Method for predicting performance parameters of oil-based drilling fluid

Abstract

The invention discloses an oil-based drilling fluid performance parameter prediction method, and relates to the technical field of oil-gas field development. According to the method, firstly, multiple training sets are obtained, each training set comprises a drilling fluid formula and temperature which serve as input and drilling fluid performance parameters which serve as output, then the training sets are used for training an Adaboost-BP prediction model, and the trained model can be used for prediction; the Adaboost-BP prediction model takes a three-layer BP neural network as a sub-model, and meanwhile, a feature enhancement-fusion module is arranged on a hidden layer of the sub-model. According to the method, multiple key indexes of the oil-based drilling fluid can be predicted at a time, and compared with a traditional method, the prediction efficiency is improved, and the application range is widened.

Description

Technical Field

[0001] This application relates to the field of oil and gas field development technology, and in particular to a method for predicting the performance parameters of oil-based drilling fluids. Background Technology

[0002] Drilling fluids play multiple roles in the drilling process, including cuttings carrying, cooling, lubrication, and pressure control. Their rheological properties directly affect the efficiency and safety of drilling operations. Drilling fluid performance parameters are crucial for optimizing drilling processes and preventing downhole accidents; therefore, the selection of drilling fluid formulations, especially oil-based drilling fluids, is a major challenge. Current technologies primarily involve manually sampling and measuring an oil-based drilling fluid formulation using experimental equipment such as a six-speed rotational viscometer. This process is time-consuming, yielding only a small amount of discrete data per day. Consequently, multiple trials are required to obtain an oil-based drilling fluid suitable for the current reservoir, resulting in parameter updates lagging significantly behind actual operating conditions.

[0003] In recent years, the rapid development of artificial intelligence and big data technologies has provided new ideas for predicting drilling fluid performance parameters. Researchers have attempted to predict drilling fluid rheological parameters using artificial intelligence methods and achieved good results. However, most studies have only used single prediction models or single optimization models without considering systematic optimization to improve the model's prediction accuracy. In addition, some studies have used image recognition technology to analyze drilling fluid flow states, achieving high prediction accuracy, but this requires magnetic stirring to generate stable images, making on-site deployment infeasible. Furthermore, existing intelligent monitoring systems generally face problems such as data quality fluctuations and difficulties in modeling multi-factor coupling. With the development of intelligent algorithms, many optimization and ensemble algorithms have been applied to intelligent prediction, greatly improving the predictive power and accuracy of models. Therefore, it is necessary to construct a drilling fluid performance parameter prediction method that is suitable for complex working conditions, possesses real-time performance, and high accuracy. Summary of the Invention

[0004] To address at least one of the aforementioned problems, this application provides a method for predicting the performance parameters of oil-based drilling fluids.

[0005] To achieve the above objectives, this application provides a method for predicting the performance parameters of oil-based drilling fluids, comprising the following steps: S1. Obtain multiple sets of oil-based drilling fluid formulations, test temperatures, and performance parameters. Match the formulations, temperatures, and performance parameters one by one to generate a sample set. Randomly select a portion of the samples in the sample set as the training set. S2. Establish an Adaboost-BP prediction model and train it using a training set to obtain a pre-trained model. The structure of the Adaboost-BP prediction model is as follows: establish k three-layer BP neural network sub-models, k≥4, optimize the hyperparameters of the sub-models based on DQN, optimize the parameters of the sub-models using the Adam optimizer, calculate the weights of the k sub-models, and sum the weighted outputs of the sub-models with their weights to obtain the output of the Adaboost-BP prediction model. The hidden layers of the three-layer BP neural network sub-models also include a feature enhancement-fusion module. The feature enhancement-fusion module includes a channel attention module, a depthwise separable convolution module, and a feature fusion module. The channel attention module performs channel weighting on the input features, the depthwise separable convolution module performs feature extraction and cross-channel fusion on the input features to obtain enhanced features, and the feature fusion module fuses the enhanced features. S3. Performance parameters of oil-based drilling fluids can be predicted based on pre-trained models.

[0006] The beneficial effects of the present invention are as follows: the method of the present invention can predict multiple key indicators of oil-based drilling fluid at one time, which improves prediction efficiency and applicability compared with traditional methods. Detailed Implementation

[0007] The technical solution of this application will be clearly described below with reference to the embodiments thereof. Obviously, the described embodiments are only some, not all, of the embodiments of this application. Other embodiments obtained by those skilled in the art based on the embodiments of this application without creative effort are all within the protection scope of this application.

[0008] This application provides a method for predicting the performance parameters of oil-based drilling fluids, including the following steps: S1. Obtain multiple sets of oil-based drilling fluid formulations, test temperatures, and performance parameters. Match the formulations, temperatures, and performance parameters one by one to generate a sample set. Randomly select a portion of the samples in the sample set as the training set. Specifically, in this embodiment, it is first necessary to obtain the formulation of the oil-based drilling fluid and its performance parameters under different temperature conditions. Generally speaking, the formulation of oil-based drilling fluid typically consists of organic clay, primary emulsifier, secondary emulsifier, lime, CaCl2 solution, base oil, and barite powder. Temperature is a crucial parameter affecting its performance, and these formulation and temperature parameters can be determined before testing. Furthermore, to minimize the influence of these parameters, in this embodiment, all materials in the formulation are uniformly converted to mass percentages.

[0009] Meanwhile, for the components of oil-based drilling fluids, such as base oil, this embodiment allows for sample selection and substitution; for example, diesel oil or white oil can be chosen as the base oil. The co-emulsifiers and auxiliary emulsifiers can also be adjusted; for example, petroleum sulfonate emulsifiers or fatty acid soap emulsifiers can be used. For the remaining components in the oil-based drilling fluid, such as organic clay, the specific amount added can be selected, or it can be added at all. When no organic clay is added, its addition amount is set to 0. Lime is used to adjust the pH, so other materials can be selected, such as calcium hydroxide or magnesium hydroxide. CaCl2 solution can be replaced with an aqueous solution of potassium chloride or even pure water. Barite powder is used to adjust the drilling fluid density and can be replaced with iron oxide powder or calcium carbonate powder. Oil-based drilling fluids mainly contain the above-mentioned components; therefore, during modeling, the binding temperatures of the seven important components in the formula need to be set as inputs. When a certain component is not in the formula, it is set to 0.

[0010] The most important performance parameters for oil-based drilling fluids are plastic viscosity, apparent viscosity, dynamic shear force, static shear force, consistency coefficient, flow index, density, and funnel viscosity. These performance parameters are affected by the oil-based drilling fluid formulation and temperature, and are thus dependent variables. In engineering, these performance parameters have a significant impact on the application scenarios and scope of oil-based drilling fluids. Among these performance parameters, density is measured using a densitometer, funnel viscosity is measured using a Marshall funnel viscometer, and apparent viscosity, plastic viscosity, dynamic shear force, static shear force, flow index, and consistency coefficient are measured using a six-speed rotary viscometer in conjunction with appropriate rheological models.

[0011] The training set should contain at least 1000 samples; if it contains fewer than 1000, the model's generalization ability will be limited. In practice, a training set of 2000 to 5000 samples is recommended; too many samples will increase costs.

[0012] S2. Establish an Adaboost-BP prediction model and train it using a training set to obtain a pre-trained model. The structure of the Adaboost-BP prediction model is as follows: establish k three-layer BP neural network sub-models, k≥4, optimize the hyperparameters of the sub-models based on DQN, optimize the parameters of the sub-models using the Adam optimizer, calculate the weights of the k sub-models, and sum the weighted outputs of the sub-models with their weights to obtain the output of the Adaboost-BP prediction model. The hidden layers of the three-layer BP neural network sub-models also include a feature enhancement-fusion module. The feature enhancement-fusion module includes a channel attention module, a depthwise separable convolution module, and a feature fusion module. The channel attention module performs channel weighting on the input features, the depthwise separable convolution module performs feature extraction and cross-channel fusion on the input features to obtain enhanced features, and the feature fusion module fuses the enhanced features. This step is the core of this embodiment. Conventional models can only predict one or two performance parameters of drilling fluid. When there are too many performance parameters, the prediction difficulty increases and the accuracy drops sharply, rendering it impractical. In contrast, this embodiment trains multiple three-layer BP neural networks using Adaboost and then performs a weighted summation, which is equivalent to multiple three-layer BP neural networks voting together, reducing the risk of overfitting from a single three-layer BP neural network.

[0013] The Adaboost-BP prediction model of this invention has an Adaboost model with a BP neural network as a sub-model. The difference lies in the fact that, in this embodiment, the BP neural network is a three-layer BP neural network with a relatively simple structure: one input layer, one hidden layer, and one output layer. The structures of the input and output layers are similar to those of a conventional BP neural network, and therefore will not be described in detail here.

[0014] For the hidden layer, the activation function is Hard-Swish. However, unlike conventional hidden layers, this embodiment includes a feature enhancement-fusion module. This module comprises a channel attention module, a depthwise separable convolution module, and a feature fusion module. The channel attention module suppresses redundancy, the depthwise separable convolution module extracts inter-component coupling relationships from the weighted features, and the feature fusion module prevents information distortion. These three modules are cascaded to output a hidden layer representation that enhances key components while maintaining stability, predicting eight drilling fluid performance parameters simultaneously.

[0015] The channel attention module consists of a global pooling layer and a one-dimensional convolution. First, global average pooling is performed on the input features. Then, the layout dependencies between channels are learned based on the one-dimensional convolution to generate channel weight vectors. The channel weight vectors are normalized using the Sigmoid function, and the normalized weight vectors are used to weight the original features by channels, thereby achieving adaptive feature recalibration. Through the channel attention module, eight components (seven components and one temperature) are scored. Dimensions with low scores are directly reduced, thereby suppressing redundancy.

[0016] The depthwise separable convolution module includes a depthwise convolutional layer and a pointwise convolutional layer. First, depthwise convolution is used to extract spatial features channel by channel. Then, pointwise convolution is used to fuse cross-channel information. This module then extracts the coupling relationship between components on the weighted features.

[0017] The feature fusion module is an existing module with two implementation methods: one is to fuse the enhanced features by adding them element by element, and the other is to fuse the enhanced features by concatenating the elements and then performing a 1×1 convolution.

[0018] To enhance the accuracy of the sub-model, this embodiment optimizes its hyperparameters using a Deep Q-Network (DQN), specifically by optimizing the learning rate and the number of hidden layer nodes. Furthermore, similar to the training of conventional backpropagation (BP) neural networks in this field, this embodiment employs Adam as the model optimizer.

[0019] Since the model in this embodiment of the invention is an Adaboost-BP prediction model based on a BP neural network, in addition to training the sub-models, it also involves the fusion of multiple sub-models: after training the i-th sub-model, where 1≤i≤k, the weighted error ε of the i-th sub-model is calculated. i Simultaneously calculate the weights of the i-th sub-model: And according to ε i Update the sample weights and train the next sub-model, continuing until all sub-models are trained; then, based on the weights of the k trained sub-networks... α i The weighted summation is the output of the Adaboost-BP prediction model.

[0020] S3. Performance parameters of oil-based drilling fluids can be predicted based on pre-trained models.

[0021] It should be understood that this application is not limited to the examples described above, and various modifications and changes can be made without departing from its scope. The true scope is indicated by this application.

Claims

1. A method for predicting performance parameters of oil-based drilling fluids, characterized in that, The method comprises the following steps: S1, obtaining multiple groups of oil-based drilling fluid formulas, test temperatures and performance parameters, corresponding the formulas, temperatures and performance parameters to generate a sample set, and randomly selecting part of the samples in the sample set as a training set; S2, establishing an Adaboost-BP prediction model and training the same by using the training set to obtain a pre-trained model; the structure of the Adaboost-BP prediction model is as follows: k three-layer BP neural network sub-models are established, k is greater than or equal to 4, the hyperparameters of the sub-models are optimized based on DQN, the parameters of the sub-models are optimized by using an Adam optimizer, the weights of the k sub-models are calculated, and the output results and the weights of the sub-models are weighted and summed to obtain the output of the Adaboost-BP prediction model; the hidden layer of the three-layer BP neural network sub-model further comprises a feature enhancement-fusion module, the feature enhancement-fusion module comprises a channel attention module, a depth separable convolution module and a feature fusion module, the input features are channel weighted by the channel attention module, the input features are extracted and cross-channel fused by the depth separable convolution module to obtain enhanced features, and the enhanced features are fused by the feature fusion module; S3, the performance parameters of the oil-based drilling fluid can be predicted based on the pre-trained model.

2. The method of claim 1, wherein, In S1, the drilling fluid formula comprises base oil and emulsifier, and the performance parameters are plastic viscosity, apparent viscosity, dynamic shear force, static shear force, consistency coefficient, flow index, density and funnel viscosity.

3. The method of claim 1, wherein, In the three-layer BP neural network sub-model, an input layer, a hidden layer and an output layer are included, and the activation function of the hidden layer is Hard-Swish.

4. The method of claim 1, wherein, The channel attention module firstly performs global average pooling on the input features, then learns the layout dependency relationship between channels based on one-dimensional convolution to generate a channel weight vector, normalizes the channel weight vector by using a Sigmoid function, and weights the original features by using the normalized weight vector.

5. The method of claim 1, wherein, The depth separable convolution module firstly extracts spatial features by using depth convolution channel by channel, and then fuses cross-channel information by point-by-point convolution.

6. The method of claim 1, wherein, In the feature fusion module, the enhanced features are fused by using an element-by-element addition method, or the elements are spliced and then the enhanced features are fused by using 1x1 convolution processing. In the feature fusion module, the enhanced features are fused by using an element-by-element addition method, or the elements are spliced and then the enhanced features are fused by using 1x1 convolution processing.

7. The method of claim 1, wherein, The method for weighted summation of the output results of the sub-models and the weights comprises the following steps: after training the i-th sub-model, 1≤i≤k, calculating the weighted error ε i of the i-th sub-model, simultaneously calculating the weight of the i-th sub-model: and updating the sample weight according to ε i , and training the next sub-model until the training of all the sub-models is completed; weighting and summing the k trained sub-networks based on the weights α i of the sub-networks, and obtaining the output of the Adaboost-BP prediction model.