A dynamic prediction method of construction pressure based on spatiotemporal data characteristics

Abstract

The present invention relates to a method for dynamic prediction of construction pressure based on spatiotemporal data characteristics, including: S1, obtaining original construction data and geological data of multiple wells; S2, extracting construction data from the start stage of pre-fluid to the pump stop stage in the construction data; S3, constructing time series data from the construction data, and combining the construction data and geological data to construct active independent variable data; S4, establishing a hybrid neural network model and an LGBM gradient boosting model; S5, respectively training the hybrid neural network model and the LGBM gradient boosting model with time series data and active independent variables; S6, obtaining construction data of a new well for fracturing, and using the trained model for prediction; S7, initializing the weight value of the prediction model, updating the iterative weight, and obtaining the final prediction value. The method provided by the present invention not only improves the prediction accuracy of the construction pressure, but also enhances the controllability and flexibility of the operation process, and provides a safer and more efficient solution for hydraulic fracturing operations.

Description

Technical Field

[0001] The invention relates to the technical field of oil and gas field development, and in particular to a construction pressure dynamic prediction method based on spatiotemporal data characteristics. Background Art

[0002] In the development of modern oil and gas fields, hydraulic fracturing technology is widely used as a key means of increasing production. However, due to the complexity of hydraulic fracturing operations, the construction process often faces a variety of safety hazards, among which sand plugging is particularly common. Sand plugging may not only cause equipment damage, loss of fracturing fluid and extension of construction period, but may even cause casualties in severe cases.

[0003] The occurrence of sand plugging events is affected by both the initial construction pressure of the fracturing stage and the spatial characteristic parameters of the fracturing section, and has typical time and space dual coupling characteristics. At present, the technology used to monitor and prevent sand plugging in hydraulic fracturing operations is mainly based on traditional monitoring of wellhead parameters, such as real-time monitoring of key parameters such as bottom hole pressure and flow. These parameters can provide initial signs of sand plugging, but usually cannot provide sufficient warning time to take effective preventive measures, so they are limited in avoiding the occurrence of sand plugging events. In addition, most of these methods rely on manual identification by technicians, which limits the accurate prediction of the specific time and location of sand plugging and reduces the efficiency of dealing with potential sand plugging events.

[0004] In this context, it has become a major challenge for oil and gas field operators to effectively predict and prevent sand plugging risks under the existing technical framework. Therefore, it is of great significance to propose a method that can accurately monitor and predict bottom hole pressure. Through continuous monitoring and analysis of bottom hole pressure, potential signs of sand plugging can be discovered in time and corresponding preventive measures can be taken. This not only improves the efficiency and accuracy of sand plugging prevention, but also greatly reduces operational risks and ensures the safety and economic benefits of oil and gas field development. Summary of the invention

[0005] In order to more effectively cope with the safety and efficiency challenges faced by hydraulic fracturing operations in modern oil and gas field development, the present invention proposes a dynamic prediction method for construction pressure based on spatiotemporal data characteristics, specifically a hybrid neural network model (i.e., CNN-LSTM-Attention hybrid neural network model) that integrates convolutional neural network (CNN), long short-term memory network (LSTM) and attention mechanism (Attention), as well as active independent variable technology. The innovation of this method lies in that it comprehensively uses machine learning technology to conduct a detailed analysis of a large amount of field construction data, so as to accurately capture the time series dependency and spatial parameter characteristics of the fracturing process. This not only includes identifying global trends and local fluctuations, but also involves learning the spatial characteristics of different fracturing layers to achieve real-time monitoring of pressure changes during fracturing. Through this method, the abnormal bottom hole pressure that occurs can be predicted and adjusted in a timely manner, effectively improving the safety and efficiency of oil and gas field development, and introducing a new intelligent solution for oil and gas field development.

[0006] The present invention is based on a hybrid neural network model and can analyze and process a large amount of data generated during the fracturing operation in real time. By combining time series analysis and anomaly detection algorithms, the system can predict the changes in bottom hole pressure in the future and provide accurate multi-step predictions based on historical and real-time data.

[0007] In particular, the spatial active independent variables in the present invention play a vital role. These spatial variables include not only traditional monitoring parameters such as displacement, sand concentration, and sand ratio, but also other key geological parameters, wellbore parameters, etc., which can be actively adjusted in response to the prediction results. In this way, the operator can not only timely check the changes in bottom hole pressure in the future, but also adjust the operation strategy to reduce the bottom hole pressure and avoid the impact of sand blockage based on the prediction results and the feedback of the active independent variables. Therefore, the prediction method of the present invention provides a more comprehensive and dynamic risk management strategy, which significantly improves the safety and efficiency of oil and gas field development.

[0008] A construction pressure dynamic prediction method based on spatiotemporal data characteristics comprises the following steps:

[0009] S1. Obtaining original construction data of a complete fracturing construction cycle of multiple wells and geological data of corresponding fracturing layers;

[0010] S2. Divide the complete fracturing construction cycle into different intervals according to the construction data, and extract the original construction data from the start stage of the pre-fluid to the pump stop stage;

[0011] S3, preprocessing the construction data obtained in step S2 and constructing time series data through a sliding window algorithm, combining the construction data with the geological data for preprocessing and then combining to construct active independent variable data, wherein the construction data in the active independent variable data does not include bottom hole pressure;

[0012] S4, establish a CNN-LSTM-Attention hybrid neural network model for receiving time series data and a LGBM gradient boosting model for receiving active independent variable data;

[0013] S5, using the time series data and active independent variables obtained in step S3 to train the CNN-LSTM-Attention hybrid neural network model and the LGBM gradient boosting model respectively, and using the Bayesian optimization algorithm to optimize the hyperparameters of the model training process;

[0014] S6, obtaining construction data and geological data of a new fracturing well, and using the model trained in step S5 to make a multi-step prediction of the bottom hole pressure at the beginning of the sand-carrying fluid stage;

[0015] S7. Initialize the weight values ​​of the prediction results of the CNN-LSTM-Attention hybrid neural network model and the LGBM gradient boosting model, calculate the root mean square error of the two models based on the prediction results of each round and the actual pressure, and use the reverse error method to iterate the weights of the prediction results of the two models to obtain the final prediction value of the construction pressure.

[0016] Furthermore, the construction data in step S1 include: bottom hole pressure, displacement, sand concentration, sand ratio, cumulative sand volume and cumulative displacement; the geological data include: Poisson's ratio, Young's modulus, formation pressure, maximum horizontal principal stress, minimum horizontal principal stress, pore pressure and vertical stress.

[0017] Furthermore, the construction data obtained in step S3 is preprocessed to obtain preprocessed data, wherein the preprocessing operation includes the following steps:

[0018] S31, delete null values ​​and interpolate missing value data;

[0019] S32, standardizing various construction data respectively;

[0020] S32. Divide the standardized construction data according to a certain ratio to obtain a training data set, a verification data set, and a test data set.

[0021] Furthermore, in step S3, the time series data is constructed by a sliding window algorithm, which includes the following steps:

[0022] The data of a part of time points in the past are constructed into a time window, and the number of time points contained in each subsequence is the time window size; moving the window forward by one or more time units in the entire data set will form multiple time windows.

[0023] Furthermore, the active independent variable data in step S3 includes displacement, sand concentration, and sand ratio.

[0024] Furthermore, in step S7, the weights of the prediction results of the two models are updated and iterated using the reverse error method.

[0025] Furthermore, the method further comprises step S8:

[0026] Modify active independent variables and view future predicted pressure to achieve bottom hole pressure control and provide construction optimization suggestions.

[0027] Furthermore, step S8 also includes:

[0028] S81, obtaining the displacement, sand concentration, sand ratio and geological data of n pieces of data before the backward prediction time point, and taking the average of the displacement, sand concentration and sand ratio respectively;

[0029] S82, setting the active variable displacement, sand concentration, and sand ratio predicted for the next m time points to the mean value obtained in step S81, while the geological data remain unchanged, and inputting them into the LGBM gradient boosting model for prediction;

[0030] S83. When it is necessary to observe the predicted pressure trend under different construction parameters, the obtained mean is adjusted and then input into the LGBM gradient boosting model for prediction.

[0031] On the other hand, the present invention provides a computer device, comprising a memory, a processor, and a computer program stored in the memory and running on the processor, characterized in that when the processor executes the computer program, the steps of the construction pressure dynamic prediction method described in any one of the above items are implemented.

[0032] On the other hand, the present invention provides a computer-readable storage medium having a computer program stored thereon, characterized in that when the computer program is executed by a processor, the steps of any of the above-mentioned methods for dynamically predicting construction pressure are implemented.

[0033] The present invention provides a method for dynamic prediction of construction pressure based on spatiotemporal data characteristics, which has the following beneficial technical effects: (1) High-precision prediction. By coupling the long short-term memory (LSTM) network and active independent variables, the present method can achieve highly accurate construction pressure prediction. The long short-term memory network optimizes the processing of time series data and improves the accuracy and reliability of prediction. (2) Real-time monitoring and analysis. The method provides real-time monitoring and analysis functions, which can quickly identify abnormal changes in construction pressure. This not only helps to discover potential problems in a timely manner, but also makes timely adjustments based on dynamic changes. (3) Reduce operational risks: By predicting and monitoring construction pressure in real time, the present invention significantly reduces the risks in the operation process and reduces equipment damage and operation interruptions caused by abnormal pressure. (4) Improve operational efficiency. With the help of predictive maintenance and active independent variable control, the system can reduce downtime and improve overall operational efficiency. BRIEF DESCRIPTION OF THE DRAWINGS

[0034] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings required for use in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative labor.

[0035] Figure 1 It is a schematic diagram of the prediction method flow of the present invention;

[0036] Figure 2 Schematic diagram of the implementation of the sliding window algorithm.

[0037] Figure 3 This is a structural diagram of the CNN-LSTM-Attention model.

[0038] Figure 4 This is a graph showing the loss decrease during the training process of the CNN-LSTM-Attention model and the CNN-LSTM model.

[0039] Figure 5 Comparison chart of predicted values ​​and true values ​​of CNN-LSTM-Attention and LGBM gradient boosting models on the test set.

[0040] Figure 6 This is a comparison chart of the predicted value and the true value of the prediction method of the present invention on the target well.

[0041] Figure 7 This is a flow chart for the practical application of the construction pressure dynamic prediction model. DETAILED DESCRIPTION

[0042] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0043] To facilitate understanding of the technical solutions provided by the embodiments of the present application, the background technology of the present application is first described.

[0044] The construction process of hydraulic fracturing generally includes the following stages: (1) the preparation stage before fracturing; (2) circulation, checking the performance of the fracturing vehicle equipment and ensuring the smooth flow of the surface process pipeline; (3) starting the high-pressure pump of the fracturing vehicle and testing the pressure-bearing performance of the equipment above the wellhead valve and the surface pipeline; (4) the pre-fluid stage, the pre-fluid is usually clean water or a low-viscosity liquid; (5) the trial extrusion stage, starting the fracturing vehicle and squeezing the fracturing fluid into the formation according to the designed displacement; (6) the sand-carrying fluid stage, injecting the sand-carrying fluid (a mixture of fracturing fluid and proppant); (7) the displacement fluid stage, pushing the sand-carrying fluid in the tubing or casing into the formation; (8) post-fracturing processing work: including stopping the pump, crack expansion monitoring, post-fracturing analysis, etc. Of course, in actual work projects, technicians can adjust the above steps according to specific circumstances.

[0045] During the fracturing process, if the transported proppant cannot enter the formation in time, but is blocked in the artificial fracture, casing or tubing, the pressure of the fracturing fluid will suddenly increase and the construction cannot continue. This phenomenon is called sand plugging. Commonly used real-time diagnostic analysis methods include net pressure fitting method, construction pressure double logarithmic curve method, "N" value method and slope reversal method, but these traditional methods are mostly based on subjective and empirical judgment in actual application, which is easy to cause misjudgment and cannot provide sufficient warning time.

[0046] The sand plugging phenomenon generally occurs after the start of the sand-carrying fluid stage, and before the sand-carrying fluid stage, the well has already gone through the stages of ball delivery and pre-fluid, and some construction data exist. However, since the time from the pre-fluid to the sand-carrying fluid stage is not long, and sometimes there is a phenomenon of abnormal bottom hole pressure just after entering the sand-carrying fluid stage, in order to enable operators to predict subsequent bottom hole pressure changes as early as possible, so as to adjust the construction pressure in advance and ultimately avoid sand plugging, the present invention is based on the existing construction data and the corresponding geological data of the fracturing layer, and adopts a machine learning method to provide a construction pressure dynamic prediction method.

[0047] like Figure 1 As shown in the flow chart in the figure, the present invention provides a method for dynamic prediction of construction pressure based on spatiotemporal data characteristics, comprising the following steps:

[0048] S1. Obtain the original construction data of the complete fracturing construction cycle of multiple wells in the study area and the geological data of the corresponding fracturing layers;

[0049] S2. Divide the complete fracturing construction cycle into different intervals according to the original construction data, and extract the original construction data from the start stage of the pre-fluid to the pump stop stage;

[0050] S3, preprocessing the obtained construction data and constructing time series data through a sliding window algorithm, and combining the construction data and / or geological data to construct active independent variable data;

[0051] S4. Build a CNN-LSTM-Attention hybrid neural network model to receive time series data, and build an LGBM gradient boosting model (LightGBM) to receive active independent variable data;

[0052] S5. Use the obtained time series data to train the CNN-LSTM-Attention hybrid neural network model, use the active independent variable data to train the LGBM gradient boosting model, and use the Bayesian optimization algorithm to optimize the hyperparameters of the model training process;

[0053] S6, obtaining existing construction data and geological data of a target fracturing well, and using the model trained in step S5 to perform multi-step future prediction of bottom hole pressure;

[0054] S7, initialize the weight values ​​of the prediction results of the two models, and calculate the root mean square errors of the two models according to the prediction results of each round and the actual pressure, and use the reverse error method to iterate the weights of the prediction results of the two models, so as to obtain a more balanced and accurate final prediction value;

[0055] S8. Modify the active independent variables and check the future predicted pressure to achieve bottom hole pressure control and provide construction optimization suggestions.

[0056] In the above step S1, the original construction data includes parameters such as bottom hole pressure, displacement, sand concentration, sand ratio, cumulative sand volume and cumulative liquid volume. The bottom hole pressure refers to the pressure formed at the bottom of the well due to the injection of high-pressure fracturing fluid into the wellbore and squeezing into the reservoir during the fracturing process. In this field, the bottom hole pressure is generally converted from the wellhead pressure, and its unit is MPa. The displacement refers to the rate or volume of the fracturing fluid injected into the well, and its unit is generally m 3 / min. Sand concentration refers to the mass of proppant (such as sand, ceramsite, etc.) in unit volume of fracturing fluid, and the unit is usually mass concentration unit, such as kg / m 3Or lb / bbl, which means the mass of proppant in each cubic meter or barrel of fracturing fluid. Sand ratio refers to the volume ratio of fracturing fluid to proppant, and is dimensionless. Cumulative sand volume refers to the total volume or total mass of proppant injected into the well during the fracturing operation. Cumulative fluid volume refers to the total volume of fracturing fluid (including pre-fluid, sand-carrying fluid and post-fluid, etc.) injected into the well during the fracturing operation. Since the occurrence of sand plugging time is affected by the above-mentioned original construction data, and the above-mentioned original construction data is data that changes with time series.

[0057] The geological data include parameters such as Poisson's ratio, Young's modulus, maximum horizontal principal stress, minimum horizontal principal stress, vertical stress, formation pressure, pore pressure, etc. The above geological parameters are conventional parameters used in the field to describe the properties of the formation and will not be repeated here. Generally speaking, the geological data of different fracturing layers show different characteristics, so the prediction of sand plugging events is also affected by the above spatial characteristic parameters.

[0058] In the above step S2, during the fracturing construction process, the main fracturing process is from the pre-fluid stage to the pump stop stage, so the data of this stage in the construction process is extracted as the original data.

[0059] In the above step S3, the original construction data needs to be preprocessed to obtain preprocessed data, wherein the preprocessing operation includes the following steps:

[0060] (S31) deleting null values ​​and interpolating missing value data;

[0061] (S32) Each of the original construction data is standardized, and the standardization formula is:

[0062]

[0063] Where Z is the standardized value, x is the original data, μ is the mean of the data set, and σ is the standard deviation of the data set;

[0064] Further, the standardized construction data is divided according to the ratio of , to obtain a training data set, a verification data set, and a test data set. In this embodiment, the division is performed according to the ratio of 6:2:2. Of course, those skilled in the art can also divide according to other ratios, and the specific division ratio is not limited in this application.

[0065] The sliding window algorithm in step S3 constructs the data of a part of time points in the past into a time window, and the number of time points contained in each subsequence is the time window size; the window is moved forward by one or more time units in the entire data set to form multiple time windows; each time window is a subsequence of a time series, which can be used as a feature vector for subsequent analysis or machine learning model training.

[0066] For example, Figure 2 As shown, the construction data from the 1st second to the n+3th second can be divided into sample 1 data set, sample 2 data set, and sample 3 data set. The sample 1 data set is the construction data from the 1st second to the nth second and the bottom hole pressure at the n+1th second, the sample 2 data set is the construction data from the 2nd second to the n+1th second and the bottom hole pressure at the n+2th second, and the sample 3 data set is the construction data from the 3rd second to the n+2th second and the bottom hole pressure at the n+3th second, and so on. Multiple time series data can be obtained for subsequent analysis or training of machine learning models.

[0067] In step S3, the active independent variable is a parameter that the staff can actively adjust to affect the bottom hole pressure, and the active independent variable does not include the parameter bottom hole pressure. Those skilled in the art can select construction data and geological data to combine and construct the active independent variable data. In one embodiment, the active independent variable can also only include construction data or geological data. For example, the active independent variable is the displacement, sand concentration, and sand ratio corresponding to each time point. These three parameters are also the main parameters that need to be adjusted in subsequent predictions.

[0068] The CNN-LSTM-Attention hybrid neural network model in step S4 combines three powerful components: CNN (convolutional neural network) unit, LSTM (long short-term memory) unit, and Attention mechanism. Figure 3The figure shows the structure of the CNN-LSTM-Attention model, which includes an input layer, a convolutional layer, an LSTM layer, an Attention layer, a fully connected layer, and an output layer. The CNN unit in this model is responsible for extracting important spatial features from the input data, and is particularly suitable for processing data with significant spatial structures, such as images or shaped time series data. The LSTM unit is responsible for processing time series data, and excellently captures and memorizes long-term and short-term temporal dependencies in the data. In addition, the added Attention mechanism can further enhance the performance of the model. It improves the model's sensitivity and processing capabilities for key information by giving the model the ability to focus on the most important parts of the input data. This enables the entire network to not only effectively process data with complex spatial and temporal features, but also adaptively focus on the most critical parts of these features, further improving the prediction or classification accuracy of the model.

[0069] Among them, the CNN convolution unit consists of two parts: the convolution layer and the pooling layer. The calculation formula of the convolution layer is:

[0070] F conv =ReLU(W*X+b) (2)

[0071] Where W is the weight of the convolution kernel, X is the input data, b is the bias term, * represents the convolution operation, and ReLU is the activation function.

[0072] The data processed by the convolution layer will enter the pooling layer to reduce the feature dimension and help extract more robust features. The maximum pooling operation is usually selected, and its formula is:

[0073]

[0074] Where P ij is the feature element after pooling, X i+m,j+n For the feature elements in the pooling window, the maximum pooling operation selects the maximum value in the area it covers.

[0075] After being processed by the CNN convolutional unit, the data is reconstructed into a format suitable for LSTM input. The calculation formula of the forget gate in the LSTM long short-term memory neural network is:

[0076] f t =σ(W f ·[h t-1 ,x t ]+b f ) (4)

[0077] In the formula, f t represents the output of the forget gate, Wf and b f is the relevant parameter, σ represents the sigmoid function, h t-1 is the hidden state at the previous moment, x t is the input at the current moment;

[0078] The input gate consists of two parts: a sigmoid function for determining the information to be updated, and a tanh function for calculating new candidate values, the calculation formula of which is as follows:

[0079] i t =σ(W i ·[h t-1 ,x t ]+b i ) (5)

[0080]

[0081] In the formula, i t represents the output of the input gate, represents the candidate value, W i , W c , b i , b c is the relevant parameter;

[0082] The cell state update formula is:

[0083]

[0084] In the formula, C t Indicates the cell state at the current moment, C t-1 Indicates the cell state at the previous moment;

[0085] The output gate determines which part of the cell state will be used as the output of the LSTM unit. The formula is as follows:

[0086] o t =σ(W o ·[h t-1 ,x t ]+b o ) (8)

[0087] h t =o t *tanh(C t ) (9)

[0088] In the formula, after passing through the sigmoid function, it is multiplied by the tanh value of the cell state to generate the LSTM unit output at the current moment, h t Input data to the attention layer;

[0089] The attention mechanism calculates an attention score for each time step in the sequence. The score measures the relative importance of the hidden state of that time step in the entire sequence. The calculation formula for the attention score data is:

[0090] s T =w a tanh(w a h t +b a ) (10)

[0091] In the formula, s T is the attention score data, w a is the attention mechanism weight vector, h t is the output result of the long short-term memory unit at time T, b a is the attention mechanism bias vector;

[0092] Then use the softmax function to convert these attention scores into probability distribution (ie, attention weights). The calculation formula for the attention probability distribution value at time T is:

[0093]

[0094] In the formula, the λ T Assign a value to the attention probability at time t, whose value is directly between 0 and 1, which indicates the importance at time t. exp is an exponential function with the natural constant e as the base.

[0095] Finally, the hidden state of each time step is multiplied by its corresponding attention weight, and then summed over all time steps to obtain the final weighted feature representation, whose weighted formula is:

[0096]

[0097] In the formula, r is the weighted feature representation, λ t Assign values ​​to the attention probabilities, h t Output value for LSTM unit.

[0098] In step S4, the LGBM gradient boosting model (LightGBM) is an efficient gradient boosting framework. It is an ensemble learning method based on the decision tree algorithm. It minimizes the loss function by iteratively adding decision trees. LightGBM shows higher efficiency and lower memory consumption when processing large data sets.

[0099] Among them, the basic formula of gradient boosting is:

[0100] F t (x) = F t-1 (x)+α·f t(x)(13)

[0101] Where F t (x) is the model after the tth iteration, F t-1 (x) is the model after the t-1th iteration, α is the learning rate, f t (x) is the newly added tree.

[0102] In each round of iteration, the loss function L is selected and its negative gradient with respect to the model output is calculated, which is used as the training target of the new tree. The calculation formula of the negative gradient is:

[0103]

[0104] Where g i is the negative gradient, L is the loss function, F(x i ) is the output result of the current iteration model, x i y i is the corresponding data point.

[0105] The LGBM gradient boosting model also uses GOSS and EFB technologies to reduce data dimensions and better handle large-scale data sets.

[0106] In step S7, the root mean square error is a common indicator for measuring the difference between the model prediction value and the actual observation value, and its formula is:

[0107]

[0108] Where y i is the true value at the i-th time point, is the predicted value at the i-th time point.

[0109] The reverse error method and the calculation model weight formula in step S7 are:

[0110]

[0111] Where σ lstm and σ lgbm are the root mean square error values ​​obtained by the LSTM model and the LGBM gradient boosting model in the previous round of prediction, W lstm and W lgbm are the weight ratios of the prediction values ​​of the two models respectively.

[0112] The final prediction value in step S7 is the prediction value of the two models multiplied by the corresponding weight value, and the calculation formula is:

[0113]

[0114] In the formula and They are the predicted values ​​of LSTM model and LGBM gradient boosting model respectively. is the final predicted value.

[0115] In step S8, the specific operation of adjusting the active independent variable is:

[0116] (S81) obtaining the displacement, sand concentration, sand ratio and geological data of n pieces of data (n>0) before the backward prediction time point, and taking the average of the displacement, sand concentration and sand ratio respectively;

[0117] (S82) The active variables displacement, sand concentration, and sand ratio predicted for m time points (m>0) in the future are set to the mean values ​​obtained in step (S81), while the geological data remain unchanged, and are input into the LGBM gradient boosting model for prediction;

[0118] (S83) When it is necessary to observe the predicted pressure trend under different construction parameters, the mean value obtained in step (S81) is adjusted and then input into the LGBM gradient boosting model for prediction.

[0119] The dynamic prediction method of construction pressure based on spatiotemporal prediction in the present invention not only improves the prediction accuracy, but also enhances the controllability and flexibility of the operation process, providing a safer and more efficient solution for hydraulic fracturing operations.

[0120] In the present invention, based on the above-mentioned construction pressure dynamic prediction method based on spatiotemporal data characteristics, a computer device is also provided in one embodiment of the present invention, wherein the above-mentioned method is run on the computer device. The computer device may include one or more processors, such as one or more central processing units (CPUs) or graphics processing units (GPUs), and each processing unit may implement one or more hardware threads. The computer device may also include any memory for storing any kind of information such as code, settings, data, etc. In a specific embodiment, a computer program on the memory and executable on the processor, when the computer program is executed by the processor, can execute instructions according to the above-mentioned method.

[0121] Without limitation, for example, the memory may include any one or more combinations of the following: any type of RAM, any type of ROM, flash memory device, hard disk, optical disk, etc. More generally, any memory may use any technology to store information. Further, any memory may provide volatile or non-volatile retention of information. Further, any memory may represent a fixed or removable component of a computer device. In one case, when the processor executes the associated instructions stored in any memory or combination of memories, the computer device may perform any operation of the associated instructions. The computer device also includes one or more drive mechanisms for interacting with any memory, such as a hard disk drive mechanism, an optical disk drive mechanism, etc.

[0122] Corresponding to the construction pressure dynamic prediction method in the present invention, the embodiment of this article also provides a computer-readable storage medium, on which a computer program is stored, and when the computer program is executed by a processor, the steps of the above method are executed.

[0123] An embodiment of the present invention further provides a computer-readable instruction, wherein when a processor executes the instruction, the program therein causes the processor to execute the method as described above.

[0124] Application Examples

[0125] In a specific implementation, the prediction method of the present invention is applied to a well in a certain block where a sand plugging event occurs, and the construction plan is optimized, which specifically includes the following steps:

[0126] S1. Obtain the original construction data of the complete fracturing construction cycle of multiple wells and the geological data of the corresponding fracturing layers. The original construction data at least includes bottom hole pressure, displacement, sand ratio, and may also include parameters such as cumulative sand volume and cumulative liquid volume. The geological data at least includes maximum horizontal principal stress, minimum horizontal principal stress, formation pressure, pore pressure and vertical stress, and may also include parameters such as Poisson's ratio and Young's modulus.

[0127] Table 1 shows some of the original construction data.

[0128] Table 1

[0129]

[0130]

[0131] Table 2 shows some geological parameters corresponding to a certain fracturing layer.

[0132] Table 2

[0133]

[0134] S2. Divide the complete fracturing construction cycle into different intervals according to the construction data, and extract the construction data from the pre-fluid start stage to the pump stop stage;

[0135] As shown in Table 3, the complete fracturing construction cycle of a well is divided into five stages: pressure test, pre-fluid, sand-carrying fluid, displacement, and pump stop.

[0136] Table 3

[0137]

[0138]

[0139] S3, preprocess the above-obtained construction data and construct time series data through a sliding window algorithm, preprocess the construction data without bottom hole pressure and construct active independent variable data with geological data. In the application example, displacement, sand concentration and sand ratio are selected as active independent variable data.

[0140] Exemplarily, in this embodiment, the time window size is set to 180, the sliding window is 5, and the time step is 60.

[0141] S4. Build a CNN-LSTM-Attention hybrid neural network model to receive time series data, and build an LGBM gradient boosting model to receive active independent variable data;

[0142] Among them, the initial parameters of the CNN-LSTM-Attention hybrid neural network model are shown in Table 4.

[0143] Table 4

[0144]

[0145]

[0146] The initial parameters of the LGBM gradient boosting model are shown in Table 5.

[0147] Table 5

[0148]

[0149] S5. Use the obtained time series data and active independent variables to train the CNN-LSTM-Attention hybrid neural network model and the LGBM gradient boosting model respectively, and use the Bayesian optimization algorithm to optimize the hyperparameters of the model training process;

[0150] When training the model, in order to comprehensively evaluate the performance of the model, this embodiment uses the following indicators for comprehensive evaluation:

[0151] (1) Mean absolute percentage error (MAPE), whose value range is [0, +∞)

[0152]

[0153] (2) Mean absolute error (MAE), whose value range is [0, +∞).

[0154]

[0155] (3) Mean square error (MSE), whose value range is [0, +∞).

[0156]

[0157] Among them, y i is the actual bottom hole pressure, is the bottom hole pressure predicted by the model. The smaller the above three indicators, MAPE, MAE, and MSE, the better the model fitting effect.

[0158] In this embodiment, if Figure 4 As shown in the figure, the loss function of CNN-LSTM-Attention and CNN-LSTM training is compared. Figure 4 It can be seen that when CNN-LSTM-Attention and CNN-LSTM are trained, the loss function can decrease steadily and finally converge, but the model with attention mechanism (i.e. CNN-LSTM-Attention model) has a faster loss decrease rate and finally converges to a smaller loss value.

[0159] In this embodiment, the Bayesian optimization algorithm is used to optimize the hyperparameters of the model training process. Bayesian optimization is an effective method for global optimization, especially suitable for optimizing costly function evaluation scenarios. The specific process is as follows:

[0160] (Ⅰ) Select the prior distribution: Bayesian optimization first selects a prior distribution to represent the objective function. This distribution is usually a Gaussian distribution because it can well estimate the function value and its uncertainty. (Ⅱ) Collect initial data points: Before starting the iteration, you need to collect some initial data points. These data points can be randomly selected or selected based on some heuristic method. (Ⅲ) Update the posterior distribution: Use the initial data points and the prior distribution to update the posterior distribution through the Bayesian rule. The posterior distribution combines prior knowledge and observed data and reflects the current best estimate of the objective function. (Ⅳ) Select the next point for evaluation. (Ⅴ) Perform function evaluation. (Ⅵ) Repeatedly update the posterior distribution. (Ⅶ) Select the best solution.

[0161] After the optimization process is completed, the best solution is selected according to the posterior distribution. This is usually the point where the function value is maximized or the best point actually evaluated, that is, the optimal hyperparameters of the model.

[0162] like Figure 5 As shown in the figure, the prediction effects of the CNN-LSTM-Attention model and the LGBM model in this embodiment on the test set after obtaining the optimal hyperparameters are trained. The relative errors of the two models are 3.84% and 2.18% respectively.

[0163] S6. Obtain the construction data and geological data of a new fracturing well. When the sand-carrying stage is just beginning, use the trained model to make a multi-step prediction of the bottom hole pressure in the future.

[0164] The prediction effect of the prediction model on the pressure of the sand-carrying fluid stage in the new well is as follows: Figure 6 As shown in Figure 2, the final prediction is obtained by weighting the prediction values ​​of the two models. The relative error between the predicted value and the actual value is 1.21%, which is consistent with Figure 5 Compared with the prediction results of , the prediction accuracy is effectively improved, which has certain guiding significance for on-site construction decision-making.

[0165] S7. Initialize the weight values ​​of the prediction results of the two models, and calculate the root mean square error of the two models based on the prediction results of each round and the actual pressure. Use the reverse error method to update and iterate the weights of the prediction results of the two models to obtain a more balanced and accurate final prediction value.

[0166] The flowchart for initializing model weights, updating model weights and obtaining the final predicted value is as follows Figure 7 As shown in the figure, the idea of ​​gradient descent is used to modify and adjust the weights of the model in real time to achieve more accurate predictions.

[0167] S8. Modify the active independent variables and check the future predicted pressure to achieve bottom hole pressure control and provide construction optimization suggestions.

[0168] like Figure 7As shown, the present invention first collects geological data and construction data of the target block, uses a sliding window algorithm to obtain a time series of construction data, combines the pre-processed construction data and geological data to construct active independent variable data, constructs a CNN-LSTM-Attention model for receiving time series, constructs an LGBM model for receiving active independent variable data, and performs weighted optimization on the predicted values ​​of the two models to obtain the bottom hole pressure in the future. During the prediction process, model evaluation and weight update can be performed through real-time monitoring of construction data, thereby continuously optimizing the prediction model. At the same time, through the predicted bottom hole pressure, on-site construction personnel can also evaluate whether there is a risk of sand plugging when adjusting construction parameters, thereby providing a prevention and management method for on-site fracturing construction and ensuring the safety and economic benefits of oil and gas field development.

[0169] The principles and implementation methods of the present invention are described in this article using specific examples. The description of the above embodiments is only used to help understand the method and core idea of ​​the present invention. At the same time, for those skilled in the art, according to the idea of ​​the present invention, there will be changes in the specific implementation methods and application scope. In summary, the content of this specification should not be understood as limiting the present invention.

Claims

1. A construction pressure dynamic prediction method based on spatiotemporal data characteristics, characterized in that: The following steps are involved: S1. Obtaining original construction data of a complete fracturing construction cycle of multiple wells and geological data of corresponding fracturing layers; S2. Divide the complete fracturing construction cycle into different intervals according to the construction data, and extract the construction data from the start stage of the pre-fluid to the pump stop stage; S3, preprocessing the construction data obtained in step S2 and constructing time series data through a sliding window algorithm, and combining the construction data and geological data after preprocessing to construct active independent variable data, where the active independent variable data is data that can be actively adjusted to affect bottom hole pressure; S4. Establish a CNN-LSTM-Attention hybrid neural network model to receive time series data, and establish an LGBM gradient boosting model to receive active independent variable data; S5, using the time series data obtained in step S3 to train the CNN-LSTM-Attention hybrid neural network model, using the active independent variable data to train the LGBM gradient boosting model, and performing hyperparameter optimization on the training process; S6, obtaining construction data and geological data of a new fracturing well, and using the model trained in step S5 to make a multi-step prediction of the bottom hole pressure at the beginning of the sand-carrying stage; S7. Initialize the weight values ​​of the prediction results of the CNN-LSTM-Attention hybrid neural network model and the LGBM gradient boosting model, calculate the errors of the two models according to the prediction results of each round and the actual pressure, optimize the weights of the prediction results of the two models, and obtain the construction pressure prediction value.

2. The construction pressure dynamic prediction method based on spatiotemporal data characteristics according to claim 1, wherein the construction data in step S1 includes: displacement, sand concentration and sand ratio; geological data include: maximum horizontal principal stress, minimum horizontal principal stress, formation pressure, pore pressure and vertical stress.

3. The construction pressure dynamic prediction method based on spatiotemporal data characteristics according to claim 1, wherein the construction data is obtained in step S3 and preprocessed to obtain preprocessed data, wherein the preprocessing operation includes the following steps: S31, delete null values ​​and interpolate missing value data; S32, standardizing various construction data respectively; S32. Divide the standardized construction data according to a certain ratio to obtain a training data set, a verification data set, and a test data set.

4. The construction pressure dynamic prediction method based on spatiotemporal data characteristics according to claim 1, wherein the step S3 constructs the time series data by a sliding window algorithm, and comprises the following steps: The data of a part of time points in the past are constructed into a time window, and the number of time points contained in each subsequence is the size of the time window; By moving the time window forward by one or more time units in the entire data set, multiple time windows are formed.

5. The construction pressure dynamic prediction method based on spatiotemporal data characteristics as described in claim 1, wherein the active independent variable data in step S3 includes displacement, sand concentration, and sand ratio.

6. The method for dynamic prediction of construction pressure based on spatiotemporal data characteristics as claimed in claim 1, wherein in step S7, the weights of the prediction results of the two models are optimized using a reverse error method.

7. The construction pressure dynamic prediction method based on spatiotemporal data characteristics according to claim 1 further comprises step S8: Make changes to the active independent variables and view the predicted construction stress.

8. The construction pressure dynamic prediction method based on spatiotemporal data characteristics according to claim 7, wherein step S8 further comprises: S81, obtaining the displacement, sand concentration, sand ratio and geological data of n pieces of data before the backward prediction time point, and taking the average of the displacement, sand concentration and sand ratio respectively; S82, setting the active variable displacement, sand concentration, and sand ratio predicted for the next m time points to the mean value obtained in step S81, the geological data remains unchanged, and inputting them into the LGBM gradient boosting model for prediction; S83. When it is necessary to observe the predicted pressure trend under different construction parameters, the obtained mean is adjusted and then input into the LGBM gradient boosting model for prediction.

9. A computer device comprising a memory, a processor, and a computer program stored in the memory and running on the processor, characterized in that: When the processor executes the computer program, the steps of the construction pressure dynamic prediction method described in any one of claims 1 to 8 are implemented.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the steps of the construction pressure dynamic prediction method described in any one of claims 1 to 8 are implemented.

Images