Offshore oil and gas field platform load prediction method considering multivariable correlation

Abstract

The invention relates to an offshore oil and gas field platform load prediction method considering multivariable correlation, and the method comprises the steps: S1, building a system-level input parameter set of an offshore oil and gas field platform for load prediction; s2, based on historical data of system-level input parameter-load of the current offshore oil and gas field platform, a correlation analysis method is adopted to quantify the influence degree result of each input parameter and load in the system-level input parameter set; s3, according to the influence degree result, collecting time sequence data of a preset number of key parameters in the system set input parameter set; and S4, inputting the collected time sequence data of the preset number of key parameters into a pre-trained prediction model, and outputting a load prediction result.

Description

Technical Field

[0001] The present invention relates to the technical field of offshore oil and gas exploitation, and in particular to an offshore oil and gas field platform load prediction method considering multivariate correlation. Background Art

[0002] The offshore oil and gas industry is one of the world's major energy pillar industries. Due to its unique geographical location, offshore oil and gas fields usually use gas turbines or diesel generators to provide power for operating platforms. These platforms are interconnected by submarine cables to build a regional power grid. Compared with land oil fields, offshore platforms have denser equipment layouts, more urgent demand for power loads, and higher requirements for energy supply reliability.

[0003] The study found that the traditional load prediction of offshore oil and gas field platforms is difficult to effectively analyze the correlation between the characteristic quantities of multiple input parameters and the load prediction results. Secondly, the traditional prediction model itself needs to be improved in terms of feature extraction methods and prediction effects. Summary of the invention

[0004] The present invention provides a method for predicting the load of an offshore oil and gas field platform taking into account the correlation of multiple variables. Firstly, the correlation between the characteristic quantities in a variety of input parameters and the load is considered to screen the strong characteristic quantities to be input into the prediction model; secondly, a prediction model based on a network architecture of a gated recurrent unit with an improved structure and a dual attention mechanism is used to improve the feature extraction effect and prediction accuracy.

[0005] To achieve the above object, the present invention adopts the following technical solutions:

[0006] In a first aspect, the present application provides a method for predicting loads on an offshore oil and gas field platform taking into account multivariate correlations, the method comprising:

[0007] S1, establish a system-level input parameter set for load forecasting on offshore oil and gas field platforms;

[0008] S2, based on the historical data of system-level input parameter-load of the current offshore oil and gas field platform, a correlation analysis method is used to quantify the influence of each input parameter in the system-level input parameter set on the load;

[0009] S3, collecting time-series data of a preset number of key parameters in the system set input parameter set according to the impact degree result;

[0010] S4, inputting the collected time series data of a preset number of key parameters into a pre-trained prediction model, and outputting the load prediction result.

[0011] In one implementation, in S1, the system-level input parameter set includes seven input parameters: liquid production, water production, gas production, number of self-flowing wellheads, number of electric submersible pumps, water injection volume, and water content in three aspects: oil field production scale, mechanical production system, and crude oil gathering and transportation system.

[0012] In one implementation, in S2, the Pearson correlation coefficient is used to quantify the influence of each input parameter on the load.

[0013] In one implementation, in S3, the preset number of key parameters is all 7 input parameters or a subset thereof.

[0014] In one implementation, in S4, the prediction model adopts a network architecture of a gated recurrent unit based on a dual attention mechanism.

[0015] In one implementation, in the dual attention mechanism,

[0016] Adding an attention mechanism to the time scale of the temporal data of the preset number of key parameters;

[0017] And according to the quantified results of the influence degree of the preset number of key parameters, different weights are assigned to the extracted feature quantities of each input parameter.

[0018] In one implementation, the method further includes: a step of pre-training the prediction model; specifically including:

[0019] Using a preset number of key parameters and load historical data data sets, and preprocessing the data sets;

[0020] Divide the preprocessed data set into training set and test set;

[0021] The parameter values ​​in the prediction model network architecture are adjusted based on the training set, and the training is stopped when the preset training accuracy is reached based on the test set to obtain a prediction model that has completed the training.

[0022] In one implementation, the preprocessing includes using the 3-sigma principle to identify abnormal data in the data set and repairing the abnormal data.

[0023] The technical solution of the present invention considers the correlation between the feature quantities in multiple input parameters and the load, and screens the strong feature quantities to be input into the prediction model; and the prediction model based on the network architecture of the gated recurrent unit with an improved structure and integrated with the dual attention mechanism can improve the feature extraction effect and prediction accuracy. BRIEF DESCRIPTION OF THE DRAWINGS

[0024] Figure 1is a schematic diagram of a main energy system for an offshore oil and gas field in an embodiment of the present invention;

[0025] Figure 2 It is a schematic diagram of the structure of the GRU according to an embodiment of the present invention;

[0026] Figure 3 is a network structure diagram of a prediction model in an embodiment of the present invention;

[0027] Figure 4 is an algorithm flow chart of a prediction model according to an embodiment of the present invention;

[0028] Figure 5 is a Pearson correlation coefficient heat map in an embodiment of the present invention;

[0029] Figure 6 is the actual value of different loads in a calculation example of the present invention;

[0030] Figure 7 It is a schematic diagram of load forecasting results of the example of the present invention and the existing method;

[0031] Figure 8 It is a schematic diagram of MAPE comparison between the example of the present invention and the existing method. DETAILED DESCRIPTION

[0032] In order to make the purpose, technical solution and advantages of the embodiment of the present invention clearer, the technical solution of the embodiment of the present invention will be clearly and completely described below in conjunction with the drawings of the embodiment of the present invention. Obviously, the described embodiment is a part of the embodiment of the present invention, not all of the embodiments. Based on the described embodiment of the present invention, all other embodiments obtained by ordinary technicians in this field belong to the scope of protection of the present invention.

[0033] In view of the defects and problems of the prior art, the present application provides a method for predicting the load of an offshore oil and gas field platform taking into account the correlation of multiple variables, the method comprising:

[0034] S1, establish a system-level input parameter set for load forecasting on offshore oil and gas field platforms;

[0035] Specifically, for example, data on seven input parameters that affect the oilfield load demand, such as liquid production, water production, gas production, number of self-flowing wellheads, number of electric submersible pumps, water injection volume, and water content, can be extracted from three aspects: oilfield production scale, mechanical production system, and crude oil gathering and transportation system.

[0036] S2, based on the historical data of system-level input parameter-load of the current offshore oil and gas field platform, a correlation analysis method is used to quantify the influence of each input parameter in the system-level input parameter set on the load;

[0037] The Pearson correlation coefficient method is used to screen the input parameters of the prediction model. The Pearson correlation coefficient is the coefficient of the quotient of the covariance and standard deviation of the estimated sample, and is calculated using the following formula:

[0038]

[0039] In the above formula are the means of n experiments respectively.

[0040] The Pearson correlation coefficient fluctuates between [-1,1]. When r>0, the two variables have a positive correlation; when r<0, it indicates that the two are negatively correlated, and the larger the absolute value of the correlation coefficient, the higher the degree of correlation between the two.

[0041] The specific value range and correlation are given in Table 1.

[0042]

[0043] Table 1

[0044] S3, collecting time-series data of a preset number of key parameters in the system set input parameter set according to the impact degree result;

[0045] Specifically, the preset number of key parameters is all 7 input parameters or a subset thereof, wherein the quantified influence degrees are sorted from high to low.

[0046] S4, inputting the collected time series data of a preset number of key parameters into a pre-trained prediction model, and outputting the load prediction result.

[0047] Specifically, the prediction model adopts the DA-GRU algorithm that incorporates the dual attention (DA) mechanism.

[0048] Before this step, the prediction model needs to be trained, including:

[0049] Get historical data;

[0050] The acquired historical data of offshore oil and gas fields are preprocessed and divided into training set and test set. To ensure the high quality of data, the dates with large-scale missing data are deleted, and the dates with a small amount of missing data are filled with the previous value. The 3-sigma principle is used to identify the outliers at each time point, and the calculation method is as follows:

[0051]

[0052] In the above formula, X i is the eigenvalue at the i-th moment; is the characteristic mean at the i-th moment; σ iis the characteristic standard deviation at the i-th moment; m is the number of moments; k is the number of samples; X ij is the jth sample of the feature. The identified abnormal data is also filled with the previous value.

[0053] In order to overcome the differences in dimensions of the feature quantities, the data set needs to be normalized and the normalization method shown in the following formula is used for calculation.

[0054]

[0055] In the above formula, x norm The data is after standardization; x is the data to be standardized; x min is the minimum value of the input data, x max is the maximum value of the input data.

[0056] The training set data is input into the DA-GRU network model, the features with greater influence in the input data are extracted, and the additive attention mechanism is used to assign weights to them to obtain a strong feature training sequence.

[0057] The internal structure of a GRU neuron is as follows Figure 2 As shown, in Figure 2 In, Z t is an update gate. It is used to control the transformation of previous state information into the current state. The larger its value is, the more information is to be stored by the current neuron. t is a reset threshold used to determine the degree to which previous information has been forgotten. h t The candidate hidden state of h t is the output state at time t. This variable retains the information of the current unit and passes it to the next unit.

[0058] Z t The calculation formula is:

[0059] Z t =σ(W z x t +U z h t-1 )

[0060] R t The calculation formula is:

[0061] R t =σ(W r x t +U r h t-1 )

[0062] The calculation formula is:

[0063]

[0064] h t The calculation formula is:

[0065]

[0066] In the above formula, x t Indicates the input data at the current moment; h t-1 is the traffic volume output of the memory unit at the previous moment (i.e., moment t-1); W z , W r and U z , U r Represents the update gate and reset gate x respectively t and h t-1 The weight matrix; W is the training parameter matrix; is the Hadamard product; tanh is the hyperbolic tangent function of the activation function, which can convert the data into values ​​in the range of [0,1] and [-1,1] respectively.

[0067] The calculation formula of the additive attention mechanism is as follows:

[0068] q t =W q x t +b q

[0069] e t =V e T tanh(W e x t +U e q t +b e )

[0070]

[0071] In the above formula, q t is the query vector; W e 、V e T , U e , W q is the learnable weight; x t is the input vector of the attention module; e t is the score of attention; b q 、b e is the learnable bias coefficient; α ti for e t The attention weight of the i-th feature in .

[0072] On the basis of retaining the basic information of the original features, the extracted feature sequence is processed by dimensionality reduction and then input into the GRU neural network model for training to obtain the preliminary prediction value.

[0073] The prediction is reverse normalized to obtain the final prediction result.

[0074] The method provided above has the following technical effects:

[0075] 1. The integration of DA mechanism can effectively improve the GRU network's ability to mine effective feature information and improve the model prediction accuracy; in the prediction of electricity, gas and oil loads in offshore oil and gas fields, the average RMSE of the GRU prediction model after integrating the DA mechanism is 13.39% lower than that of the traditional GRU.

[0076] 2. The proposed model obtains more accurate prediction results based only on the historical production data during the production process, reducing the reliance on data on uncertain factors such as weather in offshore oil and gas fields. It is not limited by the scale and energy characteristics of the oil field and has a wide range of applications.

[0077] A detailed description is given below in a specific example to demonstrate the above technical effect.

[0078] An offshore oil and gas field is taken as the research object to verify the accuracy of the proposed offshore oil and gas field load prediction model. The load data, production and water content data of the offshore oil and gas field from January 1, 2023 to January 1, 2024 are selected as experimental data, and the training set and test set are divided into two sets in a ratio of 2:1.

[0079] The experimental parameters are set as follows: the number of hidden neurons in the DA-GRU prediction model is 200, the number of epochs is 300, the initial learning rate is 0.0005, and it decays to 90% of the previous value every 10 epochs. In the attention module, the number of neurons in the temporal attention hidden layer is 56, and the number of neurons in the feature attention hidden layer is 40.

[0080] The Pearson correlation coefficient method is used to determine and quantify the influence of each characteristic parameter on the oil field load. The correlation coefficient heat map is shown in Figure 5As shown. The absolute value of the correlation coefficient between the characteristic quantities such as gas production, water production, and liquid production and different loads basically reached above 0.4, which means that there is a medium or higher degree of correlation between gas production, water production, and liquid production and load; the absolute value of the correlation coefficient between the number of self-flowing wells, the number of electric submersible pumps, and the water injection volume and load is basically below 0.4, especially the correlation coefficients between the number of electric submersible pumps and the load are 0.24, 0.31, and -0.25, respectively, so it can be considered that there is an overall extremely weak correlation between the number of electric submersible pumps and the load; but the correlation coefficient between the number of self-flowing wells and the water injection volume and the load is basically less than 0.1, which means that the number of self-flowing wells and the water injection volume have basically no effect on the load, so these three types of characteristics are not used as model inputs. Based on the above analysis, the time series consisting of gas production, water production, liquid production, and water content over a period of time is selected as the model input to predict the electric load, gas load, and diesel load of offshore oil and gas fields.

[0081] In order to verify the correctness of the model and evaluate the performance difference of the prediction model, this paper uses indicators such as root mean squared error (RMSE) and mean absolute percentage error (MAPE) to evaluate the load forecasting. The calculation formulas of MAPE and RMSE are as follows:

[0082]

[0083] In the above formula, y i is the actual load value; is the load forecast value.

[0084] The control variable method is used to predict the load in the same time series using DA-GRU, traditional GRU and DA-LSTM. The predicted values ​​are compared with the true values, and the prediction performance of the algorithm is evaluated using MAPE and RMSE. The true value of the load and the prediction results using different algorithms are shown in Figure 2. Figure 6 , Figure 7 shown. Figure 6 The horizontal scale in the middle represents the time in days, and the vertical scale represents the actual value of the load. Figure 7 The horizontal scale in the middle represents the time in days, and the vertical scale represents the relative error power of the load (load prediction power - load actual power).

[0085] from Figure 7It can be seen that the traditional GRU algorithm has a lower relative error result in the prediction of different types of loads. After adding the dual attention mechanism, its predicted relative error power is closer to the zero axis, indicating that DA-GRU has better prediction performance than the traditional GRU algorithm. Comparing the predicted relative error power of DA-GRU and DA-LSTM, the overall offset of the relative error power of DA-LSTM from the zero axis is greater than that of DA-GRU. Both of them have added the dual attention mechanism, but DA-GRU has better performance, indicating that under the unique load characteristics of offshore oil and gas fields, the basic performance of the GRU prediction algorithm is slightly better than that of LSTM.

[0086] The prediction results of different algorithms are compared with Figure 6 The actual load values ​​shown are compared and analyzed to quantitatively evaluate the performance differences of the three models in predicting different types of loads in offshore oil and gas fields. Tables 2 and 3 show the MAPE and RMSE data of the three models after predicting electricity, gas, and oil loads, respectively. Figure 8 The MAPE differences of different algorithms are shown.

[0087]

[0088] Table 2

[0089]

[0090] Table 3

[0091] From Table 2, Table 3 and Figure 8 It can be seen that the error of the DA-GRU model with the introduction of the dual attention mechanism is smaller than that of the GRU. For example, the power load prediction accuracy is 0.663% lower than that of the conventional GRU, and the RMSE is 10732.439kW.h lower than that of the traditional GRU. The average RMSE of the DA-GRU model in the three types of typical load prediction results is 13.39% lower than that of the GRU. , indicating that the dual attention mechanism has a significant performance improvement on the underlying model. The error of the DA-GRU model is smaller than that of the DA-LSTM. For example, in the power load prediction results, the MAPE predicted by the DA-GRU is 2.654% lower than that of the DA-LSTM, and the RMSE is 16334.224kW.h lower than that of the DA-LSTM. This shows that although both models are integrated with the DA mechanism on the basis of the underlying prediction model, the performance of the fusion of LSTM and DA is far inferior to that of the GRU integrated with the DA mechanism.

[0092] In the several embodiments provided by the present invention, it should be understood that the disclosed method can be implemented in other ways. For example, the device embodiments described above are only schematic. For example, the division of the above units is only a logical function division. There may be other division methods in actual implementation. For example, multiple units or components can be combined or integrated into another system, or some features can be ignored or not executed. Another point is that the mutual coupling or direct coupling or communication connection shown or discussed can be through some interfaces, indirect coupling or communication connection of devices or units, which can be electrical, mechanical or other forms.

[0093] The above-mentioned integrated unit implemented in the form of a software functional unit can be stored in a computer-readable storage medium. The above-mentioned software functional unit is stored in a storage medium, including a number of instructions for a computer device (which can be a personal computer, a server, or a network device, etc.) or a processor (Processor) to perform some steps of the above-mentioned method of each embodiment of the present invention. The aforementioned storage medium includes: U disk, mobile hard disk, read-only memory (Read-Only Memory, ROM), random access memory (Random Access Memory, RAM), disk or optical disk and other media that can store program codes.

[0094] The above are only preferred embodiments of the present invention and are not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principles of the present invention should be included in the scope of protection of the present invention.

Claims

1. A method for predicting the load of an offshore oil and gas field platform considering multivariate correlation, characterized in that: The method comprises: S1, establish a system-level input parameter set for load forecasting on offshore oil and gas field platforms; S2, based on the historical data of system-level input parameter-load of the current offshore oil and gas field platform, a correlation analysis method is used to quantify the influence of each input parameter in the system-level input parameter set on the load; S3, collecting time-series data of a preset number of key parameters in the system set input parameter set according to the impact degree result; S4, inputting the collected time series data of a preset number of key parameters into a pre-trained prediction model, and outputting the load prediction result.

2. The offshore oil and gas field platform load prediction method considering multivariate correlation according to claim 1 is characterized in that: In S1, the system-level input parameter set includes seven input parameters: liquid production, water production, gas production, number of self-flowing wellheads, number of electric submersible pumps, water injection volume, and water content in three aspects: oil field production scale, mechanical production system, and crude oil gathering and transportation system.

3. The offshore oil and gas field platform load prediction method considering multivariate correlation according to claim 2 is characterized in that: In S2, the Pearson correlation coefficient is used to quantify the influence of each input parameter on the load.

4. The offshore oil and gas field platform load prediction method considering multivariate correlation according to claim 2, characterized in that: In S3, the preset number of key parameters is all 7 input parameters or a subset thereof.

5. The offshore oil and gas field platform load prediction method considering multivariate correlation according to claim 4 is characterized in that: In S4, the prediction model adopts a network architecture of a gated recurrent unit based on a dual attention mechanism.

6. The offshore oil and gas field platform load prediction method considering multivariate correlation according to claim 5, characterized in that: In the dual attention mechanism, Adding an attention mechanism to the time scale of the temporal data of the preset number of key parameters; And according to the quantified results of the influence degree of the preset number of key parameters, different weights are assigned to the extracted feature quantities of each input parameter.

7. The offshore oil and gas field platform load prediction method considering multivariate correlation according to claim 6, characterized in that: The method further includes: a step of pre-training the prediction model; specifically including: Using a preset number of key parameters and load historical data data sets, and preprocessing the data sets; Divide the preprocessed data set into training set and test set; The parameter values ​​in the prediction model network architecture are adjusted based on the training set, and the training is stopped when the preset training accuracy is reached based on the test set to obtain a prediction model that has completed the training.

8. The offshore oil and gas field platform load prediction method considering multivariate correlation according to claim 7, characterized in that: The preprocessing includes using the 3-sigma principle to identify abnormal data in the data set and repairing the abnormal data.

Images