Gathering and transportation pipe network steam driving online prediction method

Abstract

The invention provides a gathering and transportation pipe network steam driving on-line prediction method, which comprises the following steps: acquiring steady-state and dynamic parameters from a wellhead to a united station in a gathering and transportation system, and calculating energy loss in a pipeline transportation process at a corresponding moment; standardizing the data set in the gathering and transportation process of the oil and gas field to obtain a standardized data set; determining a standardized process parameter set after feature extraction by adopting a grey correlation analysis method; performing working condition classification on the standardized process parameter set after feature extraction by adopting a K-zero near-value classifier to obtain various working condition data sets; establishing PSO-LSTM prediction models of various working conditions for various working condition data sets, and optimizing the PSO-LSTM prediction models to obtain the final optimized PSO-LSTM prediction models of gathering and transportation energy consumption of various working conditions; and predicting the energy consumption of the transported crude oil by adopting the finally optimized PSO-LSTM prediction model of the gathering and transportation energy consumption (pressure and thermal energy consumption) of various working conditions. According to the method, high-precision data-driven modeling is realized according to related variables to obtain real-time pressure and thermal loss from the production wellhead to the combination station.

Description

Technical Field

[0001] The present invention belongs to the technical field of oilfield gathering and transportation control, and particularly relates to an online prediction method. Background Art

[0002] The petroleum industry is an important pillar industry in China and is the power raw material for industries such as manufacturing and transportation. As the core of the surface oil and gas production project, the oil and gas gathering and transportation system undertakes important tasks such as exploitation, transportation, processing, and treatment. The oil and gas gathering and transportation system is composed of gathering and transportation pipelines and related stations, and its function is to collect and process the well products to meet the quality requirements of users. The technological level and energy consumption level of the gathering and transportation system not only determine the operation stability of the entire oilfield, but also determine the economic efficiency of the oilfield operation. Conducting research on the optimization and visualization of the oil and gas gathering and transportation system enables enterprises to directly achieve energy conservation and consumption reduction, reduce costs, and enhance the market competitiveness of enterprises.

[0003] The operation process mechanism of the oilfield gathering and transportation system is complex and is affected by multiple factors such as hydraulics, thermodynamics, high noise, and nonlinearity. Moreover, each system has its corresponding mechanism, and it is difficult to establish an accurate mathematical model for the mutual coupling relationship between systems. These characteristics result in the difficulty of accurately measuring the gathering and transportation energy consumption online.

[0004] Currently, for the steam drive prediction method of the gathering and transportation pipeline network, it is mainly based on experience to judge whether it is necessary to steam drive the crude oil of the oil well. This method cannot clearly judge whether heating is required, resulting in energy waste. Summary of the Invention

[0005] Aiming at the problems that the existing technology cannot establish a data-driven model of the main processes of the system based on the operation history data of the gathering and transportation system to realize the prediction of system pressure and thermal energy consumption, nor can it use the mechanism model of the on-site operation process of the system as a reference for identification to dynamically correct the data-driven model to realize the adaptive prediction control of some processes, the present invention provides an online prediction method.

[0006] To solve the above technical problems, the technical solution adopted by the present invention is: An online prediction method for steam drive of a gathering and transportation pipeline network, comprising the following steps:

[0007] Step 1: Obtain the steady-state parameters and dynamic parameters from the wellhead to the central processing facility in the gathering and transportation system, calculate the energy loss during the pipeline transportation process at the corresponding moment, and obtain the data set of the oil and gas field gathering and transportation process. Obtain the steady-state parameters and dynamic parameters from the wellhead to the central processing facility in the gathering and transportation system through the SCADA system of the oilfield central processing facility, and calculate the energy loss during the pipeline transportation process at the corresponding moment according to the mechanism model to obtain the data set of the oil and gas field gathering and transportation process DATA = {X k , T k | k = 1,..., K}, where X k ∈ Rn are the steady-state parameters and dynamic parameters in the k-th sample, T k is the energy loss during the pipeline transportation process of the k-th sample, K is the number of samples in the data set, n is the dimension of the dynamic parameters and steady-state parameters, and R is the data set space;

[0008] Step 2: Standardize the data set of the oil and gas field gathering and transportation process to obtain the standardized data set. By using a transformation function to standardize the dynamic parameters and steady-state parameters of the data set of the oil and gas field gathering and transportation process, the standardized data set SDATA = {S k , T k | k = 1,..., K}, where S k ∈R n are the standardized dynamic parameters and steady-state parameters in the k-th sample;

[0009] Step 3: Determine the importance of the standardized process parameters, extract the process parameters with importance greater than the importance threshold, and obtain the set of standardized process parameters after feature extraction. By using the grey relational analysis method to determine the importance of the standardized process parameters, extract the process parameters with importance greater than the importance threshold, and obtain the set of standardized process parameters after feature extraction CDATA = {C k , T k | k = 1,…, K}, where, are the standardized process parameters in the k-th sample after feature selection, and n c is the dimension of the standardized process parameters after feature extraction;

[0010] Step 4: Classify the set of standardized process parameters after feature extraction to obtain data sets for various working conditions. By using a K-nearest neighbor classifier to classify the set of standardized process parameters CDATA after feature extraction, data sets for various working conditions LDATA i = {L p(i) , T p(i) | p(i) = 1,…, P(i)} are obtained, and the data sets for various working conditions LDATA i are divided into a training data set xLDATA i = {L h(i) , T h(i) | h(i) = 1,…, H(i)} and a test data set cLDATA i = {L e(i) , T e(i) | e(i) = 1,…, E(i)}, where i = 1, 2,…, I, I is the number of working conditions, are the process parameters in the p-th sample of the i-th working condition, and T p(i)Let \(E_{ip}\) be the gathering and transportation energy consumption in the \(p\)-th sample of the \(i\)-th working condition, \(P(i)\) be the number of samples in the data set of the \(i\)-th working condition, \(h(i)\) be the number of the \(h\)-th sample in the training data set of the \(i\)-th working condition, \(H(i)\) be the number of samples in the training data set of the \(i\)-th working condition, \(e(i)\) be the number of the \(e\)-th sample in the test data set of the \(i\)-th working condition, and \(E(i)\) be the number of samples in the test data set of the \(i\)-th working condition; Let \(T_{ih}\) be the process parameter in the \(h\)-th sample of the \(i\)-th working condition in the training data set, \(T\) h(i) Let \(E_{ih}\) be the gathering and transportation energy consumption in the \(h\)-th sample of the \(i\)-th working condition in the training data set, Let \(T_{ie}\) be the process parameter in the \(e\)-th sample of the \(i\)-th working condition in the test data set, \(T\) e(i) Let \(E_{ie}\) be the gathering and transportation energy consumption in the \(e\)-th sample of the \(i\)-th working condition in the test data set;

[0011] Step 5: For each type of working condition data set, take the training data set \(xLDATA\) of each type of working condition i as the input, establish the PSO-LSTM prediction model for each type of working condition, use the particle swarm optimization algorithm to optimize the PSO-LSTM prediction model for each type of working condition, and use the test data set \(cLDATA\) i to test the optimized PSO-LSTM prediction model, and obtain the finally optimized PSO-LSTM prediction model for the gathering and transportation energy consumption of each type of working condition;

[0012] Step 6: Save the finally optimized PSO-LSTM prediction model for the gathering and transportation energy consumption of each type of working condition;

[0013] Step 7: Read the online measurement values of the process parameters of the gathering and transportation pipeline in real time, and use the finally optimized PSO-LSTM prediction model for the gathering and transportation energy consumption of each type of working condition to predict the energy consumption of transporting crude oil.

[0014] In some embodiments, the process parameters of the oil and gas field gathering and transportation process include: the back pressure at the wellhead per unit time, the wellhead temperature, the wellhead flow rate, the temperature at the combined station, the combined station pressure, the total gathering and transportation flow rate, and the air temperature.

[0015] In some embodiments, the calculation formula for standardizing the dynamic parameters and steady-state parameters of the data set of the oil and gas field gathering and transportation process using the transfer function is as follows:

[0016]

[0017] In some embodiments, for each type of working condition data set, take the training data set \(xLDATA\) of each type of working condition i as the input, establish the PSO-LSTM prediction model for each type of working condition, use the particle swarm optimization algorithm to optimize the PSO-LSTM prediction model for each type of working condition, and use the test data set \(cLDATA\)i Test the optimized PSO-LSTM prediction model to obtain the PSO-LSTM prediction model of the gathering and transportation energy consumption under various optimized working conditions, and execute according to the following steps:

[0018] Step 5-1: Initialize the parameters of the particle swarm optimization algorithm;

[0019] The parameters of the particle swarm optimization algorithm include: the particle population size M, the maximum number of iterations G, the inertia weight W, the learning factors C1 = C2 ∈ [0, 4], and the initialized particle swarm optimization algorithm population L a , where a = 1, 2,..., M;

[0020] Step 5-2: Obtain the PSO-LSTM parameter values according to the information of each individual in the population to get M groups of PSO-LSTM parameter values;

[0021] Step 5-3: Take the radial basis function as the PSO-LSTM kernel function, establish a PSO-LSTM prediction model for the PSO-LSTM parameter values corresponding to each individual in the population and the training data set of the i-th type of working condition, and train to obtain M PSO-LSTM prediction models;

[0022] The PSO-LSTM model of the i-th type of working condition is as follows:

[0023]

[0024] where h(i) = 1, 2,..., H(i), H(i) is the number of training samples in the training data set of the i-th type of working condition, L(i) is the new sample data for which the gathering and transportation energy consumption needs to be calculated for the i-th type of working condition, is the output value of the PSO-LSTM prediction model corresponding to the input data L(i) of the i-th type of working condition, ah(i), b(i) are the PSO-LSTM prediction model parameters of the i-th type of working condition, L h(i) is the process parameter of the h-th training sample in the training data set of the i-th type of working condition, and K(L(i), Lh(i)) is the kernel function of the PSO-LSTM prediction model of the i-th type of working condition;

[0025] The kernel function K(L(i), Lh(i)) of the PSO-LSTM prediction model of the i-th type of working condition is as follows:

[0026]

[0027] where, is the width of the PSO-LSTM kernel function of the i-th type of working condition;

[0028] Step 5-4: Input the test data of the i-th type of working condition into the PSO-LSTM model of the i-th type of working condition established by each individual in the population, and calculate the root mean square error value ε of the PSO-LSTM model of the i-th type of working condition established by each individual in the population α , and the fitness function value f of each individual in the population α = ε α ;

[0029] The root mean square error value ε of the PSO-LSTM model of the i-th type of working condition established by each individual in the population α is calculated as follows:

[0030]

[0031] where is the output value of the PSO-LSTM prediction model of the process parameters in the e-th sample in the test data set of the i-th type of working condition calculated using the measurement model corresponding to the α-th individual;

[0032] Step 5-5: Determine whether the current iteration number g reaches the maximum iteration number G. If so, the iteration ends, and the optimal PSO-LSTM parameters and model parameters ah(i), b(i) are obtained to get the finally optimized PSO-LSTM prediction model for the gathering and transportation energy consumption of the i-th type of working condition. Otherwise, execute Step 5-5;

[0033] Step 5-6: Update the population in the g-th iteration, let the iteration number g = g + 1, and return to Step 5-2.

[0034] In some embodiments, Step 5-6 includes the following steps:

[0035] Step 5-6-1: For the a-th individual L in the current iteration a，g , set the pbest coordinate of L a，g as the initial position of the current position, and calculate the fitness value f(Xi) of the individual extreme point;

[0036] The calculation formula of the fitness value f(Xi) of the individual extreme point is as follows:

[0037]

[0038] where M represents the population size, Y i is the sample output value, and y i is the actual output value;

[0039] Step 5-6-2: Evaluate each particle, calculate the fitness value of the particle. If it is better than the current individual extreme of the particle, set pbest as the position of the particle and update the individual extreme;

[0040] Step 5-6-3: Update the particle position and velocity according to the fitness value f(Xi) to obtain new particles as the particles of the (g+1)-th iteration population L a，g+1 ;

[0041] The position and velocity of the particles of the (g+1)-th iteration population L a，g+1 are calculated according to the following formulas:

[0042]

[0043] where i = 1, 2,..., M, M is the total number of particles, t is the current iteration number, W is the inertia weight, vi is the particle velocity, r1 and r2 are random numbers, Xi is the current position of the particle, is the current individual optimal position of particle i, is the current optimal position of the particle swarm, and C1 and C2 are learning factors;

[0044] Step 5-6-4: Let the iteration generation g = g + 1, and return to Step 5-2.

[0045] In some embodiments, the online measurement values of the process parameters of the gathering and transportation pipeline network are read in real time, and the PSO-LSTM prediction model of the gathering and transportation energy consumption under various optimized working conditions is used to predict the energy consumption of transporting crude oil, which is executed according to the following steps:

[0046] Step 7-1: Read the online measurement values of the process parameters of the gathering and transportation pipeline network in real time, and use the conversion function to standardize the dynamic and steady-state parameters of the data set of the oil and gas field gathering and transportation process to obtain the standardized process parameters

[0047] Step 7-2: Classify according to the distances from the standardized process parameters to the clustering centers of various working conditions, and input them into the optimized PSO-LSTM prediction model of the gathering and transportation energy consumption for determining the working condition category to obtain the online measurement values of the gathering and transportation energy consumption.

[0048] The present invention proposes an online prediction method for a gathering and transportation pipeline network under the condition of steam mixed transportation. This method can realize high-precision data-driven modeling based on relevant variables to obtain the pressure and thermal loss from the real-time production wellhead to the joint station, which can meet the production requirements; it has a wide range of applications, can be applied to the vector expression of different data sources, and has good adaptability; it can conveniently share information with other links and is convenient to provide reference information for the operations of other links. BRIEF DESCRIPTION OF THE DRAWINGS

[0049] Figure 1 is a flowchart of the online prediction method for steam drive of a gathering and transportation pipeline network in a specific embodiment of the present invention.

[0050] Figure 2It is a process flow chart for optimizing the PSO-LSTM prediction models for various working conditions using the particle swarm optimization algorithm in a specific embodiment of the present invention.

[0051] Figure 3 It is a pressure follower diagram of the gathering and transportation energy consumption of each wellhead varying with the 52nd wellhead predicted by the PSO-LSTM model in a specific embodiment of the present invention.

[0052] Figure 4 It is a temperature follower diagram of the gathering and transportation energy consumption of each wellhead varying with the 52nd wellhead predicted by the PSO-LSTM model in a specific embodiment of the present invention.

[0053] Figure 5 It is a temperature follower diagram of the gathering and transportation energy consumption of each wellhead varying with the 57th wellhead predicted by the PSO-LSTM model in a specific embodiment of the present invention.

[0054] Figure 6 It is a pressure follower diagram of the gathering and transportation energy consumption of each wellhead varying with the 57th wellhead predicted by the PSO-LSTM model in a specific embodiment of the present invention. Specific Embodiment

[0055] To facilitate a further understanding of the method and achieved effects of the present invention, the following is a detailed description with reference to the attached drawing examples. However, the attached drawings are only for reference and illustration purposes and are not used to limit the present invention.

[0056] A gathering and transportation pipeline network steam drive online prediction method in a specific embodiment of the present invention, as Figure 1 shown, includes the following steps:

[0057] Step 1: Obtain the steady-state parameters and dynamic parameters from the wellhead to the joint station in the gathering and transportation system through the SCADA system of the oilfield joint station, and calculate the energy loss during the pipeline transportation process at the corresponding moment according to the existing energy consumption mechanism model of the gathering and transportation pipeline network, to obtain the data set DATA of the oil and gas field gathering and transportation process = {X k , T k |k = 1,..., K}, where X k ∈ R n is the steady-state parameters and dynamic parameters in the kth sample, T k is the energy loss during the pipeline transportation process of the kth sample, K is the number of samples in the data set, n is the dimension of the dynamic parameters and steady-state parameters, and R is the data set space.

[0058] In this embodiment, the process parameters of 43 initial oil and gas field gathering and transportation processes include: back pressure at the wellhead per unit time (13 measurement points), wellhead temperature (13 measurement points), wellhead flow rate (13 measurement points), temperature at the combined station (1 measurement point), pressure at the combined station (1 measurement point), total gathering and transportation flow rate (1 measurement point), and air temperature (1 measurement point).

[0059] In this embodiment, the data set DATA of the oil and gas field gathering and transportation process is {X k , T k | k = 1,..., 2605}. The number of samples K in the data set is 2605, and the process parameter dimension n is 43.

[0060] Table 1 Details of the data set of the oil and gas field gathering and transportation process

[0061] Input dimension Output dimension Sample interval Number of training samples Number of test samples 43 3 24h 2410 195

[0062] Step 2: Use the conversion function to standardize the dynamic and steady-state parameters of the data set of the oil and gas field gathering and transportation process, and obtain the standardized data set SDATA = {S k , T k | k = 1,..., K}, where S k ∈ R n is the standardized dynamic and steady-state parameters in the k-th sample.

[0063] In this embodiment, the calculation formula for standardizing the dynamic and steady-state parameters of the data set of the oil and gas field gathering and transportation process using the standardization conversion function is shown in Equation (1):

[0064]

[0065] Step 3: Use the grey relational analysis method of existing multi-factor statistical analysis to determine the importance of the standardized process parameters, extract the process parameters with importance greater than the importance threshold γ, and obtain the standardized process parameter set CDATA = {C k , T k | k = 1,..., K}, where is the standardized process parameter in the k-th sample after feature selection, and n c is the dimension of the standardized process parameters after feature extraction.

[0066] In this embodiment, the importance threshold γ = 0.1. Use the grey relational analysis method to determine the importance of the standardized process parameters, and obtain the standardized process parameter set after feature extraction, as shown in Figure 2. The dimension n c of the standardized process parameters after feature extraction is 14.

[0067] Table 2 Standardization process parameters after feature extraction

[0068]

[0069]

[0070] Step 4: Use the existing classification algorithm K-nearest neighbor classifier to classify the standardized process parameter set CDATA after feature extraction to obtain various working condition data sets LDATA i ={L p(i) , T p(i) |p(i)=1,..., P(i)}, and divide each working condition data set LDATA i into a training data set xLDATA i ={L h(i) , T h(i) |h(i)=1,..., H(i)} and a test data set cLDATA i ={L e(i) , T e(i) |e(i)=1,..., E(i)}, where i = 1, 2,..., I, and I is the number of working conditions, is the process parameter in the p-th sample of the i-th working condition, T p(i) is the gathering and transportation energy consumption in the p-th sample of the i-th working condition, P(i) is the number of samples in the data set of the i-th working condition, h(i) is the h-th sample number in the training data set of the i-th working condition, H(i) is the number of samples in the training data set of the i-th working condition, e(i) is the e-th sample number in the test data set of the i-th working condition, and E(i) is the number of samples in the test data set of the i-th working condition; is the process parameter in the h-th sample of the i-th working condition in the training data set, Th(i) is the gathering and transportation energy consumption in the h-th sample of the i-th working condition in the training data set, is the process parameter in the e-th sample of the i-th working condition in the test data set, T e(i) is the gathering and transportation energy consumption in the e-th sample of the i-th working condition in the test data set.

[0071] In this embodiment, the standardized process parameter set after feature extraction is classified into I = 13 categories, and the training data sets and test data sets of each working condition are shown in Table 3.

[0072] Table 3 Training data sets and test data sets of each working condition

[0073] Input dimension Output dimension Number of samples in the training set Number of samples in the test set The first type of working condition 12 3 189 15 The second type of working condition 12 3 181 15 The third type of working condition 12 3 186 15 The fourth type of working condition 12 3 186 15 The fifth type of working condition 12 3 179 15 The sixth type of working condition 12 3 183 15 The seventh type of working condition 12 3 189 15 The eighth type of working condition 12 3 204 15 The ninth type of working condition 12 3 185 15 The tenth type of working condition 12 3 171 15 The eleventh type of working condition 12 3 191 15 The twelfth type of working condition 12 3 177 15 The thirteenth type of working condition 12 3 189 15

[0074] Step 5: For each type of working condition dataset, use the training dataset xLDATAi of each type of working condition as the input to establish a PSO-LSTM prediction model for each type of working condition. Use the existing particle swarm optimization algorithm, a search algorithm based on group collaboration, to optimize the PSO-LSTM prediction model for each type of working condition, and use the test dataset cLDATA i Test the optimized PSO-LSTM prediction model to obtain the finally optimized PSO-LSTM prediction model for the gathering and transportation energy consumption of each type of working condition.

[0075] In this embodiment, the specific process of using the particle swarm optimization algorithm to optimize the PSO-LSTM prediction model for each type of working condition is as Figure 2 shown.

[0076] Step 5-1: Initialize the parameters of the particle swarm optimization algorithm.

[0077] The parameters of the particle swarm optimization algorithm include: the size M of the particle population, the maximum number of iterations G, the inertia weight W, the learning factors C1 = C2 ∈ [0, 4], and the initialized particle swarm optimization algorithm population L a , where a = 1, 2,..., M.

[0078] In this embodiment, the parameters of the particle swarm optimization algorithm include: the size M of the particle population = 40, the maximum number of iterations G = 500, the inertia weight W = 0.8, the learning factors C1 = C2 = 2.0, and the initialized particle swarm optimization algorithm population L a , where a = 1, 2,... 40.

[0079] Step 5-2: Obtain the PSO-LSTM parameter values according to the information of each individual in the population to get M groups of PSO-LSTM parameter values.

[0080] Step 5-3: Use the radial basis function as the PSO-LSTM kernel function, and for the PSO-LSTM parameter values corresponding to each individual in the population and the PSO-LSTM prediction model established based on the training dataset of the i-th type of working condition, train to obtain M PSO-LSTM prediction models.

[0081] The PSO-LSTM model of the i-th type of working condition is shown in Equation (2):

[0082]

[0083] where h(i) = 1, 2,..., H(i), and H(i) is the number of training samples in the training dataset of the i-th type of working condition, is the new sample data for calculating gathering and transportation energy consumption under the i-th working condition, which is the output value of the PSO-LSTM prediction model corresponding to the input data L(i) of the i-th working condition. ah(i) and b(i) are the PSO-LSTM prediction model parameters of the i-th working condition, and L h(i) is the process parameter of the h-th training sample in the training dataset of the i-th working condition, and K(L(i), Lh(i)) is the kernel function of the PSO-LSTM prediction model of the i-th working condition;

[0084] The kernel function K(L(i), Lh(i)) of the PSO-LSTM prediction model of the i-th working condition is shown in Equation (3):

[0085]

[0086] where is the width of the PSO-LSTM kernel function of the i-th working condition.

[0087] In this embodiment, taking the thirteenth working condition as an example, the specific process is as Figure 2 shown. For the PSO-LSTM prediction model established from the training dataset of the thirteenth working condition, 80 PSO-LSTM prediction models are trained.

[0088] The PSO-LSTM model of the thirteenth working condition is shown in Equation (4):

[0089]

[0090] where h(13) = 1, 2,..., H(13), H(13) is the number of training samples in the training dataset of the thirteenth working condition with a value of 189, L(13) is the new sample data for calculating gathering and transportation energy consumption under the thirteenth working condition, is the output value of the PSO-LSTM prediction model corresponding to the input sample data L (13) for calculating gathering and transportation energy consumption, ah(13) and b(13) are the PSO-LSTM prediction model parameters of the thirteenth working condition, and Lh(13) is the process parameter of the h-th training sample in the training dataset of the thirteenth working condition, and K(L (i) , L h(i) ) is the kernel function of the PSO-LSTM prediction model.

[0091] In this embodiment, the kernel function K(L (13) , Lh (13) ) of the PSO-LSTM prediction model of the thirteenth working condition is shown in Equation (5):

[0092]

[0093] where Width of the PSO-LSTM kernel function for the 13th operating condition.

[0094] Step 5-4: Input the test data of the i-th operating condition into the PSO-LSTM model of the i-th operating condition established by each individual in the population, and calculate the root mean square error value ε of the PSO-LSTM model of the i-th operating condition established by each individual in the population α , that is, the fitness function value f of each individual in the population α = ε α .

[0095] The root mean square error value ε of the PSO-LSTM model of the i-th operating condition established by each individual in the population α The calculation formula is shown in Equation (6):

[0096]

[0097] where is the output value of the PSO-LSTM prediction model of the process parameters in the e-th sample in the test data set of the i-th operating condition calculated by using the measurement model corresponding to the α-th individual.

[0098] In this embodiment, taking the 13th operating condition as an example, the root mean square error value ε of the PSO-LSTM model of the 13th operating condition established by each individual in the population α The calculation formula is shown in Equation (7):

[0099]

[0100] where is the output value of the PSO-LSTM prediction model of the process parameters in the e-th sample in the test data set of the 13th operating condition calculated by using the measurement model corresponding to the α-th individual.

[0101] Step 5-5: Determine whether the current iteration number g reaches the maximum iteration algebra 500. If so, the iteration ends, and the optimal PSO-LSTM parameters and model parameters ah(13), b(13) are obtained to get the finally optimized PSO-LSTM prediction model for the gathering and transportation energy consumption of the 13th operating condition. Otherwise, execute Step 5-6.

[0102] Step 5-6: Update the population in the g-th iteration, let the iteration algebra g = g + 1, and return to Step 5-2.

[0103] The said Step 5-6 includes the following steps:

[0104] Step 5-6-1: For the a-th individual L in the current iteration a，g , take L a，gThe pbest coordinate is set to the initial position of the current position, and its corresponding individual extreme value is calculated, that is, the fitness value f(Xi) of the individual extreme value point.

[0105] The calculation formula of the fitness value f(Xi) of the individual extreme value point is shown in Equation (8):

[0106]

[0107] Among them, M represents the population size, Y i is the sample output value, y i is the actual output value.

[0108] Step 5-6-2: Evaluate each particle, calculate the fitness value of the particle. If it is better than the current individual extreme value of the particle, set pbest to the position of the particle and update the individual extreme value.

[0109] Step 5-6-3: Update the particle position and velocity according to the fitness value f(Xi) to obtain a new particle as the population particle L of the (g + 1)-th iteration a，g+1 , which is the next-generation particle.

[0110] The calculation formula of the position and velocity of the population particle L of the (g + 1)-th iteration is shown in Equation (9): a，g+1 is shown in Equation (9):

[0111]

[0112] Among them, i = 1, 2,..., M, M is the total number of particles, t is the current iteration number, W is the inertia weight, vi is the particle velocity, r1, r2 are random numbers, Xi is the current individual optimal position of particle i, is the global optimal position of the particle swarm, C1, C2 are learning factors, usually taken as 2.

[0113] Step 5-6-4: Let the iteration generation g = g + 1, and return to Step 5-2.

[0114] Step 6: Save the PSO-LSTM prediction model of the gathering and transportation energy consumption for various working conditions after final optimization.

[0115] Step 7: Read the online measurement values of the process parameters of the gathering and transportation pipeline network in real time, and use the PSO-LSTM prediction model of the gathering and transportation energy consumption (pressure and thermal energy consumption) for various working conditions to predict the energy consumption of transporting crude oil.

[0116] In this embodiment, the sample data example is shown in Table 4.

[0117] Table 4 Sample Data Example

[0118]

[0119] Step 7-1: Read the online measurement values of the process parameters of the gathering and transportation pipeline network in real time, and use a standardized conversion function to standardize the dynamic and steady-state parameters of the data set in the oil and gas field gathering and transportation process to obtain the standardized process parameter C 0 , and the sample data example is shown in Table 4

[0120] Step 7-2: Classify according to the distance between the standardized process parameter C 0 and the clustering centers of various working conditions, and input it into the optimized PSO-LSTM prediction model for the gathering and transportation energy consumption to determine the working condition category, and obtain the online measurement value of the gathering and transportation energy consumption

[0121] In this embodiment, taking the sample data example. Classify according to the distance between the standardized process parameter value and the clustering centers of various working conditions, and select the category with the smallest non-similarity index between vector objects in the sample through the majority voting method as the category of the corresponding test sample production data. That is, the non-similarity index of each variable in the sample data is the smallest at the second working condition

[0122] Table 5 Distance between the standardized process parameter and the clustering centers of various working conditions

[0123]

[0124]

[0125] The statistical analysis results of the pressure, temperature modeling and prediction errors of the steam drive online prediction method (PSO-LSTM) for the gathering and transportation pipeline network proposed by the present invention are shown in Table 6. The gathering and transportation energy consumption, pressure, and temperature model prediction results obtained by modeling and prediction with the PSO-LSTM model are compared with the real values as Figures 3 to 6 shown

[0126] Table 6 Statistical analysis of the results of the predicted values and measured values of pressure and temperature

[0127]

[0128] The above-described embodiments are only preferred embodiments given to fully illustrate the present invention, and the protection scope of the present invention is not limited thereto. Equivalent substitutions or transformations made by those skilled in the art on the basis of the present invention are all within the protection scope of the present invention

Claims

1. A method for online prediction of steam drive in a gathering and transportation pipeline network, characterized in that: The following steps are involved: Step 1, obtain the steady-state parameters and dynamic parameters from the wellhead to the joint station in the gathering and transportation system, calculate the energy loss of the pipeline transportation process at the corresponding time, and obtain the data set of the oil and gas field gathering and transportation process: obtain the steady-state parameters and dynamic parameters from the wellhead to the joint station in the gathering and transportation system through the oil field joint station SCADA system, and calculate the energy loss of the pipeline transportation process at the corresponding time according to the mechanism model, and obtain the data set of the oil and gas field gathering and transportation process DATA = {X k ,T k |k=1,…,K}, where X k ∈R n are the steady-state parameters and dynamic parameters in the kth sample, T k is the energy loss of the pipeline transportation process of the kth sample, K is the number of samples in the data set, n is the dimension of dynamic parameters and steady-state parameters, and R is the data set space; Step 2, standardize the data set of the oil and gas field gathering and transportation process to obtain a standardized data set: standardize the dynamic parameters and steady-state parameters of the data set of the oil and gas field gathering and transportation process by using a conversion function to obtain a standardized data set SDATA = {S k ,T k |k=1,…,K}, where S k ∈R n are the normalized dynamic parameters and steady-state parameters in the kth sample; Step 3, determining the importance of the standardized process parameters, extracting process parameters whose importance is greater than the importance threshold, and obtaining a standardized process parameter set after feature extraction: Determine the importance of the standardized process parameters by using the gray correlation analysis method, extract the process parameters whose importance is greater than the importance threshold, and obtain a standardized process parameter set after feature extraction CDATA = {C k ,T k |k=1,…,K}, where is the standardized process parameter in the kth sample after feature selection, n c is the dimension of the standardized process parameters after feature extraction; Step 4, classifying the working conditions of the standardized process parameter set after feature extraction to obtain various working condition data sets: Classifying the working conditions of the standardized process parameter set CDATA after feature extraction by using a K-nearest neighbor classifier to obtain various working condition data sets LDATA i ={L p(i) , T p(i) |p(i)=1,…,P(i)}, and various working condition data sets LDATA i Divided into training data set xLDATA i ={L h(i) , T h(i) |h(i)=1,…,H(i)} and the test data set cLDATA i ={L e(i) , T e(i) |e(i)=1,...,E(i)}, where i=1,2,...,I, I is the number of working conditions, is the process parameter in the pth sample of the i-th operating condition, T p(i) is the gathering and transmission energy consumption in the pth sample of the i-th working condition, P(i) is the number of samples in the i-th working condition data set, h(i) is the number of the hth sample in the i-th working condition training data set, H(i) is the number of samples in the i-th working condition training data set, e(i) is the number of the eth sample in the i-th working condition test data set, and E(i) is the number of samples in the i-th working condition test data set; is the process parameter of the hth sample of the i-th condition in the training data set, T h(i) is the transmission energy consumption in the hth sample of the i-th operating condition in the training data set, is the process parameter of the e-th sample of the i-th condition in the test data set, T e(i) is the gathering and transmission energy consumption in the e-th sample of the i-th operating condition in the test data set; Step 5: For each type of working condition data set, convert each type of working condition training data set xLDATA i As input, a PSO-LSTM prediction model for various working conditions is established, and the particle swarm optimization algorithm is used to optimize the PSO-LSTM prediction model for various working conditions. The test data set cLDATA is used i The optimized PSO-LSTM prediction model was tested to obtain the final optimized PSO-LSTM prediction model for gathering and transmission energy consumption under various working conditions; Step 6, save the final optimized PSO-LSTM prediction model of the gathering and transmission energy consumption of various working conditions; Step 7: Read the online measurement values ​​of the process parameters of the gathering and transportation pipeline network in real time, and use the PSO-LSTM prediction model of gathering and transportation energy consumption under various working conditions after the final optimization to predict the energy consumption of transporting crude oil.

2. The method for online prediction of steam drive of a gathering and transportation pipeline network according to claim 1, characterized in that: The process parameters of the oil and gas field gathering and transportation process include: wellhead back pressure per unit time, wellhead temperature, wellhead flow, temperature of the joint station, pressure of the joint station, total gathering and transportation flow, and air temperature.

3. The method for online prediction of steam drive of a gathering and transportation pipeline network according to claim 1, characterized in that: The calculation formula for standardizing the dynamic parameters and steady-state parameters of the data set of the oil and gas field gathering and transportation process using the conversion function is as follows:

4. The method for online prediction of steam drive of a gathering and transportation pipeline network according to claim 1, characterized in that: For various working condition data sets, various working condition training data sets xLDATA i As input, a PSO-LSTM prediction model for various working conditions is established, and the particle swarm optimization algorithm is used to optimize the PSO-LSTM prediction model for various working conditions. The test data set cLDATA is used i The optimized PSO-LSTM prediction model is tested to obtain the final optimized PSO-LSTM prediction model for the energy consumption of various working conditions. Follow the steps below: Step 5-1: Initialize the particle swarm optimization algorithm parameters; The particle swarm optimization algorithm parameters include: particle population size M, maximum number of iterations G, inertia weight W, learning factor C1=C2∈[0,4], initialized particle swarm optimization algorithm population L a , where a=1,2,…,M; Step 5-2: Obtain the PSO-LSTM parameter value according to the information of each individual in the population, and obtain M groups of PSO-LSTM parameter values; Step 5-3: Use the radial basis function as the PSO-LSTM kernel function, establish a PSO-LSTM prediction model for the PSO-LSTM parameter value corresponding to each individual in the population and the training data set of the i-th working condition, and train M PSO-LSTM prediction models; The PSO-LSTM model of the i-th working condition is as follows: Among them, h(i) = 1, 2, ..., H(i), H(i) is the number of training samples in the training data set of the i-th working condition, L(i) is the new sample data that needs to calculate the energy consumption of the i-th working condition, is the output value of the PSO-LSTM prediction model corresponding to the input data L(i) of the i-th working condition, ah(i), b(i) are the parameters of the PSO-LSTM prediction model of the i-th working condition, L h(i) is the process parameter of the hth training sample in the training data set of the i-th working condition, and K(L(i), Lh(i)) is the kernel function of the PSO-LSTM prediction model of the i-th working condition; The kernel function K(L(i), Lh(i)) of the PSO-LSTM prediction model for the i-th working condition is as follows: in, is the width of the PSO-LSTM kernel function for the i-th working condition; Step 5-4: Input the test data of the i-th working condition into the PSO-LSTM model of the i-th working condition established by each individual in the population, and calculate the root mean square error value of the PSO-LSTM model of the i-th working condition established by each individual in the population as ε α , the fitness function value f of each individual in the population α =ε α ; The root mean square error value ε of the PSO-LSTM model of the i-th working condition established by each individual in the population α The calculation formula is as follows: in, is the output value of the PSO-LSTM prediction model of the process parameter in the e-th sample in the i-th operating condition test data set calculated using the corresponding measurement model of the α-th individual; Step 5-5: Determine whether the current iteration number g reaches the maximum iteration number G. If so, the iteration ends, and the optimal PSO-LSTM parameters and model parameters ah(i) and b(i) are obtained to obtain the final optimized PSO-LSTM prediction model of the gathering and transmission energy consumption of the i-th working condition. Otherwise, execute step 5-5; Step 5-6: Update the population for the g-th iteration, set the iteration number g=g+1, and return to step 5-2.

5. The method for online prediction of steam drive in a gathering and transportation pipeline network according to claim 4, characterized in that: The steps 5-6 include the following steps: Step 5-6-1: For the ath individual L in the current iteration a,g , L a,g The pbest coordinates are set to the initial position of the current position, and the fitness value f(Xi) of the individual extreme point is calculated; The calculation formula of the fitness value f(Xi) of the individual extreme point is as follows: Among them, M represents the population size, Y i is the sample output value, y i is the actual output value; Step 5-6-2: Evaluate each particle and calculate the fitness value of the particle. If it is better than the current individual extreme value of the particle, set pbest to the position of the particle and update the individual extreme value. Step 5-6-3: Update the particle position and velocity according to the fitness value f(Xi), and obtain the new particle as the g+1th iteration population particle L a,g+1 ; The g+1th iteration population particle L a,g+1 The calculation formula of position velocity is as follows: Where i = 1, 2, ..., M, M is the total number of particles, t is the current number of iterations, W is the inertia weight, vi is the particle speed, r1 and r2 are random numbers, Xi is the current position of the particle, is the current optimal position of particle i, is the current optimal position of the particle group, C1 and C2 are learning factors; Step 5-6-4: Let the iterative algebra g = g+1 and return to step 5-2.

6. The method for online prediction of steam drive in a gathering and transportation pipeline network according to claim 1, characterized in that: The online measurement values ​​of the process parameters of the gathering and transportation pipeline network are read in real time, and the energy consumption of transporting crude oil is predicted by using the PSO-LSTM prediction model of gathering and transportation energy consumption of various working conditions after the final optimization, which is performed according to the following steps: Step 7-1: Read the online measured values ​​of the process parameters of the gathering and transportation network in real time, and use the conversion function to standardize the dynamic and steady-state parameters of the data set of the oil and gas field gathering and transportation process to obtain the standardized process parameters. Step 7-2: Based on the standardized process parameters The distances from the cluster centers of various working conditions are classified and input into the PSO-LSTM prediction model of the optimized gathering and transmission energy consumption of the determined working condition category to obtain the online measurement value of the gathering and transmission energy consumption.

Images