Behavior pattern prediction method and device, equipment, storage medium and computer program product

Abstract

The invention relates to the technical field of geothermal resource utilization, and discloses a behavior pattern prediction method and device, equipment, a storage medium and a computer program product, and the method comprises the steps: obtaining historical behavior data of a geothermal well, and carrying out the normalization of the historical behavior data, and obtaining target input data; importing the target input data into a preset deep learning neural network model to obtain prediction data; and predicting the behavior pattern of the geothermal well according to the prediction data. According to the behavior pattern recognition method suitable for the geothermal well, existing production data of the geothermal well are combined as prior information, attribute fusion is carried out through a feedforward neural network algorithm, the future production condition of similar production environments or the same well is predicted, and scientific guidance is provided for efficient development of geothermal energy.

Description

Technical Field

[0001] This application relates to the field of geothermal resource utilization technology, and in particular to a method, apparatus, equipment, storage medium and computer program product for predicting behavior patterns. Background Technology

[0002] Currently, geothermal resource evaluation and exploration and development mainly rely on physical models established based on the heat and mass transfer modes of the formation, and gridded numerical simulations are performed. The accuracy of the simulation requires the support of physical parameters. However, the factors affecting the geothermal well production process are complex, and if the production boundaries of geothermal wells and precise grid division are required, huge computational costs are needed, making it difficult to widely popularize and promote geothermal production simulation. Summary of the Invention

[0003] The main purpose of this application is to provide a method, device, equipment, storage medium, and computer program product for predicting behavior patterns. It aims to solve the technical problems that current geothermal resource evaluation and exploration and development mainly rely on physical models established based on the heat and mass transfer modes of the formation to conduct gridded numerical simulations. The accuracy of the simulation requires the support of physical parameters, but the factors affecting the geothermal well production process are complex, and if the production boundary of the geothermal well and accurate grid division are required, huge computational costs are needed, making it difficult to widely popularize and promote geothermal production simulation.

[0004] To achieve the above objectives, this application proposes a behavior pattern prediction method, which includes:

[0005] Acquire historical behavior data of geothermal wells and normalize the historical behavior data to obtain target input data;

[0006] The target input data is imported into a preset deep learning neural network model to obtain prediction data;

[0007] The behavior pattern of the geothermal well is predicted based on the predicted data.

[0008] Optionally, before the step of inputting the target input data into a preset deep learning neural network model to obtain the predicted data, the method further includes:

[0009] An initial deep learning neural network model was constructed based on the feedforward neural network algorithm;

[0010] The initial deep learning neural network model is trained using preset training data to obtain training output values;

[0011] The parameters of the initial deep learning neural network model are updated based on the training output values ​​to obtain the target deep learning neural network model.

[0012] Optionally, the step of constructing the initial deep learning neural network model based on the feedforward neural network algorithm includes:

[0013] A multilayer perceptron is configured based on a feedforward neural network algorithm, wherein the multilayer perceptron includes an input layer, a hidden layer, and an output layer;

[0014] Based on the target input data, a target loss function and an optimizer are selected, and an initial deep learning neural network model is established using the multilayer perceptron, the loss function, and the optimizer.

[0015] Optionally, the step of setting up the multilayer perceptron based on the feedforward neural network algorithm includes:

[0016] The dimension of the input layer of the multilayer perceptron is set based on the number of features in the target input data.

[0017] Select an appropriate activation function to design the hidden layer, and configure the hidden layer with a preset number of neurons;

[0018] The output layer is defined using a linear activation function based on the requirements of the predicted task.

[0019] A multilayer perceptron is generated by connecting the input layer, the hidden layer, and the output layer using a feedforward neural network algorithm.

[0020] Optionally, the step of updating the parameters of the initial deep learning neural network model based on the training output value to obtain the target deep learning neural network model includes:

[0021] Calculate the first loss function based on the training output value;

[0022] The gradient of the first loss function with respect to the model parameters is determined based on the backpropagation algorithm;

[0023] The training data is selected in mini-batch using the mini-batch gradient descent method, and the parameters of the initial deep learning neural network model are adjusted according to the direction of the gradient to calculate the second loss function.

[0024] When the value of the second loss function drops to a preset loss threshold, the target deep learning neural network model is obtained based on the adjusted parameters.

[0025] Optionally, before the step of calculating the first loss function based on the training output value, the method further includes:

[0026] The initial loss function is regularized by adding a squared penalty term to the weights, resulting in the first loss function.

[0027] Furthermore, to achieve the above objectives, this application also proposes a behavior pattern prediction device, the behavior pattern prediction device comprising:

[0028] The data processing module is used to acquire historical behavior data of geothermal wells and preprocess the historical behavior data to obtain target input data;

[0029] The data prediction module is used to import the target input data into a preset deep learning neural network model to obtain predicted data;

[0030] The behavior prediction module is used to predict the behavior pattern of the geothermal well based on the prediction data.

[0031] In addition, to achieve the above objectives, this application also proposes a behavior pattern prediction device, the device comprising: a memory, a processor, and a computer program stored in the memory and executable on the processor, the computer program being configured to implement the steps of the behavior pattern prediction method as described above.

[0032] In addition, to achieve the above objectives, this application also proposes a storage medium, which is a computer-readable storage medium, on which a computer program is stored, and which, when executed by a processor, implements the steps of the behavior pattern prediction method described above.

[0033] In addition, to achieve the above objectives, this application also provides a computer program product, which includes a computer program that, when executed by a processor, implements the steps of the behavior pattern prediction method described above.

[0034] This application discloses a method for acquiring historical behavior data of geothermal wells, normalizing the historical behavior data to obtain target input data, importing the target input data into a preset deep learning neural network model to obtain prediction data, and predicting the behavior pattern of the geothermal well based on the prediction data. By combining existing production data of geothermal wells as prior information and using a feedforward neural network algorithm for attribute fusion, this method predicts the future production situation of similar production environments or the same well, providing scientific guidance for the efficient development of geothermal energy. Attached Figure Description

[0035] The accompanying drawings, which are incorporated in and form part of this specification, illustrate embodiments consistent with this application and, together with the description, serve to explain the principles of this application.

[0036] To more clearly illustrate the technical solutions in the embodiments of this application or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, for those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0037] Figure 1 This is a flowchart illustrating the first embodiment of the behavior pattern prediction method of this application;

[0038] Figure 2 This is a flowchart illustrating the technical process of a geothermal well behavior pattern recognition method based on deep learning neural networks.

[0039] Figure 3 This is a flowchart illustrating the second embodiment of the behavior pattern prediction method of this application;

[0040] Figure 4 A comparison chart of temperature training data and actual data;

[0041] Figure 5 This is a flowchart illustrating the third embodiment of the behavior pattern prediction method of this application;

[0042] Figure 6 This is a software architecture diagram for a geothermal well behavior pattern recognition method based on deep learning neural networks.

[0043] Figure 7 This is a flowchart illustrating the fourth embodiment of the behavior pattern prediction method of this application;

[0044] Figure 8 This is a schematic diagram of the module structure of the behavior pattern prediction device according to an embodiment of this application;

[0045] Figure 9 This is a schematic diagram of the device structure of the hardware operating environment involved in the behavior pattern prediction method in the embodiments of this application.

[0046] The purpose, features, and advantages of this application will be further explained in conjunction with the embodiments and with reference to the accompanying drawings. Detailed Implementation

[0047] It should be understood that the specific embodiments described herein are merely illustrative of the technical solutions of this application and are not intended to limit this application.

[0048] To better understand the technical solution of this application, a detailed description will be provided below in conjunction with the accompanying drawings and specific implementation methods.

[0049] The main solution of this application embodiment is: to acquire historical behavior data of geothermal wells and normalize the historical behavior data to obtain target input data; to import the target input data into a preset deep learning neural network model to obtain prediction data; and to predict the behavior pattern of the geothermal well based on the prediction data.

[0050] With the depletion of fossil fuels and the increasing severity of environmental problems, geothermal energy, as a new type of renewable and clean energy with huge reserves, has a wide range of applications. Geothermal energy is continuously supplied from the Earth's interior, making it a sustainable energy source. The greenhouse gas emissions generated during its utilization are far lower than those from fossil fuels, and it is unaffected by weather or day / night cycles, providing a stable energy supply. However, limitations imposed by the complex underground rock and soil environment, the complex heat transfer mechanism within pipelines, the high drilling difficulty, and the high initial investment in geothermal utilization result in insufficient geothermal well data and a lack of actual engineering data. To verify and predict geothermal well behavior patterns, a deep learning neural network geothermal well pressure model is established using existing production data for simulation and prediction of the target layer pressure of the geothermal well. The complexity of underground rock and soil structures and heat transfer mechanisms makes it difficult to predict the behavior patterns of geothermal wells and establish accurate physical models, which imposes certain limitations on geothermal resource evaluation and exploration and development. Existing methods mainly establish physical models based on the heat and mass transfer modes of the formation and conduct gridded numerical simulations, such as the traditional finite element method, finite difference method, and finite volume method. The accuracy of these methods requires the support of physical parameters, but the factors affecting the geothermal well production process are complex. Moreover, if the production boundaries of geothermal wells and accurate grid division are required, huge computational costs are needed, making it difficult to widely popularize and promote geothermal production simulation.

[0051] Therefore, this application provides a method for recognizing the behavior patterns of geothermal wells, which can reduce the computational cost required for numerical simulation and accurately predict geothermal well behavior patterns through historical data fitting, providing scientific guidance for efficient geothermal development. By combining existing geothermal well production data as prior information and using a feedforward neural network algorithm for attribute fusion, it predicts the future production conditions of similar production environments or the same well.

[0052] It should be noted that the executing entity in this embodiment can be a computing service device with data processing, network communication, and program execution functions, such as a tablet computer, personal computer, or mobile phone, or an electronic device capable of performing the above functions. The following description uses a geothermal well behavior observation system as an example to illustrate this embodiment and the subsequent embodiments.

[0053] Based on this, embodiments of this application provide a behavior pattern prediction method, referring to... Figure 1 , Figure 1 This is a flowchart illustrating the first embodiment of the behavior pattern prediction method of this application.

[0054] In this embodiment, the behavior pattern prediction method includes:

[0055] Step S10: Obtain historical behavior data of geothermal wells and normalize the historical behavior data to obtain target input data.

[0056] It should be noted that geothermal wells are special wells used for developing geothermal resources. They are drilled deep into the formation to obtain geothermal resources, such as directly extracting geothermal water or steam, or injecting a medium to exchange heat with the formation and then extracting the water for power generation or heating. Historical behavior data refers to the operational data of a geothermal well over a past period, recording its working status and performance, including but not limited to parameters such as temperature, flow rate, pressure, and energy output. Normalization refers to the process of transforming data into a specific range or standard to enable effective comparison, analysis, or processing, ensuring a uniform scale for the data during calculations or model building. Target input data is the normalized historical behavior data of the geothermal well, which will serve as input for subsequent analysis, prediction, or modeling.

[0057] It should be understood that historical behavior data of geothermal wells can originate from geothermal well monitoring systems or recording devices. These systems or devices record the operating parameters of geothermal wells in real time and store them as historical data. By accessing these monitoring systems or devices, operational data of the geothermal well over a past period can be obtained. After obtaining the raw data, it is necessary to select an appropriate normalization method based on the characteristics of the data and the needs of subsequent analysis. This includes linear transformation, maximum-minimum normalization, standardization, etc., to prevent excessively large or small differences in data feature values ​​from affecting the model and causing overfitting.

[0058] In one example, data is normalized by subtracting the mean and dividing by the standard deviation. This reduces overfitting and improves the convergence speed during model training. The parameters include: x old Original data, x new Normalized data, μ is the mean, σ is the standard deviation. The formula for calculating the target input data is:

[0059]

[0060] The parameters include: x old Original data, x new Normalized data, μ is the mean of the data, and σ is the standard deviation of the data.

[0061] Step S20: Import the target input data into a preset deep learning neural network model to obtain prediction data.

[0062] It's important to understand that a deep learning neural network model is a complex machine learning model composed of neurons (or nodes) across multiple layers (input layer, hidden layer, and output layer). These neurons interact through weighted connections and activation functions to learn data representations and patterns, enabling them to handle high-dimensional data and automatically extract features. Predictive data, generated by the deep learning neural network model from the input data (i.e., the target input data), can be estimates of future states, trends, or outcomes, and can be used for decision support, planning, risk assessment, and more.

[0063] It should be understood that prediction methods based on deep learning neural network models are a type of machine learning method based on a large amount of data. They are used to learn features and recognize behaviors by constructing deep neural networks, including an input layer, multiple hidden layers and an output layer. The connections between neurons in each layer are adjusted through a large amount of training, thereby learning the features and patterns of the data and predicting the target region.

[0064] Understandably, before making predictions, it's necessary to select or define a deep learning neural network model suitable for the current task. This involves choosing the model's architecture (e.g., convolutional neural networks, recurrent neural networks, fully connected neural networks), the number and type of layers, activation functions, loss functions, and optimizers. Based on training data, optimization methods such as backpropagation and gradient descent are used to continuously adjust the model's weights and bias parameters to minimize the loss function. Once the model is trained and reaches a satisfactory performance level, it can be used to generate prediction data.

[0065] Step S30: Predict the behavior pattern of the geothermal well based on the predicted data.

[0066] It should be understood that after obtaining predictive data about geothermal wells (including predicted values ​​of key parameters such as future temperature, flow rate, and pressure) from deep learning neural network models, statistical analysis and visualization tools (such as time series analysis, regression analysis, scatter plots, and line graphs) are used to identify trends and patterns in the predictive data. Special attention is paid to predicted values ​​that may indicate changes in geothermal well performance or abnormal behavior, such as sudden temperature increases or decreases, or significant changes in flow rate. Based on the analysis results of the predictive data, a time series-based predictive model is established to predict the changing trends of geothermal well behavior patterns (such as temperature and flow rate) over a future period.

[0067] For ease of understanding, the following examples are provided, but they do not limit this application. In one example, refer to... Figure 2 , Figure 2This is a flowchart illustrating the technical process of a geothermal well behavior pattern recognition method based on deep learning neural networks. For the collected production data, data normalization and regularization are first performed to optimize for overfitting outside of the behavior patterns. The processed data is then fed into a neural network model for training, where the neural network learns the behavior patterns of geothermal wells. In the neural network, activation functions control the activation state of neurons, helping the model learn complex nonlinear relationships. The trained neural network model calculates the output value based on the input data. By comparing the model's output value with the actual value, a loss function is calculated. Based on the obtained loss function, conventional DL neural network optimization is performed, including forward and backward propagation, updating the neural network parameters to reduce prediction errors. This process is repeated multiple times until the model converges, i.e., the loss function reaches its minimum or the parameter updates are no longer significant. The final DL (Deep-Learning) model is then obtained.

[0068] In this embodiment, historical behavior data of geothermal wells is acquired and normalized to obtain target input data. This target input data is then imported into a preset deep learning neural network model to obtain prediction data. The behavior pattern of the geothermal well is predicted based on the prediction data. This method, applicable to geothermal well behavior pattern recognition, combines existing production data of geothermal wells as prior information and utilizes a feedforward neural network algorithm for attribute fusion to predict the future production conditions of similar production environments or the same well, providing scientific guidance for efficient geothermal development.

[0069] Reference Figure 3 , Figure 3 This is a flowchart illustrating the second embodiment of the behavior pattern prediction method of this application. Based on the first embodiment described above, a second embodiment of the behavior pattern prediction method of this application is proposed.

[0070] In the second embodiment, before step S20, the method further includes:

[0071] Step S201: Construct an initial deep learning neural network model based on the feedforward neural network algorithm.

[0072] It's important to understand that a feedforward neural network is a basic artificial neural network architecture. Its neurons are arranged in layers, with each neuron connected only to neurons in the previous layer, receiving the output of the previous layer and then passing it to the next layer. There is no feedback between layers. An initial deep learning neural network model is a neural network model that has not yet been trained or has only undergone preliminary training; it only contains the basic structure and parameters of the neural network.

[0073] It should be understood that when building an initial deep learning neural network model, the problem to be solved and the available dataset must first be clearly defined. The dataset should contain sufficient and representative samples. A suitable network structure should be selected based on the task requirements and the characteristics of the dataset. For example, a convolutional neural network can be chosen for image classification tasks, while a recurrent neural network can be chosen for sequence prediction tasks. After selecting a suitable network structure, the weights and biases in the network are initialized using random initialization or other initialization strategies. An appropriate loss function is selected to measure the difference between the model's predictions and the actual results, and an optimizer is chosen to update the model's parameters to minimize the loss function. Finally, the optimization algorithm used during training is determined.

[0074] Step S202: Train the initial deep learning neural network model using preset training data and obtain the training output value.

[0075] It's important to understand that pre-set training data is a dataset that has already been prepared and is used to train a deep learning neural network model. This data typically contains input features and corresponding target outputs (or labels), used to guide the model in learning how to predict outputs based on the input data. Training output values ​​are the values ​​calculated by the deep learning neural network model during training, based on the input data (i.e., the input features in the pre-set training data) and the current network parameters. These output values ​​are usually compared with the target output (the labels in the pre-set training data) to calculate the loss function and update the network parameters. As training progresses, the training output values ​​gradually approach the target output, thereby improving the model's prediction accuracy.

[0076] Step S203: Update the parameters of the initial deep learning neural network model according to the training output value to obtain the target deep learning neural network model.

[0077] Understandably, during deep learning training, the parameters of the initial deep learning neural network model can be continuously updated using optimization strategies such as backpropagation and gradient descent, based on the difference between the training output values ​​(i.e., model predictions) and the true target values ​​(labels). This process is repeated on the training dataset until a predetermined number of training epochs is reached or the loss function converges below a certain threshold. Furthermore, in each iteration, a new batch of training data is used to calculate the loss and update the parameters, which helps the model learn the global features of the data and avoid overfitting. After sufficient training iterations, when the parameters of the initial deep learning neural network model are optimized to produce satisfactory prediction results, the target deep learning neural network model is obtained.

[0078] For ease of understanding, the following examples are provided, but they do not limit this application. In one example, refer to... Figure 4 , Figure 4This is a comparison chart of temperature training data and actual data. Using flow test data from one day of a geothermal energy research frontier observation tower project as training data, the relationship between temperature, ambient temperature, and ambient pressure over time was recorded to establish a forward feedback neural network. Ambient temperature, ambient pressure, and time were defined as the three variables in the input layer. Normalization and regularization were performed on the input layer to reduce the phenomenon of individual values ​​influencing the overall trend due to excessive model complexity, also known as overfitting. To satisfy the non-linear relationship of this trend, a ReLU (Rectified Linear Unit) activation function was set. The number of hidden layers was also set; to accommodate the complex characteristics of the data changes and ensure accurate prediction results, three hidden layers were selected, each containing 128, 64, and 32 neurons respectively. The target data for the output layer, geothermal well temperature, was determined. The training data was imported into the DL neural network model for training and learning. Based on the learning results, the loss function was calculated using forward and backward propagation with the output data and the reference output layer data. Using this function as a basis, gradient descent was used to gradually adjust the hidden layer parameters to improve the neural network's fitting effect. Repeat this process until the model stabilizes (the loss value shows no significant change) or the loss value reaches the expected value. Based on the final model, behavioral pattern predictions are performed. The comparison between the predicted geothermal well temperature data and the time dataset is shown in Figure 4. The horizontal axis represents time, and the vertical axis represents temperature. Blue dots represent predicted data, and the red curve represents predicted data. The figure shows excellent fitting; the calculated coefficient of determination reaches 99.75%, indicating a high degree of fit. It can be used to predict the behavioral patterns of this geothermal well or similar geothermal wells.

[0079] In this embodiment, an initial deep learning neural network model is constructed based on a feedforward neural network algorithm; the initial deep learning neural network model is trained using preset training data to obtain training output values; the parameters of the initial deep learning neural network model are updated according to the training output values ​​to obtain a target deep learning neural network model. This allows the model to gradually learn and adapt to the characteristics of the training data during the training process, providing a systematic method to optimize and adjust the deep learning model to improve its accuracy in predicting geothermal well behavior patterns.

[0080] Reference Figure 5 , Figure 5 This is a flowchart illustrating the third embodiment of the behavior pattern prediction method of this application. Based on the second embodiment described above, a third embodiment of the behavior pattern prediction method of this application is proposed.

[0081] In the third embodiment, step S201 includes:

[0082] Step S2011: Set up a multilayer perceptron based on the feedforward neural network algorithm, wherein the multilayer perceptron includes an input layer, a hidden layer and an output layer.

[0083] It's important to understand that a multilayer perceptron (MLP) is a feedforward artificial neural network model composed of multiple layers, including an input layer, at least one hidden layer, and an output layer. Each layer consists of multiple neurons that interact through weighted connections, enabling it to handle nonlinear problems. It establishes a mapping between input and output by learning appropriate weights and biases. The input layer is the first layer of the MLP, responsible for receiving external input data. The hidden layer, located between the input and output layers, contains multiple neurons, each receiving input signals from neurons in the previous layer and processing them using an activation function (such as ReLU) to produce an output. The output layer is the last layer of the MLP, responsible for generating the model's output. The number of neurons in the output layer corresponds to the output requirements of the problem, and can be a single value or multiple categories. The feedforward neural network algorithm is a neural network training algorithm where information propagates unidirectionally from the input layer to the output layer. Weights and biases are adjusted using backpropagation to minimize the loss function.

[0084] Of course, in order to quickly construct a deep learning model that can capture the complex characteristics of geothermal well behavior patterns and improve the accuracy of predictions and the generalization ability of the model, step S2011 may include:

[0085] The dimension of the input layer of the multilayer perceptron is set based on the number of features of the target input data; an appropriate activation function is selected to design the hidden layer, and a number of neurons are configured for the hidden layer; the output layer is defined by a linear activation function according to the requirements of the prediction task; the input layer, the hidden layer and the output layer are connected based on a feedforward neural network algorithm to generate the multilayer perceptron.

[0086] It's important to understand that the input layer dimension refers to the type of input data in the input layer, typically corresponding to the number of features in the target input data. Activation functions in neural networks are used to introduce non-linearity, helping the network learn complex patterns. These include the ReLU linear rectified function, the Sigmoid function, and the Tanh hyperbolic tangent function. Neurons are the basic units of a neural network, responsible for receiving input signals, performing weighted summations, applying activation functions, and generating output signals. Linear activation functions are a special type of activation function whose output equals the weighted sum of the input signals (without bias or with a constant bias term), and are commonly used in the output layer.

[0087] For ease of understanding, the following examples are provided, but they do not limit this application. In one example, refer to... Figure 6 , Figure 6This is a software architecture diagram for a geothermal well behavior pattern recognition method based on deep learning neural networks. Based on the feature information of the input data (ambient pressure x1, time x2, and ambient temperature x3), a feedforward neural network is used to determine a multilayer perceptron consisting of a 3-dimensional input layer, three hidden layers composed of ReLU activation functions, and an output layer with a linear activation function. The 3-dimensional input layer includes changes in time, ambient temperature, and ambient pressure. The first hidden layer i contains 128 neurons; the second hidden layer j contains 64 neurons; and the third hidden layer k contains 32 neurons. These three hidden layers are neural networks controlled by ReLU activation functions. The output layer is controlled by a linear activation function, and the output value is the geothermal well temperature y1.

[0088] Step S2012: Select a target loss function and optimizer based on the target input data, and establish an initial deep learning neural network model through the multilayer perceptron, the loss function, and the optimizer.

[0089] It's important to note that the objective loss function, in deep learning, is used to quantify the difference or error between the model's predicted values ​​and the actual values. The loss function is the objective of model optimization; minimizing the loss function improves the model's predictive accuracy. An optimizer is an algorithm used to update the weights and biases of a neural network to minimize the loss function; it can include stochastic gradient descent, adaptive momentum estimation, root mean square propagation, etc.

[0090] Understandably, an appropriate loss function can be chosen based on the requirements of the prediction task (such as classification, regression, etc.). For example, for binary classification problems, the cross-entropy loss function can be chosen; for regression problems, the mean squared error loss function can be chosen. Simultaneously, an appropriate optimizer is selected based on the model's complexity and the size of the training data. After selecting the appropriate activation function and optimizer, hidden layers are designed, and the output layer is defined using a linear activation function according to the requirements of the prediction task. Finally, a feedforward neural network algorithm is used to connect the input layer, hidden layers, and output layer to form a complete neural network structure.

[0091] For ease of understanding, the following examples are provided, but they do not limit this application. In one example,

[0092] Apply the ReLU (Rectified Linear Unit) activation function to the hidden layer:

[0093]

[0094] The ReLU function is a one-sided inhibition method that prevents gradient saturation when the input is positive. It has relatively low computational cost and can suppress negative values ​​on one side, thus giving the neural network sparsity. This provides an advantage in suppressing overfitting and allows the neural network to converge faster.

[0095] Apply a linear activation function to the output layer:

[0096] h(x) = x

[0097] The function does not introduce nonlinearity and is mainly used for the output layer of regression problems.

[0098] The above steps establish a feedback-forward deep neural network model, confirming the number of hidden layers, the number of neurons, and the type of neuron activation function. A loss function and gradient descent are then introduced, with parameters including: n (number of samples), y (number of samples), and so on. true Sample value, y pred Predicted values, θ model parameters (weights and biases), L(θ) loss function, The gradient of the loss function with respect to the parameter θ, and the learning rate α, which determines the step size for each update.

[0099] During training, parameters are automatically adjusted based on sample labels and neural network output values. When the consistency between the output value and the sample label is low, the parameter adjustment is larger, and vice versa. To evaluate the consistency between the neural network output and the sample label, mean squared error (MSE) is chosen as the loss function.

[0100]

[0101] The optimizer employs gradient descent. Since the gradient of a function is the direction of its fastest growth, the opposite direction of the gradient is the direction of its fastest decrease. Utilizing this property, the neural network's learning parameters are updated along the opposite direction of the gradient to minimize the loss function value. The parameters are gradually adjusted to reduce the error.

[0102]

[0103] Where, θ new For the adjusted model parameters, θ old These are the model parameters before adjustment. This represents the gradient of the sample value with respect to the parameter θ.

[0104] In this embodiment, a multilayer perceptron is configured based on a feedforward neural network algorithm. The multilayer perceptron includes an input layer, a hidden layer, and an output layer. A target loss function and an optimizer are selected based on the target input data, and an initial deep learning neural network model is established using the multilayer perceptron, the loss function, and the optimizer. Configuring the multilayer perceptron based on the feedforward neural network algorithm allows the model to flexibly adjust its network structure according to the number of features in the target input data and the requirements of the prediction task.

[0105] Reference Figure 7 , Figure 7 This is a flowchart illustrating the fourth embodiment of the behavior pattern prediction method of this application. Based on the third embodiment described above, the fourth embodiment of the behavior pattern prediction method of this application is proposed.

[0106] In the fourth embodiment, step S203 includes:

[0107] Step S2031: Calculate the first loss function based on the training output value.

[0108] Understandably, the first loss function is the output result calculated based on the parameters of the current model to predict the input data. The initial loss value between the current prediction result and the sample result obtained according to the loss function formula is the starting point for model optimization.

[0109] Of course, to make the model simpler, smoother, and more universal, the following steps are included before step S2031:

[0110] The initial loss function is regularized by adding a squared penalty term to the weights, resulting in the first loss function.

[0111] It should be noted that the initial loss function refers to the basic loss function without any modification or adjustment, used to measure the difference between the model's predicted values ​​and the actual values. The squared penalty term for the weights is a technique used to prevent the model from overfitting. It works by penalizing the squared values ​​of the model weights (i.e., weighted summation), thereby encouraging the model weights to keep their values ​​small.

[0112] In one example, the sum of squares of the model weights W is added to the loss function, calculated as follows:

[0113]

[0114] The parameters include: L old Original loss function, L new The loss function after regularization, W iModel weights, a hyperparameter of λ-regularization strength, are added to the model's loss function as a squared penalty term to reduce noise in the training data or overfitting to specific patterns. This makes the model simpler, smoother, and improves its generalizability. Regularization also suppresses large weight values, preventing over-reliance on extreme features during training and making the model more robust. When using gradient descent optimization, regularization decays the weights with each update, thus contributing to the convergence stability of the model parameters.

[0115] Step S2032: Determine the gradient of the first loss function with respect to the model parameters based on the backpropagation algorithm.

[0116] It's important to note that backpropagation is a method used to calculate the gradient of each parameter in a neural network with respect to the loss function. Through backpropagation, we can understand the impact of each parameter on model performance, thereby guiding parameter updates. The gradient is the derivative of the loss function with respect to the model parameters, representing the speed and direction of the change in the loss function value with respect to the parameters.

[0117] Understandably, when determining the gradient of the first loss function relative to the model parameters based on the backpropagation algorithm, the error term (also called sensitivity or gradient) of each neuron is calculated layer by layer from the output layer. The error term represents the contribution of the current neuron to the final loss function. At the same time, the chain rule is used to propagate the error term from the output layer to the input layer, and during the propagation process, the gradient (i.e., partial derivative) of the weights of each neuron with respect to the loss function is calculated.

[0118] Step S2033: Select a small batch of training data using mini-batch gradient descent, adjust the parameters of the initial deep learning neural network model according to the direction of the gradient, and calculate the second loss function.

[0119] It's important to understand that mini-batch gradient descent is an optimization algorithm that accelerates the training process and reduces memory consumption by updating model parameters using mini-batch training data in each iteration. The second loss function is the recalculated loss value obtained after one or more updates to the model parameters. It is used to evaluate the model's performance after the updates.

[0120] It should be understood that when using mini-batch gradient descent to select training data, the training dataset is divided into several mini-batches, each containing a number of training samples. The number of samples in a mini-batch is typically set to a power of 2, such as 16, 32, or 64, which is beneficial for GPU acceleration. The model parameters are adjusted according to the direction of the gradient, following the direction that reduces the loss function. This process is repeated for all mini-batches until the loss function converges or the preset number of iterations is reached.

[0121] Step S2034: When the value of the second loss function drops to a preset loss threshold, the target deep learning neural network model is obtained based on the adjusted parameters.

[0122] Understandably, when the value of the second loss function drops to a preset loss threshold (also known as the stopping criterion or convergence condition), it means that the training process of the deep learning neural network model has made sufficient progress, and the model's performance has reached the expected level. When the stopping condition is met, the parameters of the current deep learning neural network model are saved, including the weights and biases of all layers, as well as any other trainable parameters.

[0123] It should be understood that, in addition to the loss threshold, other stopping conditions can be considered, such as the number of iterations, performance on the validation set no longer improving, etc. These conditions can be used in combination to ensure that the training process of the deep learning neural network model can proceed smoothly and effectively.

[0124] In this embodiment, a first loss function is calculated based on the training output value; the gradient of the first loss function relative to the model parameters is determined based on the backpropagation algorithm; a mini-batch of training data is selected using mini-batch gradient descent, and the parameters of the initial deep learning neural network model are adjusted according to the direction of the gradient, and a second loss function is calculated; when the value of the second loss function decreases to a preset loss threshold, the target deep learning neural network model is obtained based on the adjusted parameters. By introducing the calculation of the loss function, the determination of the gradient, the application of mini-batch gradient descent, and the setting of the loss threshold, the model can gradually reduce prediction errors during training until the preset loss threshold is reached.

[0125] It should be noted that the above examples are only for understanding this application and do not constitute a limitation on the behavior pattern prediction method of this application. Any simple modifications based on this technical concept are within the protection scope of this application.

[0126] This application also provides a behavior pattern prediction device, please refer to... Figure 8 The behavior pattern prediction device includes:

[0127] The data processing module 10 is used to acquire historical behavior data of geothermal wells and preprocess the historical behavior data to obtain target input data.

[0128] The data prediction module 20 is used to import the target input data into a preset deep learning neural network model to obtain prediction data;

[0129] The behavior prediction module 30 is used to predict the behavior pattern of the geothermal well based on the prediction data.

[0130] The behavioral pattern prediction device provided in this application, employing the behavioral pattern prediction method described in the above embodiments, can solve the current problem that geothermal resource evaluation and exploration mainly rely on establishing physical models based on the heat and mass transfer modes of the formation for gridded numerical simulation. The accuracy of the simulation requires support from physical parameters, but the influencing factors of geothermal well production processes are complex, and meeting the production boundaries and precise grid division of geothermal wells requires enormous computational costs, making it difficult to widely promote and popularize geothermal production simulation. Compared with the prior art, the beneficial effects of the behavioral pattern prediction device provided in this application are the same as those of the behavioral pattern prediction method provided in the above embodiments, and other technical features in the behavioral pattern prediction device are the same as those disclosed in the methods of the above embodiments, and will not be repeated here.

[0131] This application provides a behavior pattern prediction device, which includes: at least one processor; and a memory communicatively connected to the at least one processor; wherein the memory stores instructions executable by the at least one processor, which are executed by the at least one processor to enable the at least one processor to perform the behavior pattern prediction method in Embodiment 1 above.

[0132] The following is for reference. Figure 9 The diagram illustrates a structural schematic suitable for implementing the behavior pattern prediction device of the embodiments of this application. The behavior pattern prediction device in the embodiments of this application may include, but is not limited to, mobile terminals such as mobile phones, laptops, digital broadcast receivers, PDAs (Personal Digital Assistants), PADs (Portable Application Description), PMPs (Portable Media Players), in-vehicle terminals (e.g., in-vehicle navigation terminals), and fixed terminals such as digital TVs and desktop computers. Figure 9 The behavior pattern prediction device shown is merely an example and should not impose any limitations on the functionality and scope of use of the embodiments of this application.

[0133] like Figure 9As shown, the behavior pattern prediction device may include a processing unit 1001 (e.g., a central processing unit, a graphics processing unit, etc.) that can perform various appropriate actions and processes according to a program stored in a read-only memory (ROM) 1002 or a program loaded from a storage device 1003 into a random access memory (RAM) 1004. The RAM 1004 also stores various programs and data required for the operation of the behavior pattern prediction device. The processing unit 1001, ROM 1002, and RAM 1004 are interconnected via a bus 1005. An input / output (I / O) interface 1006 is also connected to the bus. Typically, the following systems can be connected to the I / O interface 1006: input devices 1007 including, for example, a touchscreen, touchpad, keyboard, mouse, image sensor, microphone, accelerometer, gyroscope, etc.; output devices 1008 including, for example, a liquid crystal display (LCD), speaker, vibrator, etc.; storage devices 1003 including, for example, magnetic tape, hard disk, etc.; and communication devices 1009. Communication device 1009 allows the behavior pattern prediction device to communicate wirelessly or wiredly with other devices to exchange data. While the figure shows behavior pattern prediction devices with various systems, it should be understood that implementation or possession of all the systems shown is not required. More or fewer systems may be implemented alternatively.

[0134] Specifically, according to the embodiments disclosed in this application, the processes described above with reference to the flowcharts can be implemented as computer software programs. For example, embodiments disclosed in this application include a computer program product comprising a computer program carried on a computer-readable medium, the computer program containing program code for performing the methods shown in the flowcharts. In such embodiments, the computer program can be downloaded and installed from a network via a communication device, or installed from storage device 1003, or installed from ROM 1002. When the computer program is executed by processing device 1001, it performs the functions defined in the methods of the embodiments disclosed in this application.

[0135] The behavioral pattern prediction device provided in this application, employing the behavioral pattern prediction method described in the above embodiments, can solve the current problem that geothermal resource evaluation and exploration mainly rely on establishing physical models based on the heat and mass transfer modes of the formation for gridded numerical simulation. The accuracy of the simulation requires support from physical parameters, but the influencing factors of geothermal well production processes are complex, and meeting the production boundaries and precise grid division of geothermal wells requires enormous computational costs, making it difficult to widely promote and popularize geothermal production simulation. Compared with the prior art, the beneficial effects of the behavioral pattern prediction device provided in this application are the same as those of the behavioral pattern prediction method provided in the above embodiments, and other technical features of this behavioral pattern prediction device are the same as those disclosed in the previous embodiment method, and will not be repeated here.

[0136] It should be understood that the various parts disclosed in this application can be implemented using hardware, software, firmware, or a combination thereof. In the description of the above embodiments, specific features, structures, materials, or characteristics can be combined in any suitable manner in one or more embodiments or examples.

[0137] The above description is merely a specific embodiment of this application, but the scope of protection of this application is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in this application should be included within the scope of protection of this application. Therefore, the scope of protection of this application should be determined by the scope of the claims.

[0138] This application provides a computer-readable storage medium having computer-readable program instructions (i.e., a computer program) stored thereon, the computer-readable program instructions being used to execute the behavior pattern prediction method in the above embodiments.

[0139] The computer-readable storage medium provided in this application may be, for example, a USB flash drive, but is not limited to, electrical, magnetic, optical, electromagnetic, infrared, or semiconductor systems, devices, or any combination thereof. More specific examples of computer-readable storage media may include, but are not limited to: electrical connections having one or more wires, portable computer disks, hard disks, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM or flash memory), optical fiber, portable compact disk read-only memory (CD-ROM), optical storage devices, magnetic storage devices, or any suitable combination thereof. In this embodiment, the computer-readable storage medium may be any tangible medium containing or storing a program that can be used by or in conjunction with an instruction execution system, system, or device. The program code contained on the computer-readable storage medium may be transmitted using any suitable medium, including but not limited to: wires, optical cables, RF (Radio Frequency), etc., or any suitable combination thereof.

[0140] The aforementioned computer-readable storage medium may be included in the behavior pattern prediction device; or it may exist independently and not assembled into the behavior pattern prediction device.

[0141] The aforementioned computer-readable storage medium carries one or more programs, which, when executed by the behavior pattern prediction device, cause the behavior pattern prediction device to perform the behavior pattern prediction method as described above.

[0142] Computer program code for performing the operations of this application can be written in one or more programming languages ​​or a combination thereof, including object-oriented programming languages ​​such as Java, Smalltalk, and C++, and conventional procedural programming languages ​​such as the "C" language or similar programming languages. The program code can be executed entirely on the user's computer, partially on the user's computer, as a standalone software package, partially on the user's computer and partially on a remote computer, or entirely on a remote computer or server. In cases involving remote computers, the remote computer can be connected to the user's computer via any type of network—including a Local Area Network (LAN) or a Wide Area Network (WAN)—or can be connected to an external computer (e.g., via the Internet using an Internet service provider).

[0143] The flowcharts and block diagrams in the accompanying drawings illustrate the architecture, functionality, and operation of possible implementations of systems, methods, and computer program products according to various embodiments of this application. In this regard, each block in a flowchart or block diagram may represent a module, segment, or portion of code containing one or more executable instructions for implementing a specified logical function. It should also be noted that in some alternative implementations, the functions indicated in the blocks may occur in a different order than those indicated in the drawings. For example, two consecutively indicated blocks may actually be executed substantially in parallel, and they may sometimes be executed in reverse order, depending on the functions involved. It should also be noted that each block in the block diagrams and / or flowcharts, and combinations of blocks in the block diagrams and / or flowcharts, can be implemented using a dedicated hardware-based system that performs the specified function or operation, or using a combination of dedicated hardware and computer instructions.

[0144] The modules described in the embodiments of this application can be implemented in software or hardware. The names of the modules do not necessarily limit the functionality of the unit itself.

[0145] The readable storage medium provided in this application is a computer-readable storage medium that stores computer-readable program instructions (i.e., computer programs) for executing the above-described behavior pattern prediction method. This addresses the current challenges in geothermal resource evaluation and exploration, which primarily rely on physical models established based on formation heat and mass transfer mechanisms for gridded numerical simulations. While the accuracy of these simulations depends on physical parameters, the geothermal well production process is influenced by complex factors. Furthermore, meeting geothermal well production boundaries and achieving precise grid division requires substantial computational costs, hindering the widespread adoption of geothermal production simulation. Compared to existing technologies, the beneficial effects of the computer-readable storage medium provided in this application are the same as those of the behavior pattern prediction method provided in the above embodiments, and will not be elaborated upon here.

[0146] This application also provides a computer program product, including a computer program that, when executed by a processor, implements the steps of the behavior pattern prediction method described above.

[0147] The computer program product provided in this application can solve the current problem that geothermal resource evaluation and exploration mainly rely on establishing physical models based on the heat and mass transfer modes of the formation and performing gridded numerical simulations. The accuracy of the simulation requires the support of physical parameters, but the influencing factors of geothermal well production processes are complex, and meeting the production boundaries of geothermal wells and precise grid division requires enormous computational costs, making it difficult to widely promote and popularize geothermal production simulation. Compared with the prior art, the beneficial effects of the computer program product provided in this application are the same as those of the behavior pattern prediction method provided in the above embodiments, and will not be elaborated upon here.

[0148] The above description is only a part of the embodiments of this application and does not limit the patent scope of this application. All equivalent structural transformations made under the technical concept of this application and using the contents of the specification and drawings of this application, or direct / indirect applications in other related technical fields, are included in the patent protection scope of this application.

Claims

1. A method for predicting behavioral patterns, characterized in that, The behavior pattern prediction method includes: Acquire historical behavior data of geothermal wells and normalize the historical behavior data to obtain target input data; The target input data is imported into a preset deep learning neural network model to obtain prediction data; The behavior pattern of the geothermal well is predicted based on the predicted data.

2. The behavior pattern prediction method as described in claim 1, characterized in that, Before the step of inputting the target input data into a preset deep learning neural network model to obtain the predicted data, the method further includes: An initial deep learning neural network model was constructed based on the feedforward neural network algorithm; The initial deep learning neural network model is trained using preset training data to obtain training output values; The parameters of the initial deep learning neural network model are updated based on the training output values ​​to obtain the target deep learning neural network model.

3. The behavior pattern prediction method as described in claim 2, characterized in that, The steps for constructing the initial deep learning neural network model based on the feedforward neural network algorithm include: A multilayer perceptron is configured based on a feedforward neural network algorithm, wherein the multilayer perceptron includes an input layer, a hidden layer, and an output layer; Based on the target input data, a target loss function and an optimizer are selected, and an initial deep learning neural network model is established using the multilayer perceptron, the loss function, and the optimizer.

4. The behavior pattern prediction method as described in claim 3, characterized in that, The steps for setting up a multilayer perceptron based on a feedforward neural network algorithm include: The dimension of the input layer of the multilayer perceptron is set based on the number of features in the target input data. Select an appropriate activation function to design the hidden layer, and configure the hidden layer with a preset number of neurons; The output layer is defined using a linear activation function based on the requirements of the predicted task. A multilayer perceptron is generated by connecting the input layer, the hidden layer, and the output layer using a feedforward neural network algorithm.

5. The behavior pattern prediction method as described in claim 2, characterized in that, The step of updating the parameters of the initial deep learning neural network model based on the training output value to obtain the target deep learning neural network model includes: Calculate the first loss function based on the training output value; The gradient of the first loss function with respect to the model parameters is determined based on the backpropagation algorithm; The training data is selected in mini-batch using the mini-batch gradient descent method, and the parameters of the initial deep learning neural network model are adjusted according to the direction of the gradient to calculate the second loss function. When the value of the second loss function drops to a preset loss threshold, the target deep learning neural network model is obtained based on the adjusted parameters.

6. The behavior pattern prediction method as described in claim 5, characterized in that, Before the step of calculating the first loss function based on the training output value, the method further includes: The initial loss function is regularized by adding a squared penalty term to the weights, resulting in the first loss function.

7. A behavior pattern prediction device, characterized in that, The device includes: The data processing module is used to acquire historical behavior data of geothermal wells and preprocess the historical behavior data to obtain target input data; The data prediction module is used to import the target input data into a preset deep learning neural network model to obtain predicted data; The behavior prediction module is used to predict the behavior pattern of the geothermal well based on the prediction data.

8. A behavior pattern prediction device, characterized in that, The device includes: a memory, a processor, and a computer program stored in the memory and executable on the processor, the computer program being configured to implement the steps of the behavior pattern prediction method as described in any one of claims 1 to 6.

9. A storage medium, characterized in that, The storage medium is a computer-readable storage medium, and a computer program is stored on the storage medium. When the computer program is executed by a processor, it implements the steps of the behavior pattern prediction method as described in any one of claims 1 to 6.

10. A computer program product, characterized in that, The computer program product includes a computer program that, when executed by a processor, implements the steps of the behavior pattern prediction method as described in any one of claims 1 to 6.

Images