Temperature model liquid production prediction interpretation model based on artificial intelligence algorithm
By optimizing the ELM neural network through an adaptive variable-temperature simulated annealing algorithm, improving the initial temperature and cooling strategy, and combining data augmentation technology, the problems of difficult parameter acquisition and low efficiency in traditional oil production prediction are solved, achieving higher accuracy and faster oil production prediction.
Patent Information
- Application Number
- CN202511016302.2
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2025-07-23
- Publication Date
- 2025-11-18
AI Technical Summary
Traditional oil production prediction methods suffer from difficulties in obtaining parameters and low prediction efficiency. The BP algorithm has a long training time, and the random initialization of input weights and biases in the ELM neural network leads to poor prediction stability. The simulated annealing algorithm has blind initial temperature settings and a single cooling strategy, resulting in slow convergence and local optima.
An adaptive variable-temperature simulated annealing (ICSA) algorithm combined with an extreme learning machine (ELM) neural network was adopted. By optimizing the input weights and bias terms, the initial temperature and cooling function of the simulated annealing algorithm were improved. Combined with Pearson correlation analysis and CTGAN data augmentation, an ICSA-ELM model was constructed.
It improves the accuracy and efficiency of oil production prediction, shortens training time, overcomes the shortcomings of random initialization, and achieves faster convergence and higher prediction accuracy.
Smart Images
Figure CN120975295A_ABST
Abstract
Description
TECHNICAL FIELD
[0001] The present application relates to the technical field of oil production prediction, and particularly relates to a temperature model liquid production prediction interpretation model based on an artificial intelligence algorithm. BACKGROUND
[0002] Traditional oil production prediction methods have the problems of difficult parameter acquisition and low prediction efficiency. The traditional BP algorithm has the problems of long training time and low precision. The input weight and bias of the ELM neural network are randomly initialized, which easily leads to poor prediction stability. The initial temperature setting is blind in the optimization of the simulated annealing algorithm, and the slow convergence and local optimum problems are caused by the single cooling strategy.
[0003] Therefore, it is necessary to provide a temperature model liquid production prediction interpretation model based on an artificial intelligence algorithm to solve the above technical problems. SUMMARY
[0004] The present application provides a temperature model liquid production prediction interpretation model based on an artificial intelligence algorithm, which aims to solve the technical problems of difficult parameter acquisition and low prediction efficiency of traditional oil production prediction methods.
[0005] The present application is implemented in the following way: a temperature model liquid production prediction interpretation model based on an artificial intelligence algorithm, comprising the following steps: Step 1, establishing an ELM neural network; Step 2, initializing the input weight and bias of the ELM neural network, and defining the parameters of the simulated annealing algorithm, including the initial temperature, the temperature reduction function, and the iteration number; Step 3, inputting the training set data, calculating the predicted value of the training set data using the ELM neural network, and calculating the mean square error (MSE) between the predicted value and the measured value; Step 4, taking the MSE as the objective function of the ICSA algorithm, and taking the weight and threshold value in the ELM neural network as the independent variable of the objective function; Step 5, entering the iteration process of the simulated annealing algorithm, generating new weight and threshold value each time, and calculating the new MSE; if the new MSE is smaller than the current error, the new weight and threshold value are accepted; otherwise, the new weight and threshold value are accepted with a probability P = exp (-Δ / T); Step 6, repeating step 4 until the predetermined iteration number is reached or the temperature is reduced to the set value; Step 7, substituting the optimal weight and threshold value obtained by the simulated annealing algorithm into the ELM to calculate the predicted value of the test data, and calculating the error between the predicted value and the expected value of the test data.
[0006] Preferably, the training data set for oil production prediction is represented as xi , P i , x i = [ x i1 , x i2 , x iN ] T ∈ R N contains the key factors affecting oil production, P i ∈ R is the output of the corresponding training data, then the mathematical expression of the ELM model is written as: (1) In formula (1), g(·) is the activation function of the hidden layer neuron; ω i is the connection weight between the input layer and the hidden layer neuron; b i is the bias of the hidden layer neuron, and the output result of the sample in the model training stage is expressed in matrix form as: (2) In formula (2), H is the output of the hidden layer node, T is the expected output of the output layer; When the training stage of the model is completed, the performance of the model on the training data should be as close to the true situation as possible, so the model only needs to satisfy formula (3): (3) Therefore, the process of solving the matrix β is converted into the process of solving equation formula (4): (4) In formula (4), is the generalized inverse matrix of H .
[0007] Preferably, in the ELM, an adaptive variable temperature simulated annealing algorithm (ICSA algorithm) is used to optimize and adjust the input layer weight ω and the bias b to improve the prediction accuracy of the ELM, that is, by searching the optimal ω and b within a certain range to improve the performance of the model.
[0008] Preferably, the simulated annealing algorithm includes initial temperature improvement and cooling function improvement, wherein the initial temperature improvement is: adjusting the temperatureT The following improvements are made: (5) (6) In formulas (5) and (6), T 0 represents the initial value of the temperature; ρ represents the complexity of the function; f , g are the maximum and minimum values of the function, respectively; m , n are the number of peaks and troughs of the function, respectively.
[0009] Preferably, the cooling function of the simulated annealing algorithm is used to control the rate of temperature drop in the search process, to balance the ability of global exploration and local development; the cooling function improvement is to propose a segmented cooling strategy, as shown in formula (8), which uses rapid cooling method at high temperature stage to shorten the solving process time; and at low temperature stage, slow cooling method is adopted to ensure that more areas can be searched to find the optimal solution more easily; (8) In formula (8), T 0 is the current temperature; α is the temperature decay rate; k is the number of iterations; T 0 > 10 represents the high-temperature annealing stage, and through this form, the effect of rapid cooling can be achieved; T 0 ≤ 10 represents the low-temperature annealing stage, and in this stage, the algorithm will perform local search; By optimizing the initial temperature and cooling function of the simulated annealing algorithm, an adaptive variable temperature simulated annealing algorithm (IC Simulated Annealing, ICSA) that converges faster and is easier to find the global optimal solution is finally obtained.
[0010] Preferably, after the model is constructed, the ICSA-ELM model needs to be verified to study the performance of the ICSA-ELM algorithm in oil well oil and water production prediction. By introducing ELM, SA-ELM, LSTM, LSTM+Attention, and MKLRPMC as comparative models, all models are respectively experimented on the basis of the above data to ensure that the model has high precision; the verification of the ICSA-ELM model includes the following steps: P1, pre-process the oil well oil production data, select appropriate feature parameters, and use CTGAN to optimize the processed data, and process and divide the data set according to the ratio of 7:3; P2, normalize the data and build an ELM model; P3, modify the initial temperature and cooling function of simulated annealing according to the method above to form the ICSA algorithm, and use the ICSA algorithm to optimize the weights and biases of the ELM model, and define the target function MSE as the optimization target; P4, in each iteration, monitor the loss and accuracy during training until a satisfactory model performance is obtained; P5, use the test set to evaluate the performance of the trained ELM model, and calculate the error between the predicted oil production and the actual oil production; P6, train other models in the same way with the same data to intuitively understand the performance of the model.
[0011] Preferably, the data preprocessing of step P1 includes the following steps: Pearson correlation analysis method is used to measure the linear similarity in oil field data, and the Pearson correlation coefficient calculation formula is: (3-14) The absolute value of the Pearson correlation coefficient is used to measure the correlation strength: when the value is closer to 1, it indicates that the linear correlation between variables is strong; on the contrary, when the value is closer to 0, it indicates that the linear correlation between variables is weak; by analyzing the selected input variables and their correlation, it is found that the feature parameters with high correlation with oil production include water production, dynamic liquid level, pump efficiency, temperature difference, water cut, etc.
[0012] Compared with the related art, the temperature model liquid production prediction interpretation model based on artificial intelligence algorithm provided by the present application has the following beneficial effects: 1. Precision improvement: ICSA algorithm optimizes the weights and biases of ELM, overcomes the defect of random initialization, and significantly improves the coincidence rate of prediction curve and measured value; 2. Efficiency optimization: ELM single forward propagation training mechanism combined with ICSA segmented cooling strategy (high temperature rapid cooling, low temperature fine search) shortens the model training time. BRIEF DESCRIPTION OF DRAWINGS
[0013] Figure 1 The CSA-ELM algorithm flowchart in the present application; Figure 2 The experimental flowchart for verifying the ICSA-ELM model in the present application; Figure 3A production rate prediction diagram for a production rate prediction software window. DETAILED DESCRIPTION
[0014] The present application will be further described below in conjunction with the accompanying drawings and embodiments.
[0015] The present application aims at the problems of traditional production rate prediction methods, such as difficulty in obtaining parameters, low prediction efficiency and long training time of artificial intelligence algorithms, and proposes a production logging output profile interpretation model combining a self-adaptive variable temperature simulated annealing algorithm and an ELM neural network.
[0016] 1. ELM neural network The extreme learning machine (ELM) is an ELM neural network that sets the parameters of the network hidden layer nodes as the target of risk minimization, and then determines the weights of the output layer through the least squares method. Compared with traditional methods, its learning process only needs one forward propagation, without iterative updating of weights, avoiding the iterative back propagation and weight updating process, and the training speed of ELM is usually much faster than that of the traditional BP algorithm, while having better prediction accuracy. In ELM, the weights and biases between the input layer and the hidden layer are randomly initialized and remain unchanged during the training process. The activation function of the hidden layer can be linear or nonlinear, such as Sigmoid, ReLU, etc. The output of the hidden layer is transmitted as input to the output layer, and the weights of the output layer are calculated by the least squares method.
[0017] The training data set of the production rate prediction is represented as x i , P i}, x i = [ x i1 , x i2 , …, x iN ] T ∈ R N contains the key factors affecting the production rate, P i and the output of the corresponding training data, then the mathematical expression of the ELM model can be written as: (1) In equation (1), g(·) is the activation function of the hidden layer neurons; ω i is the connection weight between the input layer and the hidden layer neurons; b iBias of hidden layer neurons. In the model, the weights and biases are usually set to initial values by random initialization before training, while the parameters β are unknown. Therefore, solving β parameters is crucial in the training process of the model. In the model training phase, the output results of the samples can be represented in matrix form as: (2) In equation (2), H is the output of the hidden layer node, T is the expected output of the output layer.
[0018] Generally, when the present application completes the training phase of the model, it is expected that the performance of the model on the training data will be as close to the real situation as possible, so the model only needs to satisfy equation 3: (3) Therefore, the process of solving the matrix β is converted into the process of solving equation (4).
[0019] (4) In equation (4), is the generalized inverse matrix of H .
[0020] In ELM, although the input layer weights ω and the bias b are randomly initialized, once determined, they will affect the output of the hidden layer H , and in turn affect the output weights β obtained by analytical method. If ω and b are not properly selected, it may lead to poor prediction performance of the model, and the selection of the number of hidden layer neurons and the activation function may affect the performance of ELM. In order to improve the prediction accuracy of ELM, some optimization algorithms can be used to adjust ω and b instead of choosing them completely randomly. ICSA algorithm is a global optimization technique that can improve the performance of the model by searching for the optimal ω and b within a certain range.
[0021] 2. Adaptive temperature simulation annealing algorithm In solving the maximum satisfaction problem, the traditional simulated annealing algorithm faces some challenges and limitations. First, the algorithm is usually inefficient in handling such problems, which means it takes a long time to find a solution. Second, the traditional simulated annealing algorithm often lacks directionality in its search, which can lead to blind searches that may result in the algorithm wandering in the solution space without quickly approaching the optimal solution or high-quality solution. Finally, the algorithm converges slowly, which further prolongs the solution time and may affect the quality of the solution. To overcome these shortcomings, two main strategies are proposed here to improve the traditional simulated annealing algorithm to optimize the performance of the algorithm, enabling it to solve the optimal result faster and more efficiently.
[0022] (1) Initial temperature improvement In the simulated annealing algorithm, temperature T is a crucial parameter that directly affects the search process and final result of the algorithm. T The initial value of T is positively correlated with the probability of obtaining the optimal solution, while the complexity of the function itself, such as its multi-peak nature, nonlinearity, and the size of the search space, also has a significant impact on the initial value of T . Therefore, to make the algorithm have a higher probability of obtaining the optimal solution, the improvement of is as follows: (5) (6) In formulas (5) and (6), T 0 represents the initial value of the temperature; ρ represents the complexity of the function; f , g are the maximum and minimum values of the function, respectively; m , n are the number of peaks and valleys of the function, respectively.
[0023] (2) Cooling function improvement The cooling function of the simulated annealing algorithm is used to control the rate of temperature decrease during the search process, balancing the ability of global exploration and local development. In the classic simulated annealing algorithm, the setting of the initial temperature plays a crucial role in the performance of the algorithm. Generally speaking, the initial temperature should be set high enough to ensure that the system can explore different regions of the solution space, thereby increasing the chances of finding the global optimal solution. However, if the initial temperature is too high, the algorithm may spend too much time on random searches in the high-temperature region, resulting in low computational efficiency. But a too low final temperature may cause the algorithm to fall into a local optimal solution too early, unable to continue exploring other potential solution space regions. The cooling function of the traditional simulated annealing algorithm: (7) In formula (7), T 0 T is the current temperature, α T is the temperature decay rate during annealing process.
[0024] The traditional simulated annealing method needs a large number of samplings at high temperature stage to achieve thermal equilibrium, which may lead to excessive calculation; at low temperature, the algorithm may fall into local minimum and cannot jump out, because as the temperature decreases, the probability of accepting poor solutions also decreases, which may lead the algorithm to be trapped in local optimal solution when approaching the global optimal solution. Therefore, a segmented cooling strategy is proposed here. As formula (8) shows, this strategy adopts rapid cooling method at high temperature stage to shorten the solving process time; and at low temperature stage, it takes slow cooling way to ensure that more areas can be searched and the optimal solution can be found more easily. This way makes the algorithm converge faster and reduces the time needed for solving.
[0025] (8) In formula (8), T 0 T is the current temperature; α T is the temperature decay rate; k N is the iteration number.
[0026] T 0 > 10 represents high temperature annealing stage, and through this form, rapid cooling effect can be achieved, T 0 ≤ 10 represents low temperature annealing stage, and in this stage, the algorithm will perform local search. The segmented cooling method can combine the advantages of global search and local search, and can gradually narrow the search range while exploring the solution space, improving the search efficiency.
[0027] By optimizing the initial temperature and cooling function of the simulated annealing algorithm, an adaptive variable temperature simulated annealing algorithm (IC Simulated Annealing, ICSA) that converges faster and is easier to find the global optimal solution is finally obtained.
[0028] 3. ICSA-ELM model construction In view of the problem that ELM neural network has strong randomness in input layer weight and bias, which easily leads to poor prediction effect, the improved adaptive variable temperature simulated annealing algorithm is used to optimize it, and finally the ICSA-ELM yield prediction model is constructed, and its specific process is as shown in Figure 1 The process is as follows: (1) Initialize the input weights and bias terms of the ELM neural network. Define the parameters of the simulated annealing algorithm, including the initial temperature, cooling function, number of iterations, etc.
[0029] (2) Input the training set data, calculate the predicted values of the training set data using the ELM neural network, and calculate the mean square error (MSE) between the predicted values and the measured values.
[0030] (3) Take MSE as the objective function of ICSA algorithm, and take the weights and thresholds in ELM neural network as the independent variables of the objective function.
[0031] (4) Enter the iteration process of the simulated annealing algorithm. Each iteration generates new weights and thresholds, and calculates the new MSE. If the new MSE is smaller than the current error, accept the new weights and thresholds; otherwise, accept the new weights and thresholds with a probability P = exp ( -Δ / T ) of exp((current error - new error) / current error).
[0032] (5) Repeat step 4 until the predetermined number of iterations is reached or the temperature drops to a certain level.
[0033] (6) Substitute the optimal weights and thresholds obtained by the simulated annealing algorithm into the ELM, calculate the predicted values of the test data, and calculate the error between the predicted values and the expected values of the test data.
[0034] 4. ICSA-ELM model verification and analysis In this experiment, historical data from January 2018 to April 2019 in a certain oilfield area was selected as the data sample. The sample data was divided into training sample set and test sample set according to the ratio of 7:3.
[0035] In order to study the performance of ICSA-ELM algorithm in oil well oil production and water production prediction, ELM, SA-ELM, LSTM, LSTM+Attention, MKLRP MC as a comparison model, all models are based on the above data respectively to ensure that the model has high precision.
[0036] The flow chart of the experiment design is shown in Figure 2 .
[0037] (1) Preprocess the oil well oil production data, select appropriate feature parameters, and use CTGAN to optimize the processed data, and process and divide the data set according to the ratio of 7:3.
[0038] (2) Normalize the data and build an ELM model.
[0039] (3) The initial temperature of simulated annealing and the temperature reduction function are modified according to the method above to form the ICSA algorithm, and the ICSA algorithm is used to optimize the weights and biases of the ELM model, and the target function MSE is defined as the optimization target.
[0040] (4) In each iteration, the loss and accuracy in the training process are monitored until a satisfactory model performance is obtained.
[0041] (5) The performance of the trained ELM model is evaluated using the test set. The error between the predicted oil production and the actual oil production is calculated.
[0042] (6) Other models are trained in the same way with the same data to intuitively understand the performance of the model.
[0043] Wherein, when processing oilfield data, the input feature variables have diversity and complexity. Overly complex of these variables not only leads to the expansion of the size of the neural network and the increase of the number of parameters, but also significantly slows down the convergence speed of the network. In order to solve this problem, the Pearson correlation analysis method is adopted to measure the linear similarity in the oilfield data. The formula for calculating the Pearson correlation coefficient is: (3-14) The absolute value of the Pearson correlation coefficient is a key indicator to measure the strength of correlation: when its value is closer to 1, it indicates that the linear correlation between variables is strong; on the contrary, when its value is closer to 0, it means that the linear correlation between variables is weak. Through the analysis of the correlation between the selected input variables and the correlation, it is found that the characteristic parameters with high correlation with oil production mainly include water production, dynamic liquid level, pump efficiency, temperature difference, water cut, etc.
[0044] After using the Pearson correlation coefficient to select parameters with high correlation, the data needs to be enhanced. The CTGAN model is a GAN-based method, mainly used for modeling and sampling from the distribution of table data. After determining the relevant parameters through the above operation, the CTGAN model is used to generate relevant oil production data, so as to supplement the insufficient sample data, and Table 1 is an example of oil production prediction data after expansion using CTGAN.
[0045] Table 1 Example data expanded by CTGAN
[0046] The oil production prediction interface is as shown in Figure 3 It can be seen from Figure 3 that the curves of the predicted value and the measured value are basically consistent, indicating that the method has good coincidence rate for calculating liquid production.
[0047] The above examples are only used to illustrate the technical solutions of the present application, and are not intended to limit the protection scope of the present application. Obviously, the described examples are only some of the embodiments of the present application, but not all the embodiments. Based on these examples, all other embodiments obtained by those of ordinary skill in the art without creative work fall within the scope of the present application. Although the present application is described in detail with reference to the above examples, those of ordinary skill in the art can still combine, add or delete the features in the embodiments of the present application according to the circumstances without creative work, so as to obtain different other technical solutions which do not deviate from the concept of the present application in essence, and these technical solutions also fall within the scope of the present application.
Claims
1. A temperature-based liquid production prediction and interpretation model based on artificial intelligence algorithms, characterized in that, Includes the following steps: Step 1: Establish an ELM neural network; Step 2: Initialize the input weights and bias terms of the ELM neural network, and define the parameters of the simulated annealing algorithm, including the initial temperature, cooling function, and number of iterations; Step 3: Input the training set data, use the ELM neural network to calculate the predicted values of the training set data, and calculate the mean square error (MSE) between the predicted values and the measured values. Step 4: Use MSE as the objective function of the ICSA algorithm, and use the weights and thresholds in the ELM neural network as the independent variables of the objective function; Step 5: Enter the iterative process of the simulated annealing algorithm. In each iteration, new weights and thresholds are generated, and a new MSE is calculated. If the new MSE is smaller than the current error, the new weights and thresholds are accepted; otherwise, they are accepted probabilistically. P = exp (-Δ / T) accepts new weights and thresholds; Step 6: Repeat step 4 until the predetermined number of iterations is reached or the temperature drops to the set value; Step 7: Substitute the optimal weights and thresholds obtained from the simulated annealing algorithm into ELM to calculate the predicted values of the test data, and calculate the error between the predicted values and the expected values of the test data.
2. The temperature-based liquid production prediction and interpretation model based on artificial intelligence algorithms according to claim 1, characterized in that, The training dataset for oil production prediction is represented as { x i , P i }, x i =[ x i1 , x i2 , ..., x iN ] T ∈ R N It includes key factors that affect oil production. P i If ∈R is the output of the corresponding training data, then the mathematical expression of the ELM model can be written as: （1） In equation (1), g(·) is the activation function for neurons in the hidden layer; ω i These are the connection weights between neurons in the input layer and the hidden layer. b i As the bias of the hidden layer neurons, during the model training phase, the output of the samples is represented in matrix form as follows: （2） In equation (2), H The output of the hidden layer nodes, T The expected output of the output layer; After the training phase of the model is completed, it is expected that the model's performance on the training data can be as close as possible to the real situation, so the model only needs to satisfy equation (3): （3） Therefore, the solution matrix is obtained. β The process is then transformed into solving equation formula (4): （4） In equation (4), for H The generalized inverse matrix.
3. The temperature-based liquid production prediction and interpretation model based on artificial intelligence algorithms according to claim 1, characterized in that, In ELM, the Adaptive Variable Temperature Simulated Annealing (ICSA) algorithm is used to optimize and adjust the input layer weights. ω and bias b To improve the prediction accuracy of ELM, this is achieved by searching for the optimal [element / method] within a certain range. ω and b To improve the performance of the model.
4. The temperature-based liquid production prediction and interpretation model based on artificial intelligence algorithms according to claim 3, characterized in that, The simulated annealing algorithm includes improvements to the initial temperature and the cooling function. The initial temperature improvement involves adjusting the temperature... T The following improvements will be made: （5） （6） In formulas (5) and (6), T 0 Represents the initial temperature value; ρ This represents the complexity of the function; f , g These are the maximum and minimum values of the function, respectively. m , n These represent the number of peaks and troughs of the function, respectively.
5. The temperature-based liquid production prediction and interpretation model based on artificial intelligence algorithms according to claim 4, characterized in that, The cooling function of the simulated annealing algorithm is used to control the rate of temperature decrease during the search process in order to balance the ability of global exploration and local development. The improvement of the cooling function is to propose a segmented cooling strategy, as shown in formula (8). This strategy adopts a rapid cooling method in the stage of high temperature to shorten the solution process time. In the lower temperature phase, a slow cooling method is adopted to ensure that more areas can be searched, making it easier to find the optimal solution; （8） In equation (8), T 0 The current temperature; α This refers to the temperature decay rate; k This represents the number of iterations. T 0 > 10 indicates the high-temperature annealing stage, which can achieve a rapid cooling effect; T 0 ≤10 indicates the low-temperature annealing stage, during which the algorithm performs a local search; By optimizing the initial temperature and cooling function of the simulated annealing algorithm, an adaptive variable temperature simulated annealing (ICSA) algorithm with faster convergence and easier global optimization is obtained.
6. The temperature-based liquid production prediction and interpretation model based on artificial intelligence algorithms according to claim 1, characterized in that, After the model is built, it is also necessary to validate the ICSA-ELM model to study its performance in predicting oil and water production in oil wells. ELM, SA-ELM, LSTM, LSTM+Attention, and MKLRPMC are introduced as comparison models. All models are tested on the aforementioned data to ensure the model has high accuracy. The validation of the ICSA-ELM model includes the following steps: P1. Preprocess the oil well production data, select appropriate feature parameters, and use CTGAN to optimize the processed data. Process and divide the dataset according to a 7:3 ratio. P2. Normalize the data and build an ELM model; P3. Modify the initial temperature and cooling function of simulated annealing according to the method above to form the ICSA algorithm, and use the ICSA algorithm to optimize the weights and biases of the ELM model, defining the objective function MSE as the optimization target. P4. In each iteration, monitor the loss and accuracy during training until satisfactory model performance is obtained. P5. Use the test set to evaluate the performance of the trained ELM model and calculate the error between the predicted oil production and the actual oil production. P6. Train other models using the same data and in the same way to intuitively understand the model's performance.
7. The temperature-based liquid production prediction and interpretation model based on artificial intelligence algorithms according to claim 1, characterized in that, The data preprocessing in step P1 includes the following steps: Pearson correlation analysis was used to measure the linear similarity in oilfield data. The formula for calculating the Pearson correlation coefficient is as follows: （3-14） The absolute value of the Pearson correlation coefficient is used to measure the strength of the correlation: the closer the value is to 1, the stronger the linear correlation between the variables; conversely, the closer the value is to 0, the weaker the linear correlation between the variables. By analyzing the selected input variables and their correlations, it was found that the characteristic parameters with a high correlation to oil production include water production, dynamic liquid level, pump efficiency, temperature difference, and water cut.