalized time-frequency feature that corresponds to the eigen coefficients, which is expressed as ; The matrix formed by the filtered normalized time-frequency features is , and its specific expression is: Among them, represents the filtered normalized time-frequency feature corresponding to the first sliding window signal in the modeling data, represents the filtered normalized time-frequency feature corresponding to the second sliding window signal in the modeling data, represents the filtered normalized time-frequency feature corresponding to the th sliding window signal in the modeling data.

[0012] A further improvement of the method for identifying the liquid accumulation state of gathering and transportation pipelines based on time-frequency feature screening in the present invention is that step S5 includes extracting the time-frequency features of real-time data ; Step S6 includes: S61, perform normalization processing on the time-frequency features of real-time data to obtain normalized time-frequency features, and the normalization processing method is: Among them, represents the th feature of real-time data, represents the th feature after normalization; S62, retain the normalized time-frequency features of real-time data corresponding to the sample index bits, denoted as ; S63, input the filtered normalized time-frequency features into the constructed random forest model to obtain the liquid accumulation state of the gathering and transportation pipeline.

[0013] The present invention also provides a system for identifying the liquid accumulation state of gathering and transportation pipelines based on time-frequency feature screening. The identification system is used to execute the above-mentioned identification method, and the identification system includes: A time-frequency feature extraction unit, a feature screening unit, a random forest model construction unit, and a state recognition unit. The time-frequency feature extraction unit is connected to the feature screening unit, the time-frequency feature extraction unit is connected to the state recognition unit, the feature screening unit, the random forest model construction unit, and the state recognition unit are connected, and the random forest model construction unit is connected to the state recognition unit; A time-frequency feature extraction unit, configured to extract time-frequency features and perform normalization processing; A feature screening unit, configured to use the Lasso regression method to screen time-frequency features and calculate the feature coefficients of the time-frequency features, retain the time-frequency features with non-zero feature coefficients, and record the corresponding index bits as sample index bits; A random forest model construction unit, configured to construct a random forest model; A state recognition unit, configured to extract the time-frequency features of real-time data and perform normalization processing.

[0014] The present invention effectively realizes the identification of the liquid accumulation state of the gathering and transportation pipeline by combining time-frequency feature extraction and Lasso regression feature screening and applying a random forest model. Through formula calculation, the present invention obtains comprehensive time-frequency features, including 15 time-domain features and 13 frequency-domain features, fully mining the information in the signal. Using Lasso regression for feature screening retains the key information of the time-frequency features highly correlated with the label, thus effectively improving the recognition accuracy of the model. The present invention uses a random forest as a classification model, enhancing the performance and robustness of the gathering and transportation pipeline state recognition. In addition, the data-driven recognition method of the present invention breaks through the limitations of traditional methods and has the advantages of automation and simplified work processes. It can be applied to the actual industrial environment to provide real-time prediction results of the gathering and transportation pipeline state and assist the staff in making judgments.

[0015] Additional aspects and advantages of the present invention will be given in part in the following description, become apparent in part from the following description, or be understood through the practice of the present invention. BRIEF DESCRIPTION OF THE DRAWINGS

[0016] In order to more clearly illustrate the technical solutions in the present invention or the prior art, the following will briefly introduce the drawings required for the description of the embodiments or the prior art. Obviously, the drawings in the following description are some embodiments of the present invention. For those of ordinary skill in the art, without creative efforts, other drawings can also be obtained based on these drawings.

[0017] Figure 1 It is a flowchart of a method and system for identifying the liquid accumulation state of a gathering and transportation pipeline based on time-frequency feature screening provided by an embodiment of the present invention.

[0018] Figure 2It is the confusion matrix for identifying the liquid accumulation state of gathering and transportation pipelines under the safe liquid level and warning liquid level provided by the embodiments of the present invention.

[0019] Figure 3 It is the confusion matrix for identifying the liquid accumulation state of gathering and transportation pipelines under the safe liquid level, low liquid level and high liquid level provided by the embodiments of the present invention. Specific embodiments

[0020] To make the objectives, technical solutions and advantages of the present invention clearer, the technical solutions in the present invention will be clearly and completely described below with reference to the accompanying drawings in the present invention. Apparently, the described embodiments are some, but not all, of the embodiments of the present invention. All other embodiments obtained by those of ordinary skill in the art based on the embodiments in the present invention without creative efforts shall fall within the protection scope of the present invention. The following embodiments are used to illustrate the present invention, but shall not be used to limit the scope of the present invention.

[0021] The following combines Figure 1 to describe a method for identifying the liquid accumulation state of gathering and transportation pipelines based on time-frequency feature screening, including the following steps: S1. Obtain the modeling data of several liquid accumulation states of the gathering and transportation pipeline, and calibrate the state labels of the modeling data; S2. Extract the time-frequency features of the modeling data and mark the index bits of each time-frequency feature; S3. Use the Lasso regression method to screen the time-frequency features and calculate the feature coefficients of the time-frequency features, retain the time-frequency features with non-zero feature coefficients, and record the corresponding index bits as the sample index bits; S4. Use the screened time-frequency features as modeling samples, and use the state labels corresponding to the time-frequency features in the modeling samples as supervision to construct a random forest model; S5. Obtain the real-time data of the gathering and transportation pipeline, and extract the time-frequency features of the real-time data; S6. Extract part of the data from the time-frequency features of the real-time data according to the sample index bits, and input the extracted time-frequency features into the random forest model to obtain the liquid accumulation state of the gathering and transportation pipeline.

[0022] Through precise time-frequency feature extraction and Lasso regression screening, this identification method can efficiently identify the key features related to the warning liquid level, thereby constructing a high-precision random forest model, improving the accuracy of liquid accumulation state identification, reducing the risks of false alarms and missed alarms, and providing a strong guarantee for the safe operation of gathering and transportation pipelines.

[0023] This identification method realizes the real-time monitoring and early warning of the liquid accumulation state in gathering and transportation pipelines. By obtaining real-time data and inputting it into the constructed random forest model, it can quickly judge whether the liquid accumulation state is close to or reaches the warning liquid level, thereby triggering the alarm mechanism in a timely manner and notifying relevant personnel to take corresponding measures. This ability of real-time monitoring and early warning helps to detect and handle potential safety hazards in a timely manner and avoid accidents.

[0024] This method also has the potential for continuous optimization and improvement. By continuously accumulating real-time data and model output results, a rich historical dataset can be formed for subsequent training and optimization of the model. This can not only improve the accuracy and reliability of the model in identifying the liquid accumulation state, but also adapt to changes in different working conditions and liquid accumulation states, enhancing the generality and adaptability of the method.

[0025] In a preferred embodiment of a method for identifying the liquid accumulation state in gathering and transportation pipelines based on time-frequency feature screening according to the present invention, step S1 includes the following steps: S11: Collect the detection signals of the sensors in the gathering and transportation pipelines to construct a dataset and determine the liquid accumulation state corresponding to each data point in the dataset; by collecting the detection signals of the sensors in the gathering and transportation pipelines and constructing a dataset, the change situation of the liquid accumulation state can be comprehensively and accurately recorded. This step provides a solid foundation for subsequent data analysis and model construction, ensuring the accuracy and reliability of the liquid accumulation state identification; S12: Use the sliding window method to select some data points in the dataset as modeling data and construct a two-dimensional matrix; It is assumed that the modeling data consists of sliding windows, and the length of each sliding window is , then each sliding window signal is , and the signals of each sliding window are combined to form modeling data, and the modeling data is a two-dimensional matrix , is the real number field; Using the sliding window method to select some data points in the dataset as modeling data and construct a two-dimensional matrix effectively reduces the data dimension, reduces the computational complexity, and at the same time retains the key information of the liquid accumulation state. This not only improves the efficiency of model construction, but also enhances the generalization ability of the model, enabling it to better adapt to changes in different working conditions and liquid accumulation states; the construction of the two-dimensional matrix also facilitates subsequent time-frequency feature extraction and feature screening, further improving the accuracy and robustness of the liquid accumulation state identification; S13: Calibrate the state labels of the modeling data according to the liquid accumulation state, and the values of the state labels are integers and are in a proportional relationship with the liquid level height corresponding to the liquid accumulation state, Modeling data corresponding to dimensional label vector It can be expressed as: Wherein, represents the status label corresponding to the first sliding window signal, represents the status label corresponding to the second sliding window signal, represents the th status label corresponding to the sliding window signal.

[0026] Preferably, several of the liquid accumulation states include a safety liquid level and a warning liquid level. The modeling data corresponding to the safety liquid level is calibrated as the first label, and the modeling data corresponding to the warning liquid level is calibrated as the second label.

[0027] Further, extracting the time-frequency features of the modeling data in step S2 includes the following steps: S21, determining the time-domain signal of each sliding window signal in the modeling data, and performing Fourier transform on the time-domain signal to obtain the corresponding frequency-domain signal; S22, extracting time-domain features from the time-domain signal and frequency-domain features from the frequency-domain signal; Time-domain features such as mean, variance, peak value, etc. reflect the change trend of the signal over time, while frequency-domain features such as main frequency, bandwidth, etc. reveal the distribution law of the signal in the frequency domain; S23, fusing the time-domain features and frequency-domain features to obtain fused time-frequency features; The fusion process ensures the complementarity of time-domain and frequency-domain information, providing a more comprehensive feature description for subsequent liquid accumulation state recognition; S24, performing normalization processing on the fused time-frequency features to obtain time-frequency features; The normalization processing eliminates the dimension difference between different features, improving the comparability of features and the stability of the model.

[0028] Further, the extraction method of time-domain features is: Wherein, represents the value of the time-domain signal at the th moment, represents the total number of points of the time-domain signal, represents the first feature of the time-domain signal, reflecting the DC component of the time-domain signal, represents the second feature of the time-domain signal, characterizing the average magnitude of the amplitude of the time-domain signal, represents the third feature of the time-domain signal, reflecting the average level of the energy of the time-domain signal, represents the fourth feature of the time-domain signal, measuring the degree of fluctuation of the time-domain signal deviating from the average value, represents the fifth feature of the time-domain signal, reflecting a special weighted average feature of the amplitude of the time-domain signal, represents the sixth feature of the time-domain signal, representing the effective amplitude of the time-domain signal Represents the seventh feature of the time-domain signal, reflecting the maximum amplitude, Represents the eighth feature of the time-domain signal, the maximum value of the time-domain signal, Represents the ninth feature of the time-domain signal, the minimum value of the time-domain signal, Represents the tenth feature of the time-domain signal, reflecting the waveform feature of the time-domain signal, Represents the eleventh feature of the time-domain signal, measuring the symmetry of the time-domain signal, Represents the twelfth feature of the time-domain signal, reflecting the relative relationship between the AC component and the DC component, Represents the thirteenth feature of the time-domain signal, the ratio of the effective value to the average of the absolute values, Represents the fourteenth feature of the time-domain signal, the skewness of the time-domain signal, measuring the symmetry of the probability distribution, Represents the fifteenth feature of the time-domain signal, the kurtosis of the time-domain signal, measuring the steepness of the amplitude distribution.

[0029] Furthermore, the extraction method of the frequency-domain features is as follows: Among them, Represents the total number of points of the frequency-domain signal, Represents the amplitude or energy value of the Represents the frequency corresponding to the Represents the first feature of the frequency-domain signal, reflecting its DC component, Represents the second feature of the frequency-domain signal, measuring its deviation from the average value degree of fluctuation, Represents the third feature of the frequency-domain signal, measuring the symmetry of the probability distribution, Represents the fourth feature of the frequency-domain signal, measuring the steepness of the amplitude distribution, Represents the fifth feature of the frequency-domain signal, reflecting average feature of Represents the sixth feature of the frequency-domain signal, reflecting effective amplitude, Represents the seventh feature of the frequency-domain signal, measuring deviation from degree of fluctuation, Represents the eighth feature of the frequency-domain signal, a normalized measure of the amplitude feature, Represents the ninth feature of the frequency-domain signal, a normalized measure, Represents the tenth feature of the frequency-domain signal, reflecting the ratio of the fluctuation to its effective amplitude, Represents the eleventh feature of the frequency-domain signal, measuring Symmetry of the distribution represents the twelfth feature of the frequency-domain signal and measures the steepness of the distribution represents the thirteenth feature of the frequency-domain signal, a special weighted metric

[0030] Furthermore, the fused time-frequency features are The two-dimensional matrix composed of the fused time-frequency features of several sliding-window signals is That is where represents the meaning of the real number field on a matrix with 28 columns and The method of normalization is as follows where represents the two-dimensional matrix in the row and the column data represents the average value of the column data represents the standard deviation of the column data represents the data after normalization After each sliding-window signal is normalized, the normalized time-frequency feature is The two-dimensional matrix composed of the normalized time-frequency features of each sliding-window signal is .

[0031] Preferably, step S3 specifically includes Using the Lasso regression method to set the regression model as where represents the feature coefficient vector represents the random error represents the feature coefficient corresponding to the first time-frequency feature of each sliding-window signal in the modeling data represents the feature coefficient corresponding to the second time-frequency feature of each sliding-window signal in the modeling data represents the feature coefficient corresponding to the 28th time-frequency feature of each sliding-window signal in the modeling data The regression model in the least squares regularization estimation is expressed as where represents the penalty coefficient represents the The standardized features corresponding to a sliding window signal whose value range is , indicating the status label corresponding to the th sliding window signal; The feature coefficient vector is calculated by the method of least squares regularization estimation; If it is recorded that the feature coefficient of the th feature coefficient is not 0, it can be known that there are feature coefficients of the standardized time-frequency features that are not 0. Retain the standardized time-frequency features corresponding to the feature coefficients in the standardized time-frequency features, denoted as ; The matrix formed by the screened standardized time-frequency features is , and its specific representation is: wherein, represents the screened standardized time-frequency feature corresponding to the first sliding window signal in the modeling data, represents the screened standardized time-frequency feature corresponding to the second sliding window signal in the modeling data, represents the screened standardized time-frequency feature corresponding to the th sliding window signal in the modeling data.

[0032] Specifically, step S5 includes extracting the time-frequency features of the real-time data ; Step S6 includes: S61, standardize the time-frequency features of the real-time data to obtain standardized time-frequency features. The method of standardization is: wherein, represents the th feature of the real-time data, represents the th feature after standardization; S62, retain the standardized time-frequency features of the real-time data corresponding to the number of sample index bits, denoted as ; S63, input the screened standardized time-frequency features into the constructed random forest model to obtain the liquid accumulation state of the gathering and transportation pipeline.

[0033] The present invention also provides a gathering and transportation pipeline liquid accumulation state recognition system based on time-frequency feature screening. The recognition system is used to execute the recognition method as described above. The recognition system includes: A time-frequency feature extraction unit, a feature screening unit, a random forest model construction unit, and a state recognition unit. The time-frequency feature extraction unit is connected to the feature screening unit, the time-frequency feature extraction unit is connected to the state recognition unit, the feature screening unit, the random forest model construction unit, and the state recognition unit are connected, and the random forest model construction unit is connected to the state recognition unit; The time-frequency feature extraction unit is used to extract time-frequency features and perform normalization processing; The feature screening unit is used to screen time-frequency features using the Lasso regression method, calculate the feature coefficients of the time-frequency features, retain the time-frequency features with non-zero feature coefficients, and record the corresponding index bits as sample index bits; The random forest model construction unit is used to construct a random forest model; The state recognition unit is used to extract the time-frequency features of real-time data and perform normalization processing.

[0034] The present invention effectively realizes the identification of the liquid accumulation state in gathering and transportation pipelines by combining time-frequency feature extraction and Lasso regression feature screening and applying a random forest model. Through formula calculation, the present invention obtains comprehensive time-frequency features, including 15 time-domain features and 13 frequency-domain features, fully mining the information in the signal. Using Lasso regression for feature screening retains the key information of time-frequency features highly correlated with the label, thereby effectively improving the recognition accuracy of the model. The present invention uses a random forest as a classification model, enhancing the performance and robustness of the identification of the gathering and transportation pipeline state. In addition, the present invention is based on a data-driven identification method, breaking through the limitations of traditional methods, and having the advantages of automation and simplified work processes. It can be applied to actual industrial environments, provide real-time prediction results of the gathering and transportation pipeline state, and assist staff in making judgments.

[0035] In a specific embodiment, an ultrasonic sensor is used to obtain the one-dimensional signal of the ultrasonic sensor as the initial sensor detection signal. If the types of sensors in the gathering and transportation pipeline are extended to multiple, the measurement variables will be extended to a higher dimension.

[0036] The state division of the liquid accumulation state is divided according to the actual situation by itself. Its label is an integer and increases as the liquid accumulation level of each state increases. In a specific embodiment, the liquid accumulation state is divided into two categories: a safe liquid level and a warning liquid level. The label of the safe liquid level is 0, and the label of the warning liquid level is 1. Then the first label represents the safe liquid level label, and the second label represents the warning liquid level label. In another specific implementation case, the liquid accumulation state is divided into three categories: a safe liquid level, a low liquid level, and a high liquid level. The label of the safe liquid level is 0, the label of the low liquid level is 1, and the label of the high liquid level is 2. Then the first label represents the safe liquid level label, and the second label represents the low liquid level label and the high liquid level label.

[0037] To verify the effectiveness of the recognition method of the present invention, the following experiments were conducted.

[0038] The inner diameter of the pipeline of the gathering and transportation pipeline is 110 mm. By artificially setting the liquid level in the pipeline to simulate the liquid accumulation volume in actual industry. Among them, when the used labels are the safety liquid level label is 0 and the warning liquid level label is 1 (hereinafter referred to as the binary classification situation), the samples with the liquid accumulation level in the pipeline of 0 mm - 10 mm are considered as the safety liquid level, and the samples with the liquid accumulation level of 15 mm - 100 mm are considered as the warning liquid level; when the used labels are the safety liquid level label is 0, the low liquid level label is 1, and the high liquid level label is 2 (hereinafter referred to as the ternary classification situation), the samples with the liquid accumulation level in the pipeline of 0 mm - 10 mm are considered as the safety liquid level, the samples with the liquid accumulation level of 15 mm - 50 mm are considered as the low liquid level, and the samples with the liquid accumulation level of 55 mm - 100 mm are considered as the high liquid level.

[0039] In the binary classification situation, the modeling samples altogether include 2 states, each state includes 80 samples, a total of 160 samples. Before screening, there are 28 features, that is, the standardized time-frequency feature matrix is ; after screening, 3 features are retained, that is, the standardized time-frequency feature matrix after screening is ; in the ternary classification situation, the modeling samples altogether include 3 states, each state includes 80 samples, a total of 240 samples. Before screening, there are 28 features, that is, the standardized time-frequency feature matrix is , after screening, 5 features are retained, that is, the standardized time-frequency feature matrix after screening is .

[0040] The confusion matrix in the binary classification situation is as shown in Figure 2 , the confusion matrix in the ternary classification situation is as shown in Figure 3 . In the binary classification situation, there are 20 safety liquid level samples and 20 warning liquid level samples respectively. In the ternary classification situation, there are 20 safety liquid level samples, 20 low liquid level samples and 20 high liquid level samples respectively. The recognition accuracy of the gathering and transportation pipeline state in the binary classification and ternary classification situations can reach 100% and 90% respectively, indicating the effectiveness of the recognition method of the present invention.

[0041] 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 foregoing embodiments, those of ordinary skill in the art should understand that: they can still modify the technical solutions recorded in the foregoing embodiments, or perform equivalent replacements on some of the technical features; and these modifications or replacements do not make the essence of the corresponding technical solutions deviate from the spirit and scope of the technical solutions of the various embodiments of the present invention.

Claims

1. A method for identifying the liquid accumulation state of gathering and transportation pipelines based on time-frequency feature screening, characterized in that, It includes the following steps: S1. Obtain the modeling data of several liquid accumulation states of the gathering and transportation pipeline, and calibrate the status labels of the modeling data; S2. Extract the time-frequency features of the modeling data and mark the index bits of each time-frequency feature; S3. Use the Lasso regression method to screen the time-frequency features and calculate the feature coefficients of the time-frequency features, retain the time-frequency features with non-zero feature coefficients, and record the corresponding index bits as the sample index bits; S4. Use the screened time-frequency features as modeling samples, and use the status labels corresponding to the time-frequency features in the modeling samples as supervision to construct a random forest model; S5. Obtain the real-time data of the gathering and transportation pipeline and extract the time-frequency features of the real-time data; S6. Extract some data from the time-frequency features of the real-time data according to the sample index bits, and input the extracted time-frequency features into the random forest model to obtain the liquid accumulation state of the gathering and transportation pipeline.

2. The method for identifying the liquid accumulation state of gathering and transportation pipelines based on time-frequency feature screening according to claim 1, wherein Step S1 includes the following steps: S11: Collect the detection signals of the sensors of the gathering and transportation pipeline to construct a data set, and determine the liquid accumulation state corresponding to each data point in the data set; S12: Use the sliding window method to select some data points in the data set as modeling data and construct a two-dimensional matrix; It is assumed that the modeling data consists of sliding windows, and the length of each sliding window is . Then each sliding window signal is . Combining each sliding window signal forms the modeling data, and the modeling data is a two-dimensional matrix , being the real number field; S13: Calibrate the status labels of the modeling data according to the liquid accumulation state. The values of the status labels are integers and are in direct proportion to the liquid level height corresponding to the liquid accumulation state. Modeling data Corresponding Dimensional label vector Can be expressed as: Among them, represents the status label corresponding to the first sliding window signal, represents the status label corresponding to the second sliding window signal, represents the status label corresponding to the sliding window signal.

3. The method for identifying the liquid accumulation state of gathering and transportation pipelines based on time-frequency feature screening according to claim 2, wherein In step S2, the extraction of the time-frequency features of the modeling data includes the following steps: S21. Determine the time-domain signals of each sliding window signal in the modeling data, and perform Fourier transform on the time-domain signals to obtain the corresponding frequency-domain signals; S22. Extract time-domain features from the time-domain signals and extract frequency-domain features from the frequency-domain signals; S23. Fuse the time-domain features and frequency-domain features to obtain fused time-frequency features; S24. Perform normalization processing on the fused time-frequency features to obtain time-frequency features.

4. The method for identifying the liquid accumulation state of gathering and transportation pipelines based on time-frequency feature screening according to claim 3, wherein The extraction method of the time-domain features is: Among them, represents the value of the time-domain signal at the moment, represents the total number of points of the time-domain signal, represents the first feature of the time-domain signal, represents the second feature of the time-domain signal, represents the third feature of the time-domain signal, represents the fourth feature of the time-domain signal, represents the fifth feature of the time-domain signal, represents the sixth feature of the time-domain signal, represents the seventh feature of the time-domain signal, represents the eighth feature of the time-domain signal, represents the ninth feature of the time-domain signal, represents the tenth feature of the time-domain signal, represents the eleventh feature of the time-domain signal, represents the twelfth feature of the time-domain signal, represents the thirteenth feature of the time-domain signal, represents the fourteenth feature of the time-domain signal, represents the fifteenth feature of the time-domain signal.

5. The identification method for the liquid accumulation state of gathering and transportation pipelines based on time-frequency feature screening according to claim 4, characterized in that The extraction method of the frequency-domain features is: Among them, represents the total number of points of the frequency-domain signal, represents the amplitude or energy value of the th frequency-domain signal, represents the frequency corresponding to the th frequency-domain signal, represents the first feature of the frequency-domain signal, represents the second feature of the frequency-domain signal, represents the third feature of the frequency-domain signal, represents the fourth feature of the frequency-domain signal, represents the fifth feature of the frequency-domain signal, represents the sixth feature of the frequency-domain signal, represents the seventh feature of the frequency-domain signal, represents the eighth feature of the frequency-domain signal, represents the ninth feature of the frequency-domain signal, represents the tenth feature of the frequency-domain signal, represents the eleventh feature of the frequency-domain signal, represents the twelfth feature of the frequency-domain signal, represents the thirteenth feature of the frequency-domain signal.

6. The method for identifying the liquid accumulation state of gathering and transportation pipelines based on time-frequency feature screening according to claim 5, characterized in that The fused time-frequency feature is , and the two-dimensional matrix composed of the fused time-frequency features of several sliding window signals is , The method of normalization processing is: Among them, represents the data of the th row and th column in the two-dimensional matrix, represents the average value of the data in the th column, represents the standard deviation of the data in the th column, represents the data after standardization processing; After each sliding window signal is normalized, the normalized time-frequency feature is , and the two-dimensional matrix composed of the normalized time-frequency features of each sliding window signal is .

7. The method for identifying the liquid accumulation state of gathering and transportation pipelines based on time-frequency feature screening according to claim 6, wherein Step S3 specifically includes: Set the regression model using the Lasso regression method as, Among them, represents the feature coefficient vector, represents the random error, represents the feature coefficient corresponding to the first time-frequency feature of each sliding window signal in the modeling data, represents the feature coefficient corresponding to the second time-frequency feature of each sliding window signal in the modeling data, represents the feature coefficient corresponding to the 28th time-frequency feature of each sliding window signal in the modeling data; Regression model in least squares regularization estimation is expressed as: Among them, represents the penalty coefficient, represents the standardized feature corresponding to the th sliding window signal, , represents the status label corresponding to the th sliding window signal; The eigen coefficient vector is calculated by means of least square regularization estimation ; Record the non-zero characteristic coefficients, and retain the normalized time-frequency characteristics corresponding to the characteristic coefficients during standardization, denoted as ; The matrix formed by the filtered normalized time-frequency characteristics is , and its specific representation is: Among them, represents the screened and standardized time-frequency features corresponding to the first sliding window signal in the modeling data, represents the screened and standardized time-frequency features corresponding to the second sliding window signal in the modeling data, represents the screened and standardized time-frequency features corresponding to the sliding window signal.

8. A method for identifying the liquid accumulation state of gathering and transportation pipelines based on time-frequency feature screening according to claim 7, characterized in that Step S5 includes extracting the time-frequency features of the real-time data ; Step S6 includes: S61. Perform normalization processing on the time-frequency features of the real-time data to obtain normalized time-frequency features. The method of normalization processing is: Among them, represents the th feature of the real-time data, represents the th feature after standardization; S62, retain the normalized time-frequency features of the real-time data corresponding to the number of bits of the sample index, denoted as ; S63. Input the screened normalized time-frequency features into the constructed random forest model to obtain the liquid accumulation state of the gathering and transportation pipeline.

9. An identification system for the liquid accumulation state of gathering and transportation pipelines based on time-frequency feature screening, characterized in that, The recognition system is used to execute the recognition method according to any one of claims 1 to 8. The recognition system includes: A time-frequency feature extraction unit, a feature screening unit, a random forest model construction unit, and a state recognition unit. The time-frequency feature extraction unit is connected to the feature screening unit, the time-frequency feature extraction unit is connected to the state recognition unit, the feature screening unit, the random forest model construction unit, and the state recognition unit are connected, and the random forest model construction unit is connected to the state recognition unit; A time-frequency feature extraction unit, which is used to extract time-frequency features and perform normalization processing; A feature screening unit, which is used to use the Lasso regression method to screen time-frequency features and calculate the feature coefficients of the time-frequency features, retain the time-frequency features with non-zero feature coefficients, and record the corresponding index bits as the sample index bits; A random forest model construction unit for constructing a random forest model; A state recognition unit for extracting the time-frequency features of real-time data and performing normalization processing.

Images