Metering and monitoring method and system based on pipeline multiphase flow

Abstract

The invention discloses a multi-phase flow measurement monitoring method and system based on a pipeline, and belongs to the technical field of fluid measurement, and the method comprises the steps: obtaining an actual flow velocity value and a dual-mode sensing signal of a fluid in the pipeline; acquiring historical working condition data, dynamically adjusting the sampling frequency of the bimodal sensing signal, generating an optimized sensing sampling signal, and performing time-frequency domain combined noise reduction processing to generate edge end preprocessing data; performing multi-step prediction on the edge end preprocessing data by using a time sequence prediction model to generate a dynamic compensation coefficient; according to the dynamic compensation coefficient and the actual flow velocity value, a multiphase flow value is obtained through calculation, and an abnormal event early warning signal is generated according to the multiphase flow value. Resistance type and ultrasonic dual-mode signal fusion, time-frequency domain combined noise reduction and cloud edge collaborative dynamic compensation technologies are adopted, high-precision real-time monitoring of multiphase flow can be achieved, abnormal early warning and adjusting operation are intelligently triggered based on flow fluctuation, and pipeline operation safety and measurement reliability are comprehensively improved.

Description

Technical Field

[0001] The present invention relates to the technical field of fluid measurement, and in particular, to a pipeline multiphase flow metering and monitoring method and system. Background Art

[0002] The pipeline multiphase flow metering technology is mainly used for real-time flow monitoring of oil, gas, and water mixed fluids, and involves the measurement of core technical parameters such as conductivity difference, Doppler frequency shift, and phase fraction ratio.

[0003] In the prior art, a multi-sensor combination scheme with a fixed sampling frequency is mostly adopted, and the multiphase flow rate is calculated by the weighted average method or the static calibration model. For example, a resistive sensor and an ultrasonic probe are installed in parallel, and after collecting signals at a preset frequency, the flow velocity and phase fraction ratio are linearly fitted based on the historical data of a single working condition, so as to realize the separation metering of oil, gas, and water. Such a scheme also often corrects the measurement error by means of offline calibration or interpolation compensation.

[0004] However, the data synchronization between sensors in the above technical scheme is poor, resulting in the accumulation of matching errors between the phase fraction ratio and the flow velocity; the fixed sampling frequency cannot adapt to the dynamic characteristics of multiphase flow, and key signals are easily lost in high gas content or rapid phase change scenarios; the static calibration model is difficult to correct the real-time deviation caused by pressure fluctuations and temperature changes, and lacks the ability to robustly process complex noise interference. These defects directly affect the accuracy and real-time performance of multiphase flow rate monitoring, and limit the application reliability in industrial scenarios. Summary of the Invention

[0005] To solve the above problems, the present invention provides a pipeline multiphase flow metering and monitoring method and system, which adopts resistive and ultrasonic dual-mode signal fusion, time-frequency domain joint noise reduction, and cloud-edge collaborative dynamic compensation technology, can realize high-precision real-time monitoring of multiphase flow rate, and intelligently trigger abnormal early warning and adjustment operations based on flow rate fluctuations, comprehensively improving the safety and measurement reliability of pipeline operation.

[0006] The above object can be achieved by the following solutions: A pipeline multiphase flow metering and monitoring method, including obtaining the actual flow velocity value and bimodal sensing signals of the fluid in the pipeline, where the bimodal sensing signals include resistive phase fraction signals and ultrasonic frequency shift signals; obtaining historical operating condition data, and dynamically adjusting the sampling frequency of the bimodal sensing signals based on the historical operating condition data to generate an optimized sensing sampling signal; performing time-frequency domain joint noise reduction processing on the optimized sensing sampling signal to generate edge-end preprocessed data; using a preset time series prediction model in the cloud to perform multi-step prediction on the edge-end preprocessed data to generate a dynamic compensation coefficient; calculating the multiphase flow rate value according to the dynamic compensation coefficient and the actual flow velocity value; outputting the multiphase flow rate value, and generating an abnormal event warning signal according to the multiphase flow rate value.

[0007] Optionally, the generating the optimized sensing sampling signal includes: obtaining the real-time pressure data, temperature data and historical medium ratio data of the pipeline to obtain a basic data set; generating flow velocity prediction models for different phases based on the basic data set; calculating the correlation weights between the resistive phase fraction signal and the ultrasonic frequency shift signal according to the flow velocity prediction models; adjusting the sampling frequencies of each mode in the bimodal sensing signal based on the correlation weights to generate an optimized sensing sampling signal.

[0008] Optionally, the performing time-frequency domain joint noise reduction processing on the optimized sensing sampling signal to generate edge-end preprocessed data includes: performing wavelet transform on the ultrasonic frequency shift signal in the sensing sampling signal to filter out frequency domain noise to generate a preliminary noise reduction signal; combining the pulse characteristics of the resistive phase fraction signal to perform waveform matching on the preliminary noise reduction signal to remove the remaining noise and generate edge-end preprocessed data.

[0009] Optionally, the generating the edge-end preprocessed data further includes: extracting the mutation waveform in the resistive phase fraction signal as a reference template; calculating the local correlation coefficient between the preliminary noise reduction signal and the reference template; when the local correlation coefficient is lower than a preset correlation threshold, correcting the edge-end preprocessed data through backpressure simulation testing.

[0010] Optionally, the using a preset time series prediction model in the cloud to perform multi-step prediction on the edge-end preprocessed data to generate a dynamic compensation coefficient includes: dividing the edge-end preprocessed data into a high-confidence data set and a low-confidence data set according to a preset time stamp through a preset rule; using a reinforcement learning method to optimize the compensation parameters of the low-confidence data set to generate a compensation parameter increment; inputting the compensation parameter increment into a long short-term memory network for training to generate a dynamic compensation coefficient.

[0011] Optionally, the division into a high-confidence data set and a low-confidence data set according to a preset rule includes: calculating a phase division ratio error value between the resistive phase division signal and the ultrasonic frequency shift signal; determining whether the phase division ratio error value is less than a preset noise threshold and the noise energy is lower than a preset energy threshold; if so, marking the edge-end preprocessed data as a high-confidence data set; if not, marking the edge-end preprocessed data as a low-confidence data set.

[0012] Optionally, the generation of the dynamic compensation coefficient includes: constructing a multi-objective loss function based on the high-confidence data set and the compensation parameter increment; using the multi-objective loss function as the loss function of the long short-term memory network, and optimizing the multi-objective loss function by the gradient descent method to obtain the dynamic compensation coefficient.

[0013] Optionally, the calculation of the multiphase flow rate value according to the dynamic compensation coefficient and the actual flow rate value includes: extracting the phase division ratio and the frequency shift value from the edge-end preprocessed data; performing density weighted integration on the phase division ratio by using the dynamic compensation coefficient to generate a phase division density value; performing flow rate correction calculation on the actual flow rate value by using the dynamic compensation coefficient and the frequency shift value to generate a corrected flow rate value; calculating the multiphase flow rate value based on the phase division density value and the corrected flow rate value.

[0014] Optionally, the generation of the abnormal event warning signal according to the multiphase flow rate value includes: when the fluctuation amplitude of the multiphase flow rate value exceeds the safety threshold, triggering a pipeline backpressure regulation operation; synchronously uploading a warning log including the abnormal event type and the pipe section location information to the cloud platform.

[0015] Based on the same inventive concept, the present invention also provides a pipeline multiphase flow metering and monitoring system, the system includes: a data acquisition module, configured to acquire the actual flow rate value of the fluid in the pipeline and a dual-modal sensing signal, the dual-modal sensing signal including a resistive phase division signal and an ultrasonic frequency shift signal; a signal optimization module, configured to dynamically adjust the sampling frequency of the dual-modal sensing signal based on historical operating condition data to generate an optimized sensing sampling signal; a data processing module, configured to perform time-frequency domain joint noise reduction processing on the optimized sensing sampling signal to generate edge-end preprocessed data; a compensation calculation module, configured to perform multi-step prediction on the edge-end preprocessed data by using a time series prediction model preset in the cloud to generate a dynamic compensation coefficient; a flow rate calculation module, configured to calculate a multiphase flow rate value according to the dynamic compensation coefficient and the actual flow rate value; a flow rate output and warning module, configured to output the multiphase flow rate value and generate an abnormal event warning signal according to the multiphase flow rate value.

[0016] Compared with the prior art, the present invention has the following advantages: 1. Through complementary measurement of bimodal sensing signals and optimization of dynamic sampling, the present invention significantly improves the detection accuracy of the phase fraction ratio and flow velocity of multiphase flow under complex flow patterns; the collaborative use of resistive phase fraction signals and ultrasonic frequency shift signals can not only effectively distinguish fluid phases with different conductive characteristics, but also capture changes in flow velocity based on the Doppler effect, avoiding misjudgment problems caused by sudden changes in fluid physical properties in traditional single-modal sensors. 2. By adopting a time-frequency domain joint noise reduction and cloud-edge collaborative computing model, the noise interference is greatly reduced and the data processing efficiency is improved; through a noise reduction technology that combines wavelet transform and waveform matching, while retaining real fluid phase change signals, mechanical vibration and environmental interference noises are filtered out, and combined with a dynamic compensation mechanism of edge-end preprocessing and cloud-side time series prediction, the anti-interference and real-time performance of multiphase flow rate calculation are improved simultaneously. 3. Through the closed-loop design of the dynamic compensation coefficient and the abnormal warning mechanism, the adaptability and safety of the system to working condition changes are enhanced; a multi-step prediction model based on reinforcement learning and long short-term memory network can correct measurement errors in real time, dynamically adjust flow rate calculation parameters, and quickly trigger backpressure regulation and cloud-side warning in combination with flow rate fluctuation monitoring, effectively preventing pipeline blockage or leakage accidents and reducing operation and maintenance risks.

[0017] Other features and advantages of the present invention will be described in the following specification, and some of them will become obvious from the specification, or can be understood by implementing the present invention. The objectives and other advantages of the present invention can be achieved and obtained through the structures pointed out in the specification, claims and drawings. BRIEF DESCRIPTION OF THE DRAWINGS

[0018] In order to more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the following will briefly introduce the drawings required for use in the description of the embodiments or the prior art. Obviously, the following drawings are some embodiments of the present invention. For those of ordinary skill in the art, other drawings can be obtained based on these drawings without creative efforts.

[0019] Figure 1 It is a schematic flow chart of a pipeline multiphase flow metering and monitoring method according to an embodiment of the present invention.

[0020] Figure 2 It is a change curve diagram of the resistive phase change signal according to an embodiment of the present invention.

[0021] Figure 3 It is a schematic structural diagram of a pipeline multiphase flow metering and monitoring method according to an embodiment of the present invention. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0022] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the following will clearly and completely describe the technical solutions in the embodiments of the present invention with reference to the accompanying drawings in the embodiments of 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 of the present invention without creative efforts fall within the protection scope of the present invention.

[0023] Referring to Figure 1 , an embodiment of the present invention proposes a method for pipeline multiphase flow measurement and monitoring. By adopting the technologies of resistance and ultrasonic dual-mode signal fusion, time-frequency domain joint noise reduction, and cloud-edge collaborative dynamic compensation, it can achieve high-precision real-time monitoring of multiphase flow rate, and based on the flow rate fluctuation, intelligently trigger abnormal early warnings and adjustment operations, comprehensively improving the safety of pipeline operation and measurement reliability.

[0024] The specific steps of the method in this embodiment include: Obtain the actual flow velocity value of the fluid in the pipeline and the dual-mode sensing signal, where the dual-mode sensing signal includes a resistive phase fraction signal and an ultrasonic frequency shift signal: Specifically, a circular electrode array composed of 16 groups of platinum electrodes is annularly installed on the outer surface of the pipeline. By applying a 5 kHz alternating current to measure the conductivity difference of the fluid, a resistive phase fraction signal is generated to reflect the phase fraction ratio of oil, gas, and water. At the same time, 8 groups of distributed ultrasonic probes are arranged at intervals of 1 meter along the axial direction of the pipeline. Each group of probes emits 1 MHz ultrasonic waves and receives the reflected signals. By calculating the flow velocity difference through Doppler frequency shift, an ultrasonic frequency shift signal is generated.

[0025] Among them, the dual-mode sensing signal is a composite signal that simultaneously measures electrical and acoustic characteristics; the resistive phase fraction signal is a potential distribution electrical signal based on the conductivity difference of different phase fluids; the ultrasonic frequency shift signal is an acoustic signal caused by the movement of discrete phase particles in the fluid, which changes the ultrasonic wave frequency.

[0026] Obtain historical operating condition data, and dynamically adjust the sampling frequency of the dual-mode sensing signal based on the historical operating condition data to generate an optimized sensing sampling signal; Perform time-frequency domain joint noise reduction processing on the optimized sensing sampling signal to generate edge-end preprocessed data; Use the time series prediction model preset in the cloud to perform multi-step prediction on the edge-end preprocessed data to generate a dynamic compensation coefficient; Calculate the multiphase flow rate value according to the dynamic compensation coefficient and the actual flow velocity value; Output the multiphase flow rate value, and generate an abnormal event warning signal according to the multiphase flow rate value.

[0027] This method solves the misjudgment problem of traditional single sensors under complex flow patterns through the complementary fusion of bimodal signals, dynamically adjusts the sampling frequency to improve resource utilization efficiency; joint time-frequency domain noise reduction suppresses environmental interference while retaining the true phase boundary signal; the cloud-edge collaborative model combines long-term memory and real-time optimization to significantly improve the anti-interference ability and working condition adaptability of flow calculation; the closed-loop warning mechanism enables rapid response to abnormal events such as pipeline blockage and phase separation, reducing operation and maintenance costs and safety risks.

[0028] This implementation method forms a bimodal sensing system by fusing resistive phase separation signals and ultrasonic frequency shift signals, which is used to synchronously sense the distribution and motion characteristics of different phase media in the pipeline. The system dynamically optimizes the sampling frequency of the sensing signal in combination with historical working condition data to ensure the sampling efficiency and quality of the signal under different working conditions. Subsequently, noise interference in the signal is removed through a joint time-frequency domain noise reduction method (such as wavelet transform combined with waveform matching) to improve data accuracy. The data after edge processing is sent to a cloud-based time series prediction model such as Long Short-Term Memory Network (LSTM) for signal trend prediction in the next several steps, thereby generating dynamic coefficients for error compensation. Finally, the measured actual flow velocity value is adjusted according to the dynamic compensation coefficient to achieve accurate calculation of multiphase flow rate, and whether there are abnormal events is judged based on the flow rate change, thereby triggering the corresponding warning mechanism. This implementation method realizes more accurate capture of multiphase fluid characteristics by introducing bimodal signal acquisition and dynamic sampling frequency control technologies; improves the real-time performance and accuracy of data processing by combining time-frequency joint noise reduction and cloud-edge collaborative calculation models; the introduction of dynamic compensation coefficients effectively corrects sensing errors, enhances measurement accuracy and robustness; the integration of the abnormal warning mechanism greatly improves the safety response ability and intelligent level of the system, and is applicable to complex and changeable industrial field environments.

[0029] Optionally, the generation of the optimized sensing sampling signal includes: Obtain the real-time pressure data, temperature data and historical medium ratio data of the pipeline to obtain a basic data set; Specifically, the pressure data and temperature data are collected in real time from the pressure sensor and temperature sensor installed on the pipeline wall, and the historical medium ratio data of the oil, gas and water ratios in the same pipe section in the past 30 days is extracted through the historical database, such as the ratio record with 60% oil phase, 25% gas phase and 15% water phase. The pressure sensor is a piezoelectric sensor with an accuracy class of 0.1% FS; the temperature sensor uses a PT100 platinum resistance with a resolution of 0.1 °C. The three types of data are aligned in time series to form a basic data set, which is stored in the form of an array sequence in JSON format, and each timestamp corresponds to a three-dimensional vector containing pressure, temperature and medium ratio.

[0030] Generate flow velocity prediction models for different phases based on the basic data set; Specifically, the basic data set is input into the Gaussian process regression model for training, with the historical flow rate as the dependent variable and pressure, temperature, and medium ratio as the independent variables. Gaussian process regression is a non-parametric Bayesian model based on kernel functions. By optimizing covariance function parameters such as the radial basis function kernel, the non-linear relationship between the flow rates of each phase state is learned. For example, when the gas phase is dominant, the flow rate is positively correlated with the square of the pressure; when the water phase is dominant, the flow rate has a negative exponential relationship with the medium viscosity. The model output is an independent prediction equation for different phase states of oil, gas, and water, and its mathematical expression is: ; In the formula, is the predicted value of the flow rate of the phase; is the phase flow coefficient calculated by the phase Gaussian kernel function, is the real-time pressure, is the real-time temperature, is the viscosity attenuation coefficient, is the phase fluid viscosity, is the noise term.

[0031] Calculate the correlation weight value between the resistive phase fraction signal and the ultrasonic frequency shift signal according to the flow rate prediction model; Specifically, the similarity of the two signals is analyzed through the covariance matrix. Calculate the cross-covariance between the phase fraction ratio in the resistive phase fraction signal and the flow rate of the ultrasonic frequency shift signal, and then obtain the correlation weight value of the two signals through a weight assignment algorithm, such as the entropy weight method. The phase fraction ratio in the resistive phase fraction signal is, for example, the oil phase conductivity of 0.1 S / m and the gas phase of 0.01 S / m, and the flow rate of the ultrasonic frequency shift signal is, for example, the liquid phase of 3 m / s and the gas-liquid mixed flow of 2.5 m / s. For example, when the gas phase ratio is higher than 70%, due to the decreased sensitivity of ultrasonic waves to the bubble group, the weight value of the resistive signal is increased to 0.8, and the weight value of the ultrasonic signal is decreased to 0.2. The weight formula is: ; In the formula, is the weight coefficient of the resistive phase fraction signal, is the weight coefficient of the ultrasonic frequency shift signal, is the variance of the resistive phase fraction signal, is the variance of the ultrasonic frequency shift signal.

[0032] Adjust the sampling frequency of each mode in the dual-modal sensing signal based on the correlation weight value to generate an optimized sensing sampling signal.

[0033] Specifically, the resistive phase fraction signal and the ultrasonic frequency shift signal adopt frequencies according to the weight assignment. For the sampling frequency of the resistive phase fraction signal, there is: ; In the formula, is the set total sampling frequency. For the sampling frequency of the ultrasonic frequency shift signal, there is: ; Specifically, the total sampling frequency is set to 1.5 kHz. When the weight coefficient of the resistive phase separation signal is 0.6, the sampling frequency of the resistive phase separation signal is 900 Hz at this time, and the sampling frequency of the ultrasonic frequency shift signal is 600 Hz.

[0034] Exemplarily, taking a three-phase flow containing 65% crude oil, 25% formation water, and 10% natural gas as an example, the pressure sensor is read as 3.5 MPa and the temperature is 50 °C, and the average oil phase in the historical ratio is 68% to construct a basic data set. The flow velocity prediction model predicts the oil phase flow velocity of 2.8 m / s, the gas phase of 3.2 m / s, and the water phase of 2.5 m / s. The variance of the resistive signal is calculated as 0.12, and the variance of the ultrasonic signal is 0.35. According to the formula, the weight coefficient of the resistive phase separation signal is 0.74, and the weight coefficient of the ultrasonic frequency shift signal is 0.26. Therefore, the sampling rate of the resistive electrode array is increased to 2 kHz, and the ultrasonic probe is reduced to 800 Hz; the wavelet transform filters out the 10 kHz high-frequency noise caused by pump vibration, and the waveform matching eliminates the droplet impact interference, and the output signal-to-noise ratio is increased by 22 dB. By dynamically adjusting the dual-mode sampling frequency, the resolution of the resistive signal is preferentially guaranteed in the oil and gas dominant scenario, and the ultrasonic signal is prevented from generating invalid sampling due to bubble interference; the time-frequency noise reduction is combined with the physical feature template to effectively filter out the mechanical vibration noise unrelated to the phase separation change; the dynamic weight distribution makes the subsequent calculation more dependent on the high-confidence signal type, improving the flow accuracy.

[0035] Optionally, the time-frequency domain joint noise reduction process for the optimized sensing sampling signal to generate edge-end preprocessed data includes: Performing wavelet transform on the ultrasonic frequency shift signal in the sensing sampling signal to filter out the frequency domain noise and generate a preliminary noise reduction signal; Specifically, first perform discrete wavelet transform on the ultrasonic frequency shift signal. Discrete wavelet transform is a time-frequency analysis method that decomposes the signal through the mother wavelet function at different scales and time positions, separating the high-frequency noise from the low-frequency useful signal. Specifically, the Daubechies 4 wavelet basis is used, and the decomposition level is 5 layers. Hard thresholds are set for the high-frequency components of the first to third layers to filter out the noise components, and a preliminary noise reduction signal is generated after reconstructing the signal. The frequency domain noise is the high-frequency random fluctuation caused by pipeline mechanical vibration or electromagnetic interference in the ultrasonic signal, such as high-frequency harmonic interference above 100 kHz.

[0036] Combining the pulse characteristics of the resistive phase separation signal, performing waveform matching on the preliminary noise reduction signal to eliminate the remaining noise and generate edge-end preprocessed data.

[0037] Specifically, the pulse feature of the resistive phase fraction signal is the waveform of sudden change in conductivity caused by the change of the fluid phase interface. For example, the sudden drop in conductivity at the oil-gas / water interface is manifested as a rectangular pulse with a decreasing amplitude. Such a sudden change waveform segment is extracted from the resistive signal as a feature template, and the time window length is intercepted as 50 ms. In the corresponding time window of the preliminarily denoised signal, its local similarity with the feature template is calculated through a sliding window. The waveform matching is a similarity measurement method based on the dynamic time warping algorithm, allowing time offset during waveform alignment, and minimizing the Euclidean distance between the two window signals. If the similarity is lower than the threshold of 0.7, it is determined as the remaining noise segment and set to zero for elimination.

[0038] Optionally, the generation of edge-side preprocessed data further includes: Extracting the sudden change waveform in the resistive phase fraction signal as a reference template; Specifically, the reference template is a normalized waveform intercepted from the step change region of the resistive phase fraction signal. The sudden change waveform is defined as the interval where the phase fraction changes by more than 15% within 0.5 seconds. For example, the conductivity drop section corresponding to the oil phase dropping from 60% to 40%. The sliding window method is used to traverse the signal. After screening the sudden change segments that meet the conditions, the peak points of multiple samples are aligned and the average waveform is calculated to generate a reference template with a length of 200 sampling points.

[0039] Calculating the local correlation coefficient between the preliminarily denoised signal and the reference template; Specifically, based on the reference template, the normalized cross-correlation calculation is performed at each window position of the preliminarily denoised signal. The formula is: ; In the formula, is the local correlation coefficient, is the value of the th sampling point in the current window of the preliminarily denoised signal, is the average value of all sampling points in the current window of the preliminarily denoised signal, is the value of the th sampling point in the position corresponding to the current window of the preliminarily denoised signal in the reference signal, is the average value of all sampling points in the position corresponding to the current window of the preliminarily denoised signal in the reference signal, is the number of window sampling points. If is lower than the preset threshold such as 0.75, it is determined that there is noise in this window that needs to be corrected.

[0040] When the local correlation coefficient is lower than the preset correlation threshold, the edge-side preprocessed data is corrected through a backpressure simulation test.

[0041] Specifically, the backpressure simulation test is a method for verifying the stability of fluid phase separation by generating a reverse pressure wave through adjusting the valve at the end of the pipeline. During the calibration phase, the valve opening is reduced by 20% for 10 seconds, and the resistive and ultrasonic signals under the current pressure fluctuation are collected simultaneously. If the correlation coefficient during the calibration phase is higher than the correlation threshold, the multiphase flow characteristic parameters at this time are used to overwrite the original abnormal window data; if it is still lower than the correlation threshold, the data during this period is considered unreliable and marked as a segment to be excluded.

[0042] Exemplarily, taking the oil-water two-phase flow in an oilfield pipeline as an example, as Figure 2 shown, there are 3 sudden drops in conductivity at the phase interface changes in the resistive phase fraction signal, with the amplitude decreasing from 2.1 S / m to 0.8 S / m. There is flow velocity fluctuation noise in the ultrasonic frequency shift signal, and high-frequency interference still remains after preliminary noise reduction. The first sudden drop segment (timestamp 10:05:23 - 10:05:25) of the resistive signal is selected to generate a reference template. The local correlation coefficient calculated for the same time period window of the preliminarily noise-reduced signal is 0.82, which is higher than the correlation threshold of 0.75, and the data is retained; while in the non-phase change interference segment from 10:06:15 to 10:06:17, the local correlation coefficient is calculated to be 0.48, triggering the backpressure test. After the valve is closed during the calibration phase, the local correlation coefficient of the noise segment is increased to 0.91, overwriting the original data; the corrected flow velocity of the abnormal segment is adjusted from 1.8 m / s to 2.3 m / s. Using the mutation waveform of the resistive signal as a physical feature template can effectively distinguish real phase changes from random noise, avoiding misjudging droplet impacts as phase interface changes; the backpressure test combined with the dynamic compensation mechanism repairs the measurement deviation caused by bubble accumulation in the fluid stability verification, ensuring the reliability of the preprocessed data.

[0043] Optionally, the multi-step prediction of the edge-side preprocessed data using the time series prediction model preset in the cloud to generate the dynamic compensation coefficient includes: Dividing the edge-side preprocessed data into a high-confidence data set and a low-confidence data set according to the preset time stamps through preset rules; Specifically, the edge-side preprocessed data is arranged in the order of acquisition time, and the time stamp is in the UTC format including year, month, day, hour, minute, and second. The preset rules perform the following operations: for the data at each time stamp, calculate the phase fraction ratio error value between the resistive phase fraction signal and the ultrasonic frequency shift signal, that is, the absolute value of the difference between the oil, gas, and water ratios inverted from the resistive signal and the phase fraction ratio calculated by the Doppler formula of the ultrasonic frequency shift signal. At the same time, calculate the noise energy within this time period, which is the energy integral of the residual high-frequency components in the signal after wavelet noise reduction. Determine whether the phase fraction ratio error value is less than a preset noise threshold such as 0.05, and whether the noise energy is lower than a preset energy threshold such as 10 dBm. If the above conditions are met, the corresponding data is marked as a high-confidence data set, otherwise it is marked as a low-confidence data set.

[0044] Among them, the timestamp is the absolute time identifier at the time of data acquisition, accurate to milliseconds; the noise threshold is the maximum allowable deviation of the phase fraction ratio; the energy threshold is the upper limit of energy for judging signal purity.

[0045] Optimize the compensation parameters of the low-confidence dataset using the reinforcement learning method to generate a compensation parameter increment; Specifically, the deep deterministic policy gradient algorithm is used as the reinforcement learning method. The noise energy, phase fraction error, and flow rate fluctuation in the low-confidence dataset are used as the state space, and the compensation parameter adjustment direction is used as the action space. The objective function is set to minimize the root mean square value of the prediction error in subsequent training. The agent generates a compensation parameter increment through exploration and exploitation policy iteration, and each increment includes an amplitude adjustment coefficient and a phase lag correction amount. For example, in the oil-gas mixture flow scenario, the agent generates an increment with an amplitude coefficient of 0.8 and a phase lag correction amount of 5° according to the fluctuation law dominated by bubble noise. Among them, the reinforcement learning method is a machine learning paradigm for learning the optimal policy by interacting with the environment; the compensation parameter increment is the adjustment amount value of the compensation parameter per unit time.

[0046] Input the compensation parameter increment into the long short-term memory network for training to generate a dynamic compensation coefficient.

[0047] Specifically, the long short-term memory network structure includes an input layer, two hidden layers, and an output layer. The input layer receives the compensation parameter increment in the form of a time series, the number of hidden layer units is 128, and the tanh activation function is used. During training, the high-confidence dataset is used as the ground truth, the low-confidence dataset combined with the compensation parameter increment is used as the input, and the loss function is the mean square error between the predicted flow rate and the actual flow velocity. For the dynamic compensation coefficient at time , there is: ; In the formula, is the non-linear mapping function of the long short-term memory network, is the sequence of compensation parameter increments from time to time , is the state memory of the hidden layer of the long short-term memory network at time . Among them, the long short-term memory network is a time series prediction neural network with a gating mechanism; the dynamic compensation coefficient is a time-varying parameter used to correct the deviation of the flow velocity and the phase fraction ratio.

[0048] Optionally, the division into a high-confidence dataset and a low-confidence dataset by preset rules includes: Calculate the phase fraction error value between the resistive phase fraction signal and the ultrasonic frequency shift signal; Specifically, the phase fraction values of oil, gas, and water are extracted from the resistive phase fraction signal, such as 65% for the oil phase, 20% for the gas phase, and 15% for the water phase; at the same time, the phase fraction is inversely calculated through the Doppler flow velocity formula of the ultrasonic frequency shift signal, such as 60% for the oil phase, 25% for the gas phase, and 15% for the water phase. For the phase fraction error value , there is: ; In the formula, is the number of phase states. For example, oil, gas, and water are three phases, is the phase fraction of the resistive phase fraction signal in the phase, is the phase fraction of the ultrasonic frequency shift signal in the phase. Among them, the resistive phase fraction signal is a phase fraction measurement signal based on the conductivity difference of the fluid; the ultrasonic frequency shift signal is a phase fraction signal calculated by inverting the flow velocity through the Doppler effect; the phase fraction error value is the absolute difference between the measurement results of the same phase state ratio by the two modal signals.

[0049] Judge whether the phase fraction error value is less than the preset noise threshold and the noise energy is lower than the preset energy threshold; Specifically, the noise threshold is the upper limit of the allowable phase fraction error, set according to experimental calibration or historical data statistics, such as 0.05, corresponding to a 5% deviation of the phase fraction; the noise energy is calculated by integrating the energy of the high-frequency components obtained by decomposing the signal through wavelet transform. For the noise energy , there is: , In the formula, is the total number of high-frequency layers after wavelet decomposition, is the m-th high-frequency component after wavelet decomposition, is the time interval integral. The preset energy threshold is 10 dBm. If both ≤ noise threshold and ≤ energy threshold are satisfied, then proceed to the next step. Among them, the noise threshold is an empirically set signal consistency standard; the energy threshold is the allowable upper limit of the high-frequency interference intensity.

[0050] If so, mark the edge-end preprocessed data as a high-confidence data set; If not, mark the edge-end preprocessed data as a low-confidence data set.

[0051] Specifically, for the edge-end preprocessed data within a certain time window, if , less than the noise threshold of 0.05 and , less than the energy threshold of 10 dBm, then mark it as a high-confidence data set and store it in the main partition of the database for model training; if , greater than the noise threshold of 0.05 or If it is greater than the energy threshold of 10 dBm, it is marked as a low-confidence data set and stored in the auxiliary partition for reinforcement learning optimization.

[0052] Optionally, the generating of the dynamic compensation coefficient includes: Constructing a multi-objective loss function based on the high-confidence data set and the compensation parameter increment; Specifically, the high-confidence data set provides sets of sample data (such as phase fraction ratio, flow rate, density), and the compensation parameter increment is the compensation parameter adjustment amount output by reinforcement learning. When constructing the loss function, the main objective is set as the mean square error between the predicted flow rate and the actual flow velocity, and the secondary objective is the stability constraint of the compensation parameter. The multi-objective loss function formula is: ; In the formula, is the total loss value, and are weight coefficients, is the measured flow rate of the th group of high-confidence samples, is the predicted flow rate, is the rd compensation parameter increment. Among them, the multi-objective loss function is a composite evaluation function that simultaneously optimizes the main task accuracy and parameter stability; the compensation parameter increment is a dynamic adjustment coefficient generated by reinforcement learning.

[0053] Taking the multi-objective loss function as the loss function of the long short-term memory network, optimizing the multi-objective loss function by the gradient descent method to obtain the dynamic compensation coefficient.

[0054] Specifically, the long short-term memory network uses the Adam optimizer, and the learning rate is set to 0.001. Each time during training, the input data is the time-series high-confidence data set and the compensation parameter increment, and the network output is the predicted value of the dynamic compensation coefficient. During the gradient descent process, calculate the partial derivative of the total loss value with respect to each layer's weight parameter, and update the network weights by backpropagation. After the final model converges, the average deviation between the compensated predicted flow rate and the measured flow rate is less than 1%.

[0055] Among them, the long short-term memory network is a recurrent neural network with temporal memory ability; the gradient descent method is an optimization algorithm that minimizes the loss function by iteratively adjusting parameters.

[0056] Optionally, the calculating of the multiphase flow rate value according to the dynamic compensation coefficient and the actual flow velocity value includes: Extracting the phase fraction ratio and the frequency shift value from the edge-side preprocessed data; Specifically, time window truncation is performed on the edge - end pre - processed data, and the length of each window is 1 second. The phase fraction ratios of oil, gas, and water are parsed from the resistive phase - split signal. For example, the oil - phase fraction ratio is 65%, the gas - phase fraction ratio is 25%, and the water - phase fraction ratio is 10%. The flow velocity displacement is calculated based on the Doppler frequency - shift formula from the ultrasonic frequency - shift signal and converted into an equivalent frequency - shift value. For the equivalent frequency - shift value , there is: ; In the formula, is the flow velocity displacement, is the ultrasonic reference frequency of 1 MHz. Among them, the phase fraction ratio is the volume proportion of each phase fluid in the total flow rate; the frequency - shift value is the normalized difference between the ultrasonic transmission and reception frequencies caused by the fluid motion.

[0057] The density - weighted integral of the phase fraction ratio is performed using the dynamic compensation coefficient to generate the phase - fraction density value; Specifically, according to the dynamic compensation coefficient each phase fraction ratio is corrected, and the density - weighted integral is calculated in combination with the standard density of each phase. For the phase - fraction density value , there is: ; In the formula, is the standard density of the th phase, such as 850 kg / m³ for crude oil, 0.8 kg / m³ for natural gas, and 1000 kg / m³ for water. Among them, the density - weighted integral is a composite physical quantity combining the phase fraction ratio and the substance density; the dynamic compensation coefficient is a time - varying parameter for correcting the multiphase flow measurement error.

[0058] The flow velocity correction calculation of the actual flow velocity value is performed using the dynamic compensation coefficient and the frequency - shift value to generate the corrected flow velocity value; Specifically, the actual flow velocity value can be measured by a turbine flowmeter, and the frequency - shift value is the equivalent frequency - shift value reflecting the fluid micro - element velocity difference. For the corrected flow velocity value , there is: ; In the formula, is the actual flow velocity value, is the conversion coefficient, calibrated by the probe spacing and the sound speed, such as . Among them, the flow velocity correction calculation is a method for compensating the actual flow velocity measurement deviation through the frequency - shift signal; the conversion coefficient is a fixed correlation factor between the geometric parameters of the ultrasonic measurement system and the sound speed.

[0059] The multiphase flow rate value is calculated based on the phase - fraction density value and the corrected flow velocity value.

[0060] Specifically, the multiphase flow rate value is the total mass flow rate flowing through the pipe cross-section per unit time. For the multiphase flow rate value , there is: ; In the formula, is the pipe cross-sectional area.

[0061] Exemplarily, in a subsea oil pipeline, the edge-end preprocessing data shows that the oil phase is 68%, the gas phase is 10%, and the water phase is 22%. The dynamic compensation coefficient , and the weight needs to be reduced due to bubble interference. The actual flow velocity , and the frequency shift value, i.e., the equivalent frequency shift value , corresponds to the flow fluctuation. The density-weighted integral gives , and the corrected flow velocity is calculated. The final flow rate is the multiphase flow rate value . Compared with the uncompensated situation , , the compensated flow rate is closer to the off-line sampled measured value of 41.2 kg / s, and the error is reduced from 11.6% to 1.2%. By adaptively weakening the double interference of bubbles on density weighting and flow velocity through the dynamic compensation coefficient, overestimating the contribution of the light gas phase is avoided, and the calculated value of the mixed flow rate is more in line with the actual conveying mass, especially suitable for the oil-gas-water multiphase flow scenario with fluctuating gas holdup.

[0062] Optionally, generating an abnormal event warning signal according to the multiphase flow rate value includes: When the fluctuation amplitude of the multiphase flow rate value exceeds the safety threshold, trigger the pipeline backpressure regulation operation; Specifically, the fluctuation amplitude of the multiphase flow rate value is calculated in real time. The sliding time window is set to 3 seconds, and the percentage difference between the maximum and minimum flow rates within the window is calculated. The safety threshold is a preset absolute value (such as 15%). When the instantaneous fluctuation amplitude exceeds this threshold, trigger the backpressure regulation operation. The backpressure regulation controls the opening of the regulating valve at the pipeline end to reduce the valve opening by 40% within 30 seconds, increasing the front-end pressure by 0.5 - 1.2 MPa to suppress the sudden change in flow velocity. For example, when the flow rate is detected to drop suddenly from 50 kg / s to 32 kg / s (fluctuation amplitude 36%), the valve opening is adjusted from 80% to 48%, increasing the pipeline resistance to restore flow stability. Among them, the fluctuation amplitude of the multiphase flow rate value is the relative percentage of the flow rate change per unit time; the pipeline backpressure regulation operation is an emergency measure to balance the flow rate by controlling the valve opening to adjust the pipeline end resistance.

[0063] Synchronously upload the warning log containing the abnormal event type and pipe segment location information to the cloud platform.

[0064] Specifically, the abnormal event types are matched with a preset database according to the flow fluctuation patterns, such as "slug flow mutation", "phase separation blockage", etc. The pipe segment location information is determined through the mapping of pipeline pressure gradient analysis and GPS coordinates. The warning log is uploaded in JSON format, including the timestamp, fluctuation amplitude, regulating valve action parameters, and location coordinates, such as the 15th stake section of the oil pipeline at longitude 118.76°E and latitude 32.04°N. The data is encrypted and transmitted to the cloud storage through the MQTT protocol, triggering the audible and visual alarms and work order dispatching of the operation and maintenance terminal. Among them, the abnormal event types are pre-defined fluid dynamic abnormal classifications; the pipe segment location information is the association identifier of the physical location and the pipeline topological structure.

[0065] Through the closed-loop linkage of flow dynamic monitoring and pressure regulation, this method actively intervenes in the fluid state at the initial stage of abnormal fluctuations and simultaneously constructs an accurate fault traceability information chain. The technical effects are reflected in reducing the risk of equipment damage caused by phase separation or sudden flow changes in multiphase flow transportation through the mechanical control and data collaboration mechanism, while improving the timeliness of abnormal event handling and the location positioning accuracy, and ensuring the safety and economy of pipeline system operation and maintenance.

[0066] Based on the same inventive concept, as Figure 3 shown, the present invention also provides a system, which includes: A data acquisition module, configured to acquire the actual flow velocity value of the fluid in the pipeline and the dual-modal sensing signals, where the dual-modal sensing signals include resistive phase separation signals and ultrasonic frequency shift signals; A signal optimization module, configured to dynamically adjust the sampling frequency of the dual-modal sensing signals based on historical operating condition data to generate optimized sensing sampling signals; A data processing module, configured to perform time-frequency domain joint noise reduction processing on the optimized sensing sampling signals to generate edge-end preprocessed data; A compensation calculation module, configured to perform multi-step prediction on the edge-end preprocessed data by using a time series prediction model preset in the cloud to generate dynamic compensation coefficients; A flow calculation module, configured to calculate the multiphase flow rate value according to the dynamic compensation coefficients and the actual flow velocity value; A flow output and warning module, which outputs the multiphase flow rate value and generates an abnormal event warning signal according to the multiphase flow rate value.

[0067] It should be noted that the electrical connections between the above-mentioned various units do not necessarily represent direct connections of the circuits. Indirect connection methods can be applied to the embodiments of the present invention as long as the purpose of the present invention is achieved. The above are only exemplary embodiments of the present invention and cannot be used to limit the scope of the present invention.

[0068] That is, any equivalent changes and modifications made in accordance with the teachings of the present invention still fall within the scope covered by the present invention. Those skilled in the art will readily conceive of other embodiments of the present invention after considering the specification and the disclosure of the practical truth. This application is intended to cover any variations, uses, or adaptations of the present invention that follow the general principles of the present invention and include the common general knowledge or conventional technical means in the technical field not recorded in the present invention.

Claims

1. A pipeline multiphase flow measurement and monitoring method, characterized in that: The method comprises: Acquire an actual flow velocity value of the fluid in the pipeline and a dual-mode sensing signal, wherein the dual-mode sensing signal includes a resistive phase separation signal and an ultrasonic frequency shift signal; Acquire historical operating condition data, and dynamically adjust the sampling frequency of the dual-modal sensing signal based on the historical operating condition data to generate an optimized sensing sampling signal; Performing a time-frequency domain joint noise reduction process on the optimized sensor sampling signal to generate edge preprocessing data; Using the time series prediction model preset in the cloud to perform multi-step prediction on the edge preprocessed data to generate a dynamic compensation coefficient; The multiphase flow rate value is calculated according to the dynamic compensation coefficient and the actual flow rate value; The multiphase flow rate value is output, and an abnormal event warning signal is generated according to the multiphase flow rate value.

2. The pipeline multiphase flow measurement and monitoring method according to claim 1 is characterized in that: The generating of the optimized sensor sampling signal comprises: Obtain the real-time pressure data, temperature data and historical medium ratio data of the pipeline to obtain the basic data set; generating a flow velocity prediction model of different phases based on the basic data set; Calculating the correlation weight of the resistive phase-division signal and the ultrasonic frequency-shift signal according to the flow velocity prediction model; The sampling frequency of each mode in the dual-modal sensing signal is adjusted based on the association weight to generate an optimized sensing sampling signal.

3. The pipeline multiphase flow measurement and monitoring method according to claim 1 is characterized in that: The performing time-frequency domain joint noise reduction processing on the optimized sensor sampling signal to generate edge preprocessing data includes: Performing wavelet transformation on the ultrasonic frequency shift signal in the sensor sampling signal to filter out frequency domain noise and generate a preliminary noise reduction signal; Combined with the pulse characteristics of the resistive phase division signal, waveform matching is performed on the preliminary noise reduction signal to remove residual noise and generate edge preprocessing data.

4. The pipeline multiphase flow measurement and monitoring method according to claim 3 is characterized in that: The generating edge preprocessing data further comprises: Extracting a sudden change waveform in the resistive phase division signal as a reference template; Calculating a local correlation coefficient between the preliminary noise reduction signal and the reference template; When the local correlation coefficient is lower than a preset correlation threshold, the edge preprocessing data is corrected through a back pressure simulation test.

5. The pipeline multiphase flow measurement and monitoring method according to claim 1 is characterized in that: The step of using a preset time series prediction model on the cloud to perform multi-step prediction on the edge preprocessed data to generate a dynamic compensation coefficient includes: Dividing the edge preprocessed data into a high-confidence data set and a low-confidence data set according to a preset timestamp and a preset rule; Optimizing the compensation parameters of the low-confidence data set using a reinforcement learning method to generate compensation parameter increments; The compensation parameter increments are input into a long short-term memory network for training to generate a dynamic compensation coefficient.

6. The pipeline multiphase flow measurement and monitoring method according to claim 5 is characterized in that: The dividing into a high-confidence data set and a low-confidence data set according to a preset rule includes: Calculating a phase-division proportional error value between the resistive phase-division signal and the ultrasonic frequency-shift signal; Determining whether the phase-division ratio error value is less than a preset noise threshold and the noise energy is lower than a preset energy threshold; If yes, marking the edge preprocessed data as a high confidence data set; If not, the edge preprocessed data is marked as a low confidence data set.

7. The pipeline multiphase flow measurement and monitoring method according to claim 5 is characterized in that: Generating the dynamic compensation coefficient comprises: Constructing a multi-objective loss function based on the high confidence data set and the compensation parameter increment; The multi-objective loss function is used as the loss function of the long short-term memory network, and the multi-objective loss function is optimized by the gradient descent method to obtain a dynamic compensation coefficient.

8. The pipeline multiphase flow measurement and monitoring method according to claim 7 is characterized in that: The multiphase flow rate value calculated according to the dynamic compensation coefficient and the actual flow rate value comprises: Extracting phase ratio and frequency shift value from the edge preprocessing data; Performing density-weighted integration on the phase fraction ratio using the dynamic compensation coefficient to generate a phase fraction density value; Performing flow velocity correction calculation on the actual flow velocity value using the dynamic compensation coefficient and the frequency shift value to generate a corrected flow velocity value; The multiphase flow rate value is calculated based on the phase density value and the corrected flow rate value.

9. The pipeline multiphase flow measurement and monitoring method according to claim 1 is characterized in that: Generating an abnormal event warning signal according to the multiphase flow rate value comprises: When the fluctuation amplitude of the multiphase flow value exceeds a safety threshold, a pipeline back pressure adjustment operation is triggered; Synchronously upload the warning log containing abnormal event type and pipe section location information to the cloud platform.

10. A pipeline multiphase flow measurement and monitoring system, applied to the pipeline multiphase flow measurement and monitoring method according to any one of claims 1 to 9, characterized in that: The system comprises: A data acquisition module, used to acquire the actual flow velocity value of the fluid in the pipeline and a dual-mode sensing signal, wherein the dual-mode sensing signal includes a resistive phase separation signal and an ultrasonic frequency shift signal; A signal optimization module, used for dynamically adjusting the sampling frequency of the dual-mode sensor signal based on historical operating condition data to generate an optimized sensor sampling signal; A data processing module, used for performing a time-frequency domain joint noise reduction process on the optimized sensor sampling signal to generate edge preprocessing data; A compensation calculation module, used to perform multi-step prediction on the edge preprocessed data using a time series prediction model preset in the cloud to generate a dynamic compensation coefficient; A flow calculation module, used for calculating the multiphase flow value according to the dynamic compensation coefficient and the actual flow velocity value; The flow output and warning module outputs the multiphase flow value and generates an abnormal event warning signal according to the multiphase flow value.

Images