CN104732303A - Oil field output prediction method based on dynamic radial basis function neural network - Google Patents

Abstract

A method for predicting oilfield production based on dynamic radial basis function neural network, comprising the following steps: A. Determine factors affecting production according to oilfield conditions, obtain historical data and divide it into training data sets and detection data sets; B. use isolated The difference normalization method is used to normalize the data set; C. Use the sensitivity method to dynamically adjust the RBF neural network structure, and establish a temporary RBF neural network prediction model; D. Use the state transition probability matrix to correct the model error to obtain a stable RBF Neural network oilfield production prediction model; E. use the detection data set obtained in step A to verify the model, and judge whether it meets expectations; F. use the expected production prediction model obtained in step E to predict oil production. This method avoids the problem of too many or too small neurons in the RBF hidden layer, and the obtained model has the function of self-adaptive adjustment; the prediction error is corrected twice to make the prediction result more accurate and reasonable.

Description

A Method of Oilfield Production Prediction Based on Dynamic Radial Basis Function Neural Network

technical field

The present invention relates to a method for predicting oilfield production based on a dynamic radial basis function neural network, in particular to a method for dynamically optimizing the structure of a radial basis function neural network through a sensitivity method, combined with a state transition probability method to correct residuals, and realizing Oilfield Production Prediction Methods.

Background technique

As the lifeblood of the national economy, oil production directly affects the country's economic development. For oil field production, to ensure a good economic benefit, there must be a high and stable oil production. Ensuring high and stable oilfield production is the central task of oilfield development and production. Therefore, the accurate prediction of oil production in oilfields has always been one of the important research tasks of oilfield development workers.

Factors affecting oil production in oil fields can be roughly divided into two categories: geological factors and human factors. Geological factors, to some extent, are immutable or the changes are small. However, the range of human factors is much wider, ranging from mining methods, well patterns, well spacing, injection-production intensity, drilling adjustment wells, shutting down and scrapping old wells or switching injection and production, etc., to various artificial measures (including compression Cracking, acidification, hole repairing and layer adjustment, electric pump modification, hydraulic pump, overhaul, etc.), each change of human factors will affect the change of oil production in the oil field. Therefore, the methods for predicting oil production in oilfields are also based on geological factors, human factors, or a combination of the two. They are mainly divided into two types: one is to study and predict the oil production of oilfields as a whole from the perspective of system theory. The other is to study the effect of increasing production and injection of individual measures. The first type of methods mainly include statistical formula method (empirical method), water drive characteristic curve method, material balance equation method, and oil and gas reservoir numerical simulation method. However, the statistical formula method (empirical method), the water drive characteristic curve method and the material balance equation method have certain defects: first, the influence of reservoir heterogeneity on oil production cannot be directly considered; second, it is impossible to consider various human factors. The impact of changes in factors on oilfield production. Theoretically speaking, the numerical simulation method of oil and gas reservoirs can comprehensively and directly consider the influence of geological factors of reservoirs and changes of various human factors on oilfield production. However, its dependence on geological data is too large, often resulting in errors in the understanding of reservoir geological conditions, making the prediction results of oil and gas reservoir numerical simulation useless. Although the understanding of reservoir geology can be corrected by finely fitting the oilfield development history, the fitting results are ambiguous. Accurate fitting of the oilfield development history often requires researchers to have solid geological knowledge, oil Knowledge of reservoir, oil recovery technology, mathematical calculation and computer knowledge, and the workload is heavy. The second type of method is to study the influence of various human factors on oil production in isolation, which goes against the fact that oil field development is a large system. Oilfield development is a complex nonlinear dynamical system. Oilfield oil production forecast is a multi-factor nonlinear forecasting problem. Therefore, in order to ensure the scientificity and accuracy of oilfield production forecasting, a new oilfield production forecasting method is urgently needed, so that the production forecasting results are more accurate, objective and reasonable.

Contents of the invention

From the perspective of artificial intelligence, the present invention uses the sensitivity method to optimize and adjust the radial basis function (RBF) neural network structure, uses the data samples of oil field production influencing factors to train the RBF neural network, and uses the state transition probability method to correct the neural network, so as to realize The output prediction result is more accurate, objective and reasonable.

In order to achieve the above purpose, a method for predicting oilfield production based on dynamic radial basis function neural network is provided, which mainly includes the following steps:

A. Get data

According to the actual situation of the oilfield, determine the indicators of factors affecting the oilfield production, obtain the historical data set and divide it into a training data set and a detection data set;

B. Normalization processing

To normalize the historical data set, the normalization method can use the deviation standardization method to convert the data of different dimensions into a unified processing format. The conversion function is as follows:

$x^{*}=\frac{x-x_{\min }}{x_{\max }-x_{\min }}$ (Formula 1)

Where x _max is the maximum value of the sample data, and x _min is the minimum value of the sample data;

C. Establishment and training of prediction model

In the RBF neural network, let K be the number of neurons in the hidden layer, x(x ₁ ,…,x _m ) be the input vector, α _k is the connection weight between the kth hidden layer neuron and the output layer neuron, φ _k is the output of the kth hidden layer neuron, so the output of the RBF neural network can be described as:

$\mathrm{the\; y}=\underset{k=1}{\overset{K}{\mathrm{\Σ}}}{\mathrm{\α}}_k{\mathrm{\φ}}_k\left(x\right)$ (Formula 2)

(1) A hidden layer neuron is given as an RBF neural network with any natural number for training, and the training times are set.

(2) Calculate the output value of each hidden layer neuron. From the formula (1), the output of the kth hidden layer neuron can be obtained as:

${\mathrm{\φ}}_k\left(x\right)=e^{\frac{-\left|\right|x-{\mathrm{\μ}}_k\left|\right|}{{\mathrm{\σ}}_k^2}}$ (Formula 3)

(3) Conduct sensitivity analysis on the output of each neuron, and calculate its contribution to the output. The weighted value of the output of the neurons in the hidden layer is used as the input of the sensitivity method, and the contribution of the output of the neurons in the hidden layer to the output of the neural network is calculated by using the following formula:

$S_h=\frac{{\mathrm{var}}_h\left[\mathrm{E.}\left(\mathrm{the\; y}\right|Z_h={\mathrm{\α}}_h{\mathrm{\φ}}_h\left(x\right))\right]}{\mathrm{var}\left(\mathrm{the\; y}\right)}$ (Formula 4)

Among them, Z=[Z ₁ ,Z ₂ ,...,Z _k ] is the input vector of the sensitivity method, y is the output of the neural network, and the relationship between y and Z can be expressed as y=f(Z ₁ ,Z ₂ ,...,Z _k ), var _h [E(y|Z _h ＝α _h φ _h (x))] is the influence of Z _h on the variance of y when Z h is equal to α _h φ _h (x), var(y) is the variance of y, and _Sh is the first-order sensitivity representation of α _h φ _h to the output y. Normalize _Sh :

${\mathrm{ST}}_h=\frac{S_h}{\underset{i=1}{\overset{K}{\mathrm{\Σ}}}{S}_i}$ (Formula 5)

(4) Select the value of ε and adjust the neural network structure according to the output contribution value of hidden layer neurons. The value of ε is generally smaller than the target error value, split the hidden layer neurons with the largest contribution value and greater than ε ₁ , and delete the hidden layer neurons with a contribution value less than ε ₂ , where ε ₁ >ε ₂ , and finally achieve Adjust the neural network structure. Define the error objective function as (N is the number of training samples):

$\mathrm{E.}=\frac{1}{2N}\underset{j=1}{\overset{N}{\mathrm{\Σ}}}{e}_j^2$ (Formula 6)

(5) According to the target error function, the gradient descent algorithm is used to adjust the output weight, central value and function width of the hidden layer neurons of the neural network:

$a_i(t+1)=a_i\left(t\right)-{\mathrm{\η}}_1{\mathrm{ST}}_i\frac{\∂\mathrm{E.}}{{\∂a}_i\left(t\right)}$ (Formula 7)

${\mathrm{\μ}}_i(t+1)={\mathrm{\μ}}_i\left(t\right)-{\mathrm{\η}}_2{\mathrm{ST}}_i\frac{\∂\mathrm{E.}}{{\∂\mathrm{\μ}}_i\left(t\right)}$ (Formula 8)

${\mathrm{\σ}}_i(t+1)={\mathrm{\σ}}_i\left(t\right)-{\mathrm{\η}}_3{\mathrm{ST}}_i\frac{\∂\mathrm{E.}}{{\∂\mathrm{\σ}}_i\left(t\right)}$ (Formula 9)

Among them, η ₁ , η ₂ , η ₃ are parameter learning step sizes.

(6) Stop calculation when the desired error or calculation step is reached.

D. Residual Correction

On the basis of the established neural network model, the error sequence is calculated by comparing the predicted value of the training sample with the actual output value. The error sequence is regarded as a Markov process, and the state is divided, and the probability is replaced by the frequency, and the transition probability matrix of the error state is calculated.

E. Model Validation

Use the detection data set to test the model. If the error between the output prediction value and the actual comparison value reaches the expected expectation, the neural network model is trained successfully, and the model can be used to predict oil production; otherwise, the model training is immature and needs to be restarted. to train.

F. Oil Production Forecast

Obtain real forecast basic data and input it into the trained and optimized RBF neural network. The output of the RBF neural network is the predicted value of oilfield production.

The beneficial effect of the present invention is that, compared with the general neural network prediction model, the judgment basis is more objective, and through the adjustment of the neural network structure, the finally obtained neural network structure is relatively compact, has good self-adaptive ability, and makes the evaluation result more scientific and accurate , fair and reasonable.

Description of drawings

Fig. 1 is a flow chart of oil field production prediction method based on dynamic radial basis function neural network.

Detailed ways

The present invention will be described in further detail below in conjunction with the accompanying drawings.

Step 1: Get Data

According to the actual situation of the oilfield, determine the indicators of factors affecting the oilfield production, obtain the historical data set and divide it into a training data set and a detection data set;

The second step: normalization processing

The historical data set is normalized, and the normalization method can use the deviation standardization method to transform the data of different dimensions into a unified processing format. The conversion function is as follows:

${\mathrm{<span\; class="notranslate">\; x</span>}}^{<span\; class="notranslate">\; *</span>}<span\; class="notranslate">\; =</span>\frac{\mathrm{<span\; class="notranslate">\; x</span>}<span\; class="notranslate">\; -</span>{\mathrm{<span\; class="notranslate">\; x</span>}}_{\mathrm{<span\; class="notranslate">\; min</span>}}}{{\mathrm{<span\; class="notranslate">\; x</span>}}_{\mathrm{<span\; class="notranslate">\; max</span>}}<span\; class="notranslate">\; -</span>{\mathrm{<span\; class="notranslate">\; x</span>}}_{\mathrm{<span\; class="notranslate">\; min</span>}}}$

Where x _max is the maximum value of the sample data, and x _min is the minimum value of the sample data;

The third step: the establishment and training of the prediction model

In the RBF neural network, let K be the number of neurons in the hidden layer, x(x ₁ ,…,x _m ) be the input vector, α _k is the connection weight between the kth hidden layer neuron and the output layer neuron, φ _k is the output of the kth hidden layer neuron, so the output of the RBF neural network can be described as:

$\mathrm{<span\; class="notranslate">\; the\; y</span>}<span\; class="notranslate">\; =</span>\underset{\mathrm{<span\; class="notranslate">\; k</span>}<span\; class="notranslate">\; =</span>\mathrm{<span\; class="notranslate">\; 1</span>}}{\overset{\mathrm{<span\; class="notranslate">\; K</span>}}{\mathrm{<span\; class="notranslate">\; \&Sigma;</span>}}}{\mathrm{<span\; class="notranslate">\; \&alpha;</span>}}_{\mathrm{<span\; class="notranslate">\; k</span>}}{\mathrm{<span\; class="notranslate">\; \&phi;</span>}}_{\mathrm{<span\; class="notranslate">\; k</span>}}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; x</span>}<span\; class="notranslate">\; )</span>$

(1) A hidden layer neuron is given as an RBF neural network with any natural number for training.

(2) Calculate the output value of each hidden layer neuron. From the formula (1), the output of the kth hidden layer neuron can be obtained as:

${\mathrm{<span\; class="notranslate">\; \&phi;</span>}}_{\mathrm{<span\; class="notranslate">\; k</span>}}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; x</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; =</span>{\mathrm{<span\; class="notranslate">\; e</span>}}^{\frac{<span\; class="notranslate">\; -</span><span\; class="notranslate">\; |</span><span\; class="notranslate">\; |</span>\mathrm{<span\; class="notranslate">\; x</span>}<span\; class="notranslate">\; -</span>{\mathrm{<span\; class="notranslate">\; \&mu;</span>}}_{\mathrm{<span\; class="notranslate">\; k</span>}}<span\; class="notranslate">\; |</span><span\; class="notranslate">\; |</span>}{{\mathrm{<span\; class="notranslate">\; \&sigma;</span>}}_{\mathrm{<span\; class="notranslate">\; k</span>}}^{\mathrm{<span\; class="notranslate">\; 2</span>}}}}$

(3) Conduct sensitivity analysis on the output of each neuron, and calculate its contribution to the output. The weighted value of the output of the neurons in the hidden layer is used as the input of the sensitivity method, and the contribution of the output of the neurons in the hidden layer to the output of the neural network is calculated by using the following formula:

${\mathrm{<span\; class="notranslate">\; S</span>}}_{\mathrm{<span\; class="notranslate">\; h</span>}}<span\; class="notranslate">\; =</span>\frac{{\mathrm{<span\; class="notranslate">\; var</span>}}_{\mathrm{<span\; class="notranslate">\; h</span>}}<span\; class="notranslate">\; [</span>\mathrm{<span\; class="notranslate">\; E.</span>}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; the\; y</span>}<span\; class="notranslate">\; |</span>{\mathrm{<span\; class="notranslate">\; Z</span>}}_{\mathrm{<span\; class="notranslate">\; h</span>}}<span\; class="notranslate">\; =</span>{\mathrm{<span\; class="notranslate">\; \&alpha;</span>}}_{\mathrm{<span\; class="notranslate">\; h</span>}}{\mathrm{<span\; class="notranslate">\; \&phi;</span>}}_{\mathrm{<span\; class="notranslate">\; h</span>}}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; x</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; )</span><span\; class="notranslate">\; ]</span>}{\mathrm{<span\; class="notranslate">\; var</span>}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; the\; y</span>}<span\; class="notranslate">\; )</span>}$

Among them, Z=[Z ₁ ,Z ₂ ,...,Z _k ] is the input vector of the sensitivity method, y is the output of the neural network, and the relationship between y and Z can be expressed as y=f(Z ₁ ,Z ₂ ,...,Z _k ), var _h [E(y|Z _h ＝α _h φ _h (x))] is the influence of Z _h on the variance of y when Z h is equal to α _h φ _h (x), var(y) is the variance of y, and _Sh is the first-order sensitivity representation of α _h φ _h to the output y. Perform Fourier transform on the input variable α _h φ _h (where the range of α _h φ _h is [a _h ,b _h ]):

${\mathrm{<span\; class="notranslate">\; Z</span>}}_{\mathrm{<span\; class="notranslate">\; h</span>}}<span\; class="notranslate">\; =</span>\frac{{\mathrm{<span\; class="notranslate">\; a</span>}}_{\mathrm{<span\; class="notranslate">\; h</span>}}<span\; class="notranslate">\; +</span>{\mathrm{<span\; class="notranslate">\; b</span>}}_{\mathrm{<span\; class="notranslate">\; h</span>}}}{\mathrm{<span\; class="notranslate">\; 2</span>}}<span\; class="notranslate">\; +</span>\frac{{\mathrm{<span\; class="notranslate">\; b</span>}}_{\mathrm{<span\; class="notranslate">\; h</span>}}<span\; class="notranslate">\; -</span>{\mathrm{<span\; class="notranslate">\; a</span>}}_{\mathrm{<span\; class="notranslate">\; h</span>}}}{\mathrm{<span\; class="notranslate">\; \&pi;</span>}}\mathrm{<span\; class="notranslate">\; arcsin</span>}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; sin</span>}<span\; class="notranslate">\; (</span>{\mathrm{<span\; class="notranslate">\; w</span>}}_{\mathrm{<span\; class="notranslate">\; h</span>}}\mathrm{<span\; class="notranslate">\; the\; s</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; )</span>$

Among them, w _h is the appropriate frequency selected, and the Fourier amplitude on the fundamental frequency is used to calculate the sensitivity. Formula 4 is finally transformed into a transformation:

${\mathrm{<span\; class="notranslate">\; S</span>}}_{\mathrm{<span\; class="notranslate">\; h</span>}}<span\; class="notranslate">\; =</span>\frac{<span\; class="notranslate">\; (</span>{\mathrm{<span\; class="notranslate">\; A</span>}}_{{\mathrm{<span\; class="notranslate">\; kw</span>}}_{\mathrm{<span\; class="notranslate">\; h</span>}}}^{\mathrm{<span\; class="notranslate">\; 2</span>}}<span\; class="notranslate">\; +</span>{\mathrm{<span\; class="notranslate">\; B</span>}}_{{\mathrm{<span\; class="notranslate">\; kw</span>}}_{\mathrm{<span\; class="notranslate">\; h</span>}}}^{\mathrm{<span\; class="notranslate">\; 2</span>}}<span\; class="notranslate">\; )</span>}{\underset{\mathrm{<span\; class="notranslate">\; j</span>}<span\; class="notranslate">\; =</span>\mathrm{<span\; class="notranslate">\; 1</span>}}{\overset{<span\; class="notranslate">\; +</span><span\; class="notranslate">\; \&infin;</span>}{\mathrm{<span\; class="notranslate">\; \&Sigma;</span>}}}<span\; class="notranslate">\; (</span>{\mathrm{<span\; class="notranslate">\; A</span>}}_{\mathrm{<span\; class="notranslate">\; j</span>}}^{\mathrm{<span\; class="notranslate">\; 2</span>}}<span\; class="notranslate">\; +</span>{\mathrm{<span\; class="notranslate">\; B</span>}}_{\mathrm{<span\; class="notranslate">\; j</span>}}^{\mathrm{<span\; class="notranslate">\; 2</span>}}<span\; class="notranslate">\; )</span>}$

Among them, -π<s<π, $A_j=\frac{1}{2\mathrm{\&pi;}}{\&Integral;}_{-\&infin;}^{\&infin;}f\left(\mathrm{the\; s}\right)\&times;\cos \left(w_j\mathrm{the\; s}\right)\mathrm{ds},$ To normalize S _h , record ST _h as the contribution value made by the hidden layer neuron output to the output of the neural network:

${\mathrm{<span\; class="notranslate">\; ST</span>}}_{\mathrm{<span\; class="notranslate">\; h</span>}}<span\; class="notranslate">\; =</span>\frac{{\mathrm{<span\; class="notranslate">\; S</span>}}_{\mathrm{<span\; class="notranslate">\; h</span>}}}{\underset{\mathrm{<span\; class="notranslate">\; i</span>}<span\; class="notranslate">\; =</span>\mathrm{<span\; class="notranslate">\; 1</span>}}{\overset{\mathrm{<span\; class="notranslate">\; K</span>}}{\mathrm{<span\; class="notranslate">\; \&Sigma;</span>}}}{\mathrm{<span\; class="notranslate">\; S</span>}}_{\mathrm{<span\; class="notranslate">\; i</span>}}}$

Define the error objective function as (N is the number of training samples):

$\mathrm{<span\; class="notranslate">\; E.</span>}<span\; class="notranslate">\; =</span>\frac{\mathrm{<span\; class="notranslate">\; 1</span>}}{\mathrm{<span\; class="notranslate">\; 2</span>}\mathrm{<span\; class="notranslate">\; N</span>}}\underset{\mathrm{<span\; class="notranslate">\; j</span>}<span\; class="notranslate">\; =</span>\mathrm{<span\; class="notranslate">\; 1</span>}}{\overset{\mathrm{<span\; class="notranslate">\; N</span>}}{\mathrm{<span\; class="notranslate">\; \&Sigma;</span>}}}{\mathrm{<span\; class="notranslate">\; e</span>}}_{\mathrm{<span\; class="notranslate">\; j</span>}}^{\mathrm{<span\; class="notranslate">\; 2</span>}}$

(4) Select the value of ε and adjust the neural network structure according to the output contribution value of hidden layer neurons. The value of ε is generally smaller than the target error value, split the hidden layer neurons with the largest contribution value and greater than ε ₁ , and adjust the neural network structure; assuming that the number of hidden layer neurons before splitting is K, the running time is t, and the contribution value The hidden layer neuron greater than ε ₁ is j, then the initial parameters of the newly added neuron K+1 and the parameters of neuron j are:

a _K+1 (t)=λ×a _j (t)

μ _K+1 (t) = μ _j (t)

σ _K+1 (t) = σ _j (t)

${\mathrm{<span\; class="notranslate">\; a</span>}}_{\mathrm{<span\; class="notranslate">\; j</span>}}^{<span\; class="notranslate">\; ,</span>}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; t</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; =</span><span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; 1</span>}<span\; class="notranslate">\; -</span>\mathrm{<span\; class="notranslate">\; \&lambda;</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; \&times;</span>{\mathrm{<span\; class="notranslate">\; a</span>}}_{\mathrm{<span\; class="notranslate">\; j</span>}}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; t</span>}<span\; class="notranslate">\; )</span>$

${\mathrm{<span\; class="notranslate">\; \&mu;</span>}}_{\mathrm{<span\; class="notranslate">\; j</span>}}^{<span\; class="notranslate">\; ,</span>}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; t</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; =</span>{\mathrm{<span\; class="notranslate">\; \&mu;</span>}}_{\mathrm{<span\; class="notranslate">\; j</span>}}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; t</span>}<span\; class="notranslate">\; )</span>$

${\mathrm{<span\; class="notranslate">\; \&sigma;</span>}}_{\mathrm{<span\; class="notranslate">\; j</span>}}^{<span\; class="notranslate">\; ,</span>}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; t</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; =</span>{\mathrm{<span\; class="notranslate">\; \&sigma;</span>}}_{\mathrm{<span\; class="notranslate">\; j</span>}}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; t</span>}<span\; class="notranslate">\; )</span>$

Among them, λ is an arbitrary constant in (0, 0.3) (set according to the actual needs of the oil field), and the parameters of the structure-invariant neurons are adjusted according to formulas (7) to (9).

(5) Delete hidden layer neurons whose contribution value is less than ε ₂ , and adjust the neural network structure; assuming that the running time is t, the hidden layer neuron whose contribution value is less than ε ₂ is i, and the Euclidean distance from neuron i The nearest neuron is ii, delete neuron i, and the parameters of neuron ii are:

${\mathrm{<span\; class="notranslate">\; a</span>}}_{\mathrm{<span\; class="notranslate">\; i</span>}}^{<span\; class="notranslate">\; ,</span>}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; t</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; =</span>{\mathrm{<span\; class="notranslate">\; a</span>}}_{\mathrm{<span\; class="notranslate">\; i</span>}}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; t</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; =</span>{\mathrm{<span\; class="notranslate">\; a</span>}}_{\mathrm{<span\; class="notranslate">\; i</span>}}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; t</span>}<span\; class="notranslate">\; )</span>\frac{{\mathrm{<span\; class="notranslate">\; \&phi;</span>}}_{\mathrm{<span\; class="notranslate">\; i</span>}}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; x</span>}<span\; class="notranslate">\; )</span>}{{\mathrm{<span\; class="notranslate">\; \&phi;</span>}}_{\mathrm{<span\; class="notranslate">\; i</span>}}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; x</span>}<span\; class="notranslate">\; )</span>}$

${\mathrm{<span\; class="notranslate">\; \&mu;</span>}}_{\mathrm{<span\; class="notranslate">\; i</span>}}^{<span\; class="notranslate">\; ,</span>}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; t</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; =</span>{\mathrm{<span\; class="notranslate">\; \&mu;</span>}}_{\mathrm{<span\; class="notranslate">\; i</span>}}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; t</span>}<span\; class="notranslate">\; )</span>$

${\mathrm{<span\; class="notranslate">\; \&sigma;</span>}}_{\mathrm{<span\; class="notranslate">\; i</span>}}^{<span\; class="notranslate">\; ,</span>}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; t</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; =</span>{\mathrm{<span\; class="notranslate">\; \&sigma;</span>}}_{\mathrm{<span\; class="notranslate">\; i</span>}}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; t</span>}<span\; class="notranslate">\; )</span>$

The parameters of other neurons are adjusted according to formulas (7)-(9).

(6) When the expected error or calculation step is reached, the calculation is stopped, and the model training ends.

Step 4: Residual Correction

On the basis of the established neural network model, the error sequence is calculated by comparing the predicted value of the training sample with the actual output value. The error sequence is regarded as a Markov process, and the state is divided, and the probability is replaced by the frequency, and the transition probability matrix of the error state is calculated.

Assume that there are k states in the state of the training data set, that is, there are states S ₁ , S ₂ ,...,S _k . Assuming that we are in state S _i now, and the probability of transferring to state S _j in the next step is recorded as P _ij , then the total state transition can be expressed by the following matrix:

in, $\underset{j=1}{\overset{k}{\mathrm{\&Sigma;}}}{p}_{\mathrm{ij}}=p_{i1}+p_{i2}+...+p_{\mathrm{ik}}=1,i=\mathrm{1,2},...,k.$

The relationship between transition probability matrices is: P ^(k) = P ^(k-1) × P, where P is a one-step transition probability matrix, and P ^(k) is a k-step transition probability matrix.

The relationship between transition states is: S ^(k) = S ⁽⁰⁾ × P ^(k) , where S ^(k) is the state vector after k steps of transition, and S ⁽⁰⁾ is the initial state vector.

Step 5: Model Validation

Use the detection data set to test the model. If the error between the output prediction value and the actual comparison value reaches the expected expectation, the neural network model is trained successfully, and the model can be used to predict oil production; otherwise, the model training is immature and needs to be restarted. Return to the third step for training.

Step 6: Oil Production Forecast

Obtain real forecast basic data and input it into the trained and optimized RBF neural network. The output of the RBF neural network is the predicted value of oilfield production.

Of course, the above-mentioned embodiments of the present invention are only descriptions of the present invention and cannot limit the present invention. Changes, modifications, additions or replacements made by those skilled in the art within the scope of the present invention shall also belong to the present invention scope of protection.

Claims (1)

1. a kind of oil field production prediction method based on dynamic radial basis function neural network is characterized in that, mainly comprises the following steps: A. Get data According to the actual situation of the oilfield, determine the indicators of factors affecting the oilfield production, obtain the historical data set and divide it into a training data set and a detection data set; B. Normalization processing To normalize the historical data set, the normalization method can use the deviation standardization method to convert the data of different dimensions into a unified processing format. The conversion function is as follows: $x^{*}=\frac{x-x_{\min }}{x_{\max }-x_{\min }}$ (Formula 1) Where x _max is the maximum value of the sample data, and x _min is the minimum value of the sample data; C. Establishment and training of prediction model In the RBF neural network, let K be the number of neurons in the hidden layer, x(x ₁ ,…,x _m ) be the input vector, α _k is the connection weight between the kth hidden layer neuron and the output layer neuron, φ _k is the output of the kth hidden layer neuron, so the output of the RBF neural network can be described as: $\mathrm{the\; y}=\underset{k=1}{\overset{K}{\mathrm{\&Sigma;}}}{\mathrm{\&alpha;}}_k{\mathrm{\&phi;}}_k\left(x\right)$ (Formula 2) (1) Given a hidden layer neuron is an RBF neural network of any natural number to train, set the number of times of training; (2) Calculate the output value of each hidden layer neuron, the output of the kth hidden layer neuron can be obtained from the formula (1): ${\mathrm{\&phi;}}_k\left(x\right)=e^{\frac{-\left|\right|x-{\mathrm{\&mu;}}_k\left|\right|}{{\mathrm{\&sigma;}}_k^2}}$ (Formula 3) (3) Conduct sensitivity analysis on the output of each neuron, calculate its contribution to the output, and use the weighted value of the output of the hidden layer neuron as the input of the sensitivity method, and use the following formula to calculate the output of the hidden layer neuron. Contribution of the neural network to the output: $S_h=\frac{{\mathrm{var}}_h\left[\mathrm{E.}\left(\mathrm{the\; y}\right|Z_h={\mathrm{\&alpha;}}_h{\mathrm{\&phi;}}_h\left(x\right))\right]}{\mathrm{var}\left(\mathrm{the\; y}\right)}$ (Formula 4) Among them, Z=[Z ₁ ,Z ₂ ,...,Z _k ] is the input vector of the sensitivity method, y is the output of the neural network, and the relationship between y and Z can be expressed as y=f(Z ₁ ,Z ₂ ,...,Z _k ), var _h [E(y|Z _h ＝α _h φ _h (x))] is the influence of Z _h on the variance of y when Z h is equal to α _h φ _h (x), var(y) is the variance of y, S _h is the first-order sensitivity representation of α _h φ _h to the output y, and S _h is normalized: ${\mathrm{ST}}_h=\frac{S_h}{\underset{i=1}{\overset{K}{\mathrm{\&Sigma;}}}{S}_i}$ (Formula 5) (4) Select the ε value and adjust the neural network structure according to the output contribution value of the hidden layer neurons. The value of ε is generally smaller than the target error value, and split the hidden layer neurons with the largest contribution value and greater than ε _1. For the contribution value The hidden layer neurons smaller than ε ₂ are deleted, here ε ₁ >ε ₂ , and finally the neural network structure is adjusted, and the error objective function is defined as (N is the number of training samples): $\mathrm{E.}=\frac{1}{2N}\underset{j=1}{\overset{N}{\mathrm{\&Sigma;}}}{e}_j^2$ (Formula 6) (5) According to the target error function, the gradient descent algorithm is used to adjust the output weight, central value and function width of the hidden layer neurons of the neural network: $a_i(t+1)=a_i\left(t\right)-{\mathrm{\&eta;}}_1{\mathrm{ST}}_i\frac{\&PartialD;\mathrm{E.}}{{\&PartialD;a}_i\left(t\right)}$ (Formula 7) ${\mathrm{\&mu;}}_i(t+1)={\mathrm{\&mu;}}_i\left(t\right)-{\mathrm{\&eta;}}_2{\mathrm{ST}}_i\frac{\&PartialD;\mathrm{E.}}{{\&PartialD;\mathrm{\&mu;}}_i\left(t\right)}$ (Formula 8) ${\mathrm{\&sigma;}}_i(t+1)={\mathrm{\&sigma;}}_i\left(t\right)-{\mathrm{\&eta;}}_3{\mathrm{ST}}_i\frac{\&PartialD;\mathrm{E.}}{{\&PartialD;\mathrm{\&sigma;}}_i\left(t\right)}$ (Formula 9) Wherein, η ₁ , η ₂ , η ₃ are parameter learning steps; (6) Stop calculation when the expected error or calculation step is reached; D. Residual Correction On the basis of the established neural network model, the predicted value of the training sample is compared with the actual output value, and the error sequence is calculated. The error sequence is regarded as a Markov process, and the state is divided, and the probability is replaced by the frequency, and the error state is calculated. Transition probability matrix; E. Model Validation Use the detection data set to test the model. If the error between the output prediction value and the actual comparison value reaches the expected expectation, the neural network model is trained successfully, and the model can be used to predict oil production; otherwise, the model training is immature and needs to be restarted. conduct training; F. Oil Production Forecast Obtain real forecast basic data and input it into the trained and optimized RBF neural network. The output of the RBF neural network is the predicted value of oilfield production.