Shale gas horizontal well construction parameter optimization method based on artificial neural network
By optimizing the construction parameters based on artificial neural networks, the problem of difficulty predicting the difficulty of the middle production of shale gas level wells is solved, and the optimization of the production capacity and construction parameters of shale gas levels and construction parameters is achieved.
Patent Information
- Application Number
- CN202310931014.4
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2023-07-27
- Publication Date
- 2025-01-28
AI Technical Summary
When designing shale gas horizontal wells, due to their complex geological conditions and dynamic production characteristics, existing empirical formula design methods are difficult to accurately predict production capacity, which makes it difficult to design construction parameters.
Adopt construction parameters based on artificial neural networks, and establish an artificial neural network model by collecting and returning to the geological and construction factors and test yields of the illegal level of shale gas level wells. Under the guidance, determine the optimal construction parameters.
This method can more accurately predict the capacity of shale gas level wells, reduce the difficulty of parameter design, improve the capacity of the shale gas level well, and provide scientific guidance for construction.
Smart Images
Figure CN119378345A_ABST
Abstract
Description
Technical Field
[0001] The invention relates to the technical field of shale gas horizontal well design, and in particular to a shale gas horizontal well construction parameter optimization method based on artificial neural network. Background Art
[0002] The design of construction parameters of shale gas horizontal wells affects the production capacity of shale gas horizontal wells. When designing shale gas horizontal wells, the corresponding construction parameters are generally designed according to empirical formulas combined with the geological parameters of the horizontal wells, in order to maximize the production capacity of the horizontal wells. However, due to the complex flow mechanism of shale gas, the difficulty in predicting production dynamics, and the many factors affecting high-yield wells, the design method based on empirical formulas has poor accuracy. Therefore, if a set of construction factor parameter optimization methods suitable for shale gas horizontal wells can be established, it will have positive practical significance for improving the production capacity of shale gas horizontal wells and guiding construction. Summary of the invention
[0003] The technical problem to be solved by the present invention is to provide a shale gas horizontal well construction parameter optimization method based on artificial neural network, so as to solve the problem of difficulty in designing horizontal well parameters to improve horizontal well production capacity.
[0004] The technical solution adopted by the present invention to solve the technical problem is: a method for optimizing the construction parameters of shale gas horizontal wells based on artificial neural network, which is characterized by comprising:
[0005] Collect and count the parameters of the productivity influencing factors affecting the productivity of shale gas horizontal wells and the test production of the horizontal wells, normalize the parameters of the productivity influencing factors and the test production, and divide the training set and the test set in proportion. The productivity influencing factors include geological factors and construction factors.
[0006] Taking the parameters of the productivity influencing factors in the training set as input variables and the horizontal well test production as output variables, an artificial neural network model is established according to the mapping relationship between the parameters of the productivity influencing factors and the horizontal well test production;
[0007] Using a particle algorithm to optimize the parameters of the artificial neural network model to obtain an optimal neural network model, wherein the parameters include the number of hidden layers and the number of neurons;
[0008] The optimal neural network model is verified by using the test set, and the capacity influencing factor parameters of the test set are used as input to obtain the predicted output value, and the error between the predicted output value and the test output data in the test set is calculated. If the error meets the requirements, the optimal neural network model meets the requirements;
[0009] Under the geological factor parameter conditions of the horizontal well to be constructed, a genetic algorithm is used to calculate the fitness function of the optimal neural network model, so as to obtain the optimal values of the construction factor parameters of the horizontal well to be constructed under the geological parameter conditions.
[0010] Furthermore, the correlation coefficient R between the production capacity influencing factors is less than 0.5.
[0011] Furthermore, the geological factors include one or more of vertical depth, pressure coefficient, porosity, total organic carbon content, total gas content, Young's modulus, Poisson's ratio and brittleness index.
[0012] Furthermore, the construction factors include one or more of the number of clusters, displacement, flowback rate, liquid intensity, sand addition intensity, slippery water ratio and quartz sand ratio.
[0013] Furthermore, the artificial neural network model is a BP neural network model.
[0014] Furthermore, the predicted yield value is compared with the test yield data in the test set for error calculation, and the error calculation adopts the determination coefficient R 2 , root mean square error RMSE, regression line slope k, and Weir model consistency index IA.
[0015] Furthermore, the method also includes a step of determining the relative importance of the capacity influencing factors of the optimal neural network model, so as to give priority to satisfying the construction factor parameters with large relative importance during construction.
[0016] Furthermore, the steps of determining the relative importance of the production capacity influencing factors of the optimal neural network model include:
[0017] Establish a partial dependency graph corresponding to each capacity influencing factor;
[0018] According to the slope of the curve in the partial dependence diagram, the relative importance of all capacity influencing factors is ranked from large to small, so that the relative importance of capacity influencing factors in the optimal artificial neural network model can be obtained.
[0019] Furthermore, the normalization process adopts a min-max standardization method.
[0020] Furthermore, the division ratio of the training set and the test set is 9:1.
[0021] The beneficial effects of the present invention are as follows: the shale gas horizontal well construction parameter optimization method based on artificial neural network of the present invention proposes a method of firstly using an artificial network model to perform nonlinear relationship modeling according to the influencing variables of the unit section length test production of the shale gas horizontal well, and then using a genetic algorithm to determine the optimal construction parameters according to the geological factors of the unit section length test production of the shale gas horizontal well. The accuracy of the method is not limited by idealized assumptions and parameter settings, and the nonlinear relationship between input and output variables can be automatically learned and solved by only relying on the model input variables, thereby well solving the problem of difficulty in horizontal well parameter design caused by the complex geological conditions of the shale reservoir, and having positive practical significance for improving the productivity of shale gas horizontal wells and guiding construction. BRIEF DESCRIPTION OF THE DRAWINGS
[0022] Figure 1 It is a schematic flow diagram of the optimization method of the present invention;
[0023] Figure 2 It is a flow chart for optimizing construction factor parameters using genetic algorithm;
[0024] Figure 3 It is the distribution histogram of geological engineering parameters of 317 shale gas horizontal wells;
[0025] Figure 4 It is the distribution histogram of the test production of 317 shale gas horizontal wells;
[0026] Figure 5 is the heat map of correlation coefficients between capacity variable parameters;
[0027] Figure 6 It is a schematic diagram of the optimal neural network model structure;
[0028] Figure 7 is the partial dependence plot of the capacity variable parameters;
[0029] Figure 8 It is a relative importance ranking diagram of capacity variable parameters. DETAILED DESCRIPTION
[0030] The present invention is further described below in conjunction with the accompanying drawings and embodiments.
[0031] like Figure 1 , Figure 2 As shown, a method for optimizing construction factor parameters of shale gas horizontal wells based on an artificial neural network (ANN) model of the present invention comprises the following steps:
[0032] Step 1: Collect and count the parameters of the production capacity influencing factors affecting the production capacity of shale gas horizontal wells and the test production of the horizontal wells, normalize the parameters of the production capacity influencing factors and the test production, and divide the training set and the test set in proportion. The production capacity influencing factors include geological factors and construction factors.
[0033] Step 2: using the parameters of the production capacity influencing factors in the training set as input variables and the horizontal well test production as output variables, and establishing an artificial neural network model according to the mapping relationship between the parameters of the production capacity influencing factors and the horizontal well test production;
[0034] Step 3: Using a particle swarm optimization algorithm (PSO) to optimize the parameters of the artificial neural network model to obtain an optimal neural network model, wherein the parameters include the number of hidden layers and the number of neurons;
[0035] Step 4: Use the test set to verify the optimal neural network model, take the capacity influencing factor parameters of the test set as input, obtain the predicted output value, and calculate the error between the predicted output value and the test output data in the test set. If the error meets the requirements, the optimal neural network model meets the requirements;
[0036] Step 5: According to the geological factor parameters of the horizontal well to be constructed, the range of the construction factor parameters of the horizontal well to be constructed can be determined based on experience or historical data, so that a plurality of construction factor parameter populations can be formed. Under the geological factor parameter conditions of the horizontal well to be constructed, the geological factor parameters of the horizontal well to be constructed are used as the input of the optimal neural network model, and then the genetic algorithm is used to optimize the above optimal neural network model to obtain the maximum unit section length test production purpose, thereby obtaining the optimal value of the construction factor parameters of the horizontal well to be constructed under the geological factor parameter conditions of the horizontal well to be constructed (see the specific optimization process for details). Figure 2 ).
[0037] The optimal value of the construction factor parameter is used to guide the construction of the horizontal well to be constructed, so as to improve the production of the horizontal well.
[0038] The design method of the present invention establishes an artificial neural network model by collecting and statistically analyzing the production capacity geological factor parameters and construction factor parameters that affect the production capacity of shale gas horizontal wells and the test production of the horizontal wells as a training set, and then optimizes the parameters of the artificial neural network model through a particle algorithm to obtain an optimal artificial neural network model. Finally, under the geological factor parameter conditions of the horizontal well to be constructed, the fitness function of the optimal artificial neural network model is calculated, thereby obtaining the optimal values of the construction factor parameters of the horizontal well to be constructed under the geological factor parameter conditions.
[0039] It solves the problem of difficulty in designing horizontal well parameters, and has positive practical significance for improving and ensuring the maximization of shale gas horizontal well production capacity.
[0040] Among them, in step 1, the normalization process adopts the min-max normalization method, and the formula is as follows:
[0041]
[0042]
[0043] Among them, in the formula, x i * is the normalized capacity influencing factor parameter; x i is the unnormalized capacity influencing factor parameter; x min For x i The minimum value of the parameter of the same capacity influencing factor; x max For x i The maximum value of the parameters of the same capacity influencing factors; y i * is the normalized test output; y i is the unnormalized test output; y min is the minimum value of the test output; y max The maximum value of the test output.
[0044] In the present invention, the correlation between the production capacity influencing factors of the artificial neural network model cannot be too strong, otherwise it will affect the accuracy of its prediction. Optimally, the correlation coefficient R between the production capacity influencing factors is calculated, and a heat map of the correlation coefficients between the production capacity influencing factors is drawn to ensure that the R values between the variables are all less than 0.5 to ensure that there is a weak correlation between most of the production capacity influencing factors. The calculation formula of the correlation coefficient R is an existing formula and is not specifically described in the present invention.
[0045] The geological factors in the present invention include one or more of vertical depth, pressure coefficient, porosity, total organic carbon content, total gas content, Young's modulus, Poisson's ratio and brittleness index. The construction factors include one or more of cluster number, displacement, flowback rate, fluid intensity, sand addition intensity, slick water ratio and quartz sand ratio.
[0046] The artificial neural network model in the present invention adopts the BP neural network model. In some embodiments, the artificial neural network model also adopts the CNN or RNN model.
[0047] The particle swarm optimization algorithm (PSO) is used to optimize the parameters of the trained artificial neural network model. This is an existing method. Specifically, the algorithm considers the best position of the particle swarm in history and the best position of each particle in history. The particle swarm gradually moves to the optimal position until the maximum iteration is reached. According to the principle of minimum mean square error, the particle swarm optimization algorithm is used to optimize the structure of the ANN, which is defined as:
[0048]
[0049] In the formula, N is the number of data samples; y i * and i It is the predicted value and true value of the test production per unit section length of shale gas horizontal wells in the data sample.
[0050] The particle position update formula used in this article is:
[0051]
[0052] i=1,2L mk=1,2L
[0053]
[0054] In the formula, the position of a single particle is recorded as a real vector, V ik+1 n and V ik n represents the velocity of particle i after iterating k+1 and k times in the n-th dimension; X ik+1 n and X ik n represents the position of particle i after iterating k+1 and k times in the nth dimension; c1 and c2 are learning factors; r1 and r2 are random values between 0 and 1; P ik n and G k n They represent the optimal position of a single particle and the optimal position of a particle group, respectively.
[0055]
[0056] Where k is the current iteration number of the colony particle population, T max The maximum number of iterations set for the colony particle population, ω max is the maximum inertia weight, ω min is the minimum inertia weight. max The general value is 0.9, min The general value is 0.4.
[0057] Taking the minimum MSE value of the particle swarm (corresponding to the optimal position of the swarm) as the goal, the PSO algorithm is used to optimize the number of hidden layers and the number of neurons in each hidden layer, so as to obtain the optimal neural network model structure and model parameters.
[0058] The error calculation between the predicted yield value and the test yield data in the test set is performed using the determination coefficient R 2, one or more of the root mean square error RMSE, the regression line slope k, and the Willmott consistency index IA.
[0059] Coefficient of determination R 2 : R 2 The value is used to measure the degree of correlation between two variables. The closer the R 2 value is to 1, the stronger the variable correlation and the more accurate the model prediction. According to the recommendation of statistics, R 2 > 0.64 is a good prediction model. The expression of R2 is:
[0060]
[0061] Root mean square error RMSE: RMSE is the standard deviation of the error between the actual value and the predicted value of the test production per unit length of shale gas. The smaller the RMSE, the more accurate the model prediction. The expression of RMSE is:
[0062]
[0063] Regression line slope k: k is the slope of the linear regression equation fitted between the actual value and the predicted value. The closer k is to 1 and the intercept is closer to 0, the more accurate the model prediction. According to the recommendation of statistics, 0.85 < K < 1.15 is a good prediction model. The expression of k is:
[0064]
[0065] Willmott consistency index IA: The range of IA is [0,1]. The closer IA is to 1, the more accurate the model prediction. IA = 0 indicates that the model prediction is invalid. According to the recommendation of statistics, IA > 0.80 is a good prediction model. The expression of IA is:
[0066]
[0067] In the above formulas, N is the number of data samples; yi and yi* are the actual value (test production data) and the predicted production value respectively; and y* are the average predicted production value and the average actual value respectively.
[0068] In the design method of the present invention, it also includes the step of determining the relative importance of the production capacity influencing factors of the optimal neural network model, that is, determining the importance of the influence of the production capacity influencing factors on the output of the optimal neural network model. In this way, when considering construction parameters, the construction factors with higher importance are focused on or preferentially satisfied.
[0069] The present invention uses a partial dependency diagram to determine the relative importance of the factors affecting the capacity of the optimal neural network model. A partial dependency diagram (PDP diagram) is a method for studying the dependence of predictions on input variables. A partial dependency diagram can represent the marginal impact of one or two features on the prediction results of a machine learning model, that is, how the variables affect the prediction results. The partial dependency function for regression is defined as:
[0070]
[0071] In the formula, x S is the characteristic variable of interest; x C denoted all other variables.
[0072] By x C By integrating, we can get a value that depends only on x S The function f(x S ). This function is called partial dependence function, which can realize the S In actual operation, the Monte Carlo method is usually used to obtain the partial dependence function by calculating the average value of the training set. The specific formula is as follows:
[0073]
[0074] In the formula, n represents the sample size.
[0075] The specific steps of using partial dependence diagram to determine the relative importance of the production capacity influencing factors of the optimal neural network model are as follows:
[0076] According to the partial dependency function, a partial dependency graph corresponding to each capacity influencing factor is established;
[0077] According to the slope of the curve in the partial dependence diagram, the relative importance of each input variable to the prediction of shale gas horizontal well production capacity is obtained. The larger the slope of the curve in the partial dependence diagram, the stronger the relative importance of the corresponding input variable. The relative importance of all production capacity influencing factors is ranked from large to small, so that the relative importance of production capacity influencing factors in the optimal artificial neural network model can be obtained.
[0078] Example:
[0079] Step 1: Collect and count the parameters of the production capacity influencing factors that affect the production capacity of shale gas horizontal wells and the test production of the horizontal wells, normalize the production variable parameters and the test production, and divide the training set and the test set into a ratio of 9:1. The production capacity influencing factor parameters include geological factor parameters and construction factor parameters.
[0080] In the embodiment of the present invention, the parameters of the factors affecting the productivity of 317 shale gas horizontal wells in the Sichuan Basin are statistically analyzed. The parameters of the factors affecting the productivity include 8 geological factor parameters (vertical depth, pressure coefficient, porosity, total organic carbon content, total gas content, Young's modulus, Poisson's ratio, brittleness index) and 7 construction factor parameters (number of clusters, displacement, backflow rate, liquid intensity, sand addition intensity, slick water ratio, quartz sand ratio) and their unit section length test production. Its distribution histogram is as follows Figure 3 and Figure 4 shown.
[0081] Calculate the correlation coefficient R between the parameters of the factors affecting production capacity, Figure 5 The correlation coefficient R between the parameters of the production capacity influencing factors is given. In the figure, the serial numbers 1-15 represent vertical depth, pressure coefficient, porosity, total organic carbon content, total gas content, Young's modulus, Poisson's ratio, brittleness index, number of clusters, displacement, backflow rate, slick water ratio, quartz sand ratio, liquid strength, and sand addition strength. It can be seen from the figure that the R values between most parameters of the production capacity influencing factors are less than 0.5, indicating that there is a weak correlation between most parameters of the production capacity influencing factors.
[0082] Step 2: Using the parameters of the production capacity influencing factors in the training set as input variables and the horizontal well test production as output variables, an artificial neural network model is established according to the mapping relationship between the parameters of the production capacity influencing factors and the horizontal well test production. The artificial neural network model adopts the BP neural network model. The number of neurons in the input layer of the artificial neural network model is determined by the number of production capacity influencing factors, and the number of neurons in the output layer is determined by the number of output quantities.
[0083] Artificial neural networks have the ability to self-learn and adapt. They can analyze and grasp the potential laws between the input and output data provided in advance, and finally use new input data to infer the output results based on these laws. Each neuron is a basic computing unit that executes the following commands:
[0084] y=max(0,∑ i w i x i +b)
[0085] In the formula, xi is the neuron input; ωi is the weight; b is the threshold; y is the neuron output.
[0086] All neurons are connected in a layered structure where the mapping between input and output is done using the following formula:
[0087] h i =max(0,w i ·h i-1 +b i ),1≤i≤L,h0=x
[0088] y=max(0,V·h L )
[0089] Where L is the number of layers; W1-WL is a matrix; V and b1-bL are model parameters learned from the dataset.
[0090] Step 3: Use a particle algorithm to optimize the parameters of the artificial neural network model to obtain an optimal neural network model, wherein the parameters include the number of hidden layers and the number of neurons.
[0091] The tuning ranges of the number of hidden layers and neurons of the artificial neural network model are set to 1-4 and 1-120; the particle swarm size is set to 30, the maximum number of iterations is set to 200, and c1 and c2 are set to 1.2; with the goal of obtaining the minimum MSE value of the particle swarm, the change law of the minimum MSE value with the number of iterations under different neural network structures is obtained, so as to obtain the optimal neural network model structure. Figure 6 The figure shows the variation between the minimum MSE value of the particle swarm (corresponding to the optimal position of the swarm) and the number of iterations when using different numbers of hidden layers (1, 2, 3, 4). When the number of iterations is greater than 1, the minimum mean square error of the particle swarm decreases, indicating that the particle algorithm is effective for the structural adjustment of the artificial neural network. In the first 21 iterations, the minimum mean square error of the particle swarm gradually decreases with the increase of the number of iterations. The ANN model with 1 hidden layer (the hidden layer contains 69 neurons) can obtain a minimum mean square error of 2.681×10 -6 , the optimized artificial neural network model structure is as follows Figure 6 shown.
[0092] Step 4: Use the test set to verify the optimal neural network model.
[0093] The determination coefficient R between the predicted value and the actual value of shale gas unit section length test production is calculated using training samples and test samples. 2 , root mean square error RMSE, regression line slope k, Weir model consistency index IA and other statistical descriptors are used to evaluate the reliability of the trained ANN model. According to statistics, a good prediction can be obtained through R 2 >0.64,0.85 <K<1.15,IA> 0.80 evaluation. R of the training sample 2 =0.876, RMSE = 0.002, k = 0.986, IA = 0.968. R of the test sample 2 =0.847, RMSE=0.0009, k=0.957, IA=0.965. The trained neural network model meets the requirements.
[0094] Step 5: Determine the relative importance of the capacity influencing factors of the optimal neural network model.
[0095] 5.1, such as Figure 7 As shown, a partial dependence diagram corresponding to each capacity influencing factor parameter is established;
[0096] 5.2. According to the slope of the curve in the partial dependence diagram, the relative importance of all capacity influencing factors is ranked from large to small, so that the capacity influencing factors with relatively strong importance in the optimal artificial neural network model can be obtained. The importance ranking of capacity influencing factors is as follows: Figure 8 shown.
[0097] Step 6: Under the geological factor parameter conditions of the horizontal well to be constructed, a genetic algorithm is used to calculate the fitness function of the optimal neural network model, so as to obtain the optimal values of the construction factor parameters of the horizontal well to be constructed under the geological factor parameter conditions.
[0098] The eight geological factor parameters of the horizontal well to be constructed are taken as the input variables of the optimal neural network model, and the seven construction factor parameters are optimized to obtain the maximum unit section length test production. The eight geological factor parameters are shown in Table 1. The optimization process is as follows Figure 2 shown.
[0099] Table 1 Geological parameters of shale gas wells
[0100]
[0101] The optimization results show that when the number of clusters in a single section is designed to be 4, the displacement is 18.99m3 / min, the return rate is 29.36%, the proportion of slick water is 94.37%, the proportion of quartz sand is 78.58%, the liquid intensity is 43.85m3 / m, and the sand addition intensity is 4.42t / m, the maximum unit section length test output of 0.0167×104m3 / m can be obtained.
[0102] This construction parameter is used to guide the horizontal well construction, focusing on or giving priority to satisfying the construction factor parameters that have the greatest impact on production capacity.
Claims
1. A shale gas horizontal well construction parameter optimization method based on artificial neural network, characterized in that: include: Collect and count the parameters of the productivity influencing factors affecting the productivity of shale gas horizontal wells and the test production of the horizontal wells, normalize the parameters of the productivity influencing factors and the test production, and divide the training set and the test set in proportion. The productivity influencing factors include geological factors and construction factors. Taking the parameters of the productivity influencing factors in the training set as input variables and the horizontal well test production as output variables, an artificial neural network model is established according to the mapping relationship between the parameters of the productivity influencing factors and the horizontal well test production; Using a particle algorithm to optimize the parameters of the artificial neural network model to obtain an optimal neural network model, wherein the parameters include the number of hidden layers and the number of neurons; The optimal neural network model is verified by using the test set, and the capacity influencing factor parameters of the test set are used as input to obtain the predicted output value, and the error between the predicted output value and the test output data in the test set is calculated. If the error meets the requirements, the optimal neural network model meets the requirements; Under the geological factor parameter conditions of the horizontal well to be constructed, a genetic algorithm is used to calculate the fitness function of the optimal neural network model, so as to obtain the optimal values of the construction factor parameters of the horizontal well to be constructed under the geological parameter conditions.
2. The method for optimizing shale gas horizontal well construction parameters based on artificial neural network according to claim 1, characterized in that: The correlation coefficient R between the production capacity influencing factors is less than 0.
5.
3. The method for optimizing shale gas horizontal well construction parameters based on artificial neural network according to claim 1 or 2, characterized in that: The geological factors include one or more of vertical depth, pressure coefficient, porosity, total organic carbon content, total gas content, Young's modulus, Poisson's ratio and brittleness index.
4. The method for optimizing shale gas horizontal well construction parameters based on artificial neural network according to claim 1 or 2, characterized in that: The construction factors include one or more of the number of clusters, displacement, flowback rate, liquid intensity, sand addition intensity, slippery water ratio and quartz sand ratio.
5. The method for optimizing shale gas horizontal well construction parameters based on artificial neural network according to claim 1, characterized in that: The artificial neural network model is a BP neural network model.
6. The method for optimizing shale gas horizontal well construction parameters based on artificial neural network according to claim 1, characterized in that: The error between the predicted yield value and the test yield data in the test set is calculated using the determination coefficient R 2 , root mean square error RMSE, regression line slope k, and Weir model consistency index IA.
7. The method for optimizing shale gas horizontal well construction parameters based on artificial neural network according to claim 1, characterized in that: The method also includes a step of determining the relative importance of the capacity influencing factors of the optimal neural network model, so as to give priority to satisfying the construction factor parameters with large relative importance during construction.
8. The method for optimizing shale gas horizontal well construction parameters based on artificial neural network according to claim 7, characterized in that: The steps to determine the relative importance of the factors affecting the production capacity of the optimal neural network model include: Establish a partial dependency graph corresponding to each capacity influencing factor; According to the slope of the curve in the partial dependence diagram, the relative importance of all capacity influencing factors is ranked from large to small, so that the relative importance of capacity influencing factors in the optimal artificial neural network model can be obtained.
9. The method for optimizing shale gas horizontal well construction parameters based on artificial neural network according to claim 1, characterized in that: The normalization process adopts the min-max standardization method.
10. The method for optimizing shale gas horizontal well construction parameters based on artificial neural network according to claim 1, characterized in that: The division ratio of the training set and the test set is 9:1.