Method, system and equipment for predicting recoverable reserves of tight oil reservoir and medium

Abstract

The invention belongs to the technical field of tight oil reservoir exploitation, and provides a method, a system, equipment and a medium for predicting recoverable reserves of a tight oil reservoir, and the method comprises the following steps: collecting single well historical production data of the tight oil reservoir, and arranging and constructing a time sequence according to a time sequence; and calculating autocorrelation and partial autocorrelation functions, and selecting an AR, MA or ARMA model as a prediction model according to function features. After the model is constructed, historical production data is used for training, and a trained prediction model is obtained; and inputting single well historical production data into the model, and outputting a future predicted final recoverable reserve prediction value. The method can accurately predict the predicted final recoverable reserves of the tight oil reservoir single well.

Description

Technical Field

[0001] The present invention belongs to the technical field of tight oil reservoir exploitation technology, and in particular relates to a method, system, equipment and medium for predicting recoverable reserves of tight oil reservoirs. Background Art

[0002] China has rich tight oil resources with a wide distribution range. The recoverable resources are close to 5 billion tons, ranking third in the world. At present, China's tight oil exploration has also made progress in many basins, demonstrating the huge potential of my country in tight oil exploration and development. Overall, China has rich tight oil resources, but its effective exploration and development is still in the exploratory stage.

[0003] Tight oil reservoirs are an important unconventional oil and gas resource with complex geological conditions, low permeability and poor reservoir properties. This makes the conventional method of exploiting tight oil reservoirs have low production capacity. Although it is difficult to exploit, tight oil and gas reserves are abundant. Their successful development can supplement my country's oil and gas resources, improve the energy structure, and is of great significance to ensuring my country's energy security. Accurate prediction of single well EUR (estimated ultimate recoverable reserves) is of great significance for the economic evaluation of oil reservoirs and the formulation of development plans.

[0004] The quality of tight oil resources in different regions varies. In some areas, the oil reservoirs are small in scale and poor in quality. The initial production is high, but it will soon enter a declining stage. The production cost of tight oil is relatively high. If the oil price fluctuates greatly, when the oil price is low, the company may reduce investment, affecting the application of mining technology and production scale, resulting in a decline in production. During the mining process, improper operation or the use of chemicals may cause damage to the reservoir, reduce its permeability and porosity, and affect production. Although some progress has been made in tight oil extraction technology in recent years, there are still some technical bottlenecks. For example, the limited effect of fracturing and the imperfection of enhanced recovery technology may lead to a decline in production.

[0005] Tight oil reservoirs have complex geological conditions and changeable seepage characteristics. Traditional EUR prediction methods often fail to achieve satisfactory results. Commonly used EUR prediction methods have problems such as low accuracy and poor adaptability.

[0006] The long-term production prediction methods of oil and gas wells can be roughly divided into three categories:

[0007] Empirical production decline models, analytical / semi-analytical solution models and numerical simulations.

[0008] The empirical production decline model is widely used because it only requires oil and gas production data and is simple to operate. The most commonly used and easy-to-use empirical decline methods are: Arps production decline method; SEPD method based on Arps exponential decline and improved based on a large number of examples; Duong method based on the decline model dominated by fracture flow. These three methods are simpler than other empirical methods, and the calculation results are more objective and accurate. They are currently the most widely used in oil fields.

[0009] (1) The Arps model includes three modes: exponential decline, hyperbolic decline and harmonic decline. The application of these models depends on the characteristics of the actual production data and the decline trend. In practical applications, it is necessary to select a suitable decline model for production prediction based on the production history of the specific oil field, and to determine the decline type based on the actual production data. Commonly used judgment methods include graphical method, trial and error method, curve displacement method, typical curve fitting method and binary regression method. Only after determining the decline type and related parameters can the corresponding formula be used for production prediction. Therefore, this method has relatively large limitations.

[0010] (2) SEPD model, namely the stretched exponential production decline model, attempts to more accurately describe the production characteristics of unconventional oil and gas resources by modifying the traditional exponential decline model. Since the production process of unconventional oil and gas reservoirs is different from that of traditional oil and gas reservoirs, especially the production decline behavior in the early stage of production, the SEPD model provides a more flexible framework to adapt to this change, and shows certain limitations in practical applications. On the one hand, it has not undergone rigorous theoretical derivation and relies more on empirical data, which increases the uncertainty of the prediction results. On the other hand, when the SEPD model is applied to certain specific types of oil and gas reservoirs, such as tight gas reservoirs with extremely low permeability, it may lead to conservative predictions of technically recoverable reserves, affecting the formulation of production decisions. Therefore, this method also has certain limitations.

[0011] (3) The Duong production decline method is a nonlinear production prediction model that is particularly suitable for tight oil and gas wells. It models the linear relationship of production data in double logarithmic coordinates and then predicts future production and cumulative production. A significant advantage of the Duong model is that it can adapt to production wells under different permeability conditions. However, when the production well enters the late stage, that is, when the boundary effect occurs, the Duong model may overestimate future gas production. Therefore, this method also has certain limitations.

[0012] In short, the above methods all have some defects. The Arps production decline method is not applicable in many cases, the SEPD method prediction results are often too small, the Duong method prediction results are often too large, and are easily affected by fluctuations in production data. Some experts and scholars have summarized the applicability of various methods and put forward some views, but have not conducted in-depth analysis of the applicability of these methods in combination with the decline law and examples. Summary of the invention

[0013] To solve the above problems, the present disclosure provides a method, system, device and medium for predicting recoverable reserves of tight oil reservoirs, which adopts a time series prediction model to more accurately predict the recoverable reserves of tight oil reservoirs.

[0014] The following is the technology of the present invention:

[0015] A method for predicting recoverable reserves of a tight oil reservoir, comprising:

[0016] Collect historical production data of individual wells in tight oil reservoirs;

[0017] Arrange historical production data in chronological order to construct a time series;

[0018] Calculate the autocorrelation function and partial autocorrelation function of the time series; if the partial autocorrelation function is truncated after the pth order and the autocorrelation function shows a tail, select the AR model as the time series prediction model; if the autocorrelation function is truncated after the qth order and the partial autocorrelation function shows a tail, select the MA model as the time series prediction model; if both the autocorrelation function and the partial autocorrelation function show a tailing feature, select the ARMA model as the time series prediction model; where p is the autoregressive order and q is the sliding average order;

[0019] Build a time series prediction model and use historical production data to train the time series prediction model to obtain a trained prediction model;

[0020] The historical production data of a single well in a tight oil reservoir is input into the trained prediction model, and the predicted value of the future estimated ultimate recoverable reserves of the single well in the tight oil reservoir is output.

[0021] Further,

[0022] Before calculating the autocorrelation function and partial autocorrelation function of the time series, perform the following steps:

[0023] The non-stationary historical data Yt in the time series is subjected to d (d = 0, 1, dots, n) times of difference processing. After each difference processing, the historical data is subjected to a run test until the differenced data passes the stationarity test, thereby obtaining a new stationary data series X1, X2, ..., X t-d ;

[0024] Take the first N groups of data as observation data and perform zero mean processing, that is: Get a new sequence X′ t As a time series.

[0025] Further,

[0026] The expression of the ARMA model is:

[0027]

[0028] When q=0, the ARMA model becomes an AR model:

[0029]

[0030] When p = 0, the RMA model becomes the MA model:

[0031] X t =∈ t -θ1∈ t -1-...-θ q ∈ t-q , t∈Z (3)

[0032] Where, the non-negative integer p is the autoregressive order, is the autoregressive coefficient; the non-negative integer q is the sliding average order, θ1,...,θ q is the sliding average coefficient; X t is the sequence related to spare parts consumption data, ε t is WN(0,σ2).

[0033] Further,

[0034] The constructing of the time series prediction model comprises:

[0035] When constructing the ARMA model, the sample moment estimation method, the least squares estimation method or the maximum likelihood estimation method are used to estimate the autoregressive coefficient, the moving average coefficient and the white noise variance of the ARMA.

[0036] The AIC and BIC criteria were used to determine the order of the ARMA model.

[0037] Further,

[0038] After selecting the ARMA model as the time series forecasting model, the stability of the time series forecasting model is tested, which includes:

[0039] For the characteristic equation of the autoregressive part in the ARMA model, if all the roots of the characteristic equation of the autoregressive part are outside the unit circle, the model is judged to be stationary.

[0040] Further,

[0041] After selecting the ARMA model as the time series forecasting model, the reversibility of the time series forecasting model is tested, which includes:

[0042] For the characteristic equation of the sliding average part, check whether its roots are outside the unit circle. When all roots are outside the unit circle, the model satisfies reversibility.

[0043] Further,

[0044] After selecting the time series prediction model, determine whether the residual sequence of the time series prediction model is white noise. If not, reselect the model.

[0045] A system for predicting recoverable reserves of tight oil reservoirs, characterized by comprising:

[0046] Data acquisition module, used to collect historical production data of single wells in tight oil reservoirs;

[0047] The sequence building module is used to arrange the historical production data in chronological order and build a time series;

[0048] Model selection module, used to calculate the autocorrelation function and partial autocorrelation function of the time series; if the partial autocorrelation function is truncated after the pth order and the autocorrelation function shows a tail, select the AR model as the time series prediction model; if the autocorrelation function is truncated after the qth order and the partial autocorrelation function shows a tail, select the MA model as the time series prediction model; if both the autocorrelation function and the partial autocorrelation function show a tailing feature, select the ARMA model as the time series prediction model; where p is the autoregressive order and q is the sliding average order;

[0049] Model training module, used to build a time series prediction model, train the time series prediction model using historical production data, and obtain a trained prediction model;

[0050] The prediction module is used to input the historical production data of a single well in a tight oil reservoir into a trained prediction model, and output the predicted value of the future estimated ultimate recoverable reserves of the single well in the tight oil reservoir.

[0051] Compared with the prior art, the present invention has the following advantages:

[0052] By collecting the historical production data of single wells in tight oil reservoirs with multiple parameters such as production, pressure, and temperature, and then constructing a time series in chronological order, the trend of data changes over time can be clearly presented, making it easier to discover the laws.

[0053] Then the autocorrelation function and partial autocorrelation function are calculated. The calculation of autocorrelation function and partial autocorrelation function is crucial for accurately selecting the model; the autocorrelation function can reveal the dynamic structure of the data, such as the production of tight oil reservoirs, which changes over time by considering the correlation of data at different time lags. The tailing characteristics indicate that it is affected by the comprehensive influence of data over a long period of time, reflecting the coherence changes caused by multiple factors in the mining process. The partial autocorrelation function measures the correlation between variables without the influence of intermediate variables, and can determine the key lag order that affects the current data point. The truncation phenomenon means that the direct influence of the previous data is weakened after exceeding a certain order. When the partial autocorrelation function is truncated after the pth order and the autocorrelation function is tailing, the data is mainly affected by the linear influence of its own data before the pth order, which is consistent with the inertia of the tight oil reservoir system. The AR(p) model predicts the future value by the linear combination of the previous p-period data plus the error term to adapt to it. For example, when the partial autocorrelation function of a reservoir's production is truncated at a specific order, the model can effectively utilize this relationship; if the autocorrelation function is truncated after the qth order and the partial autocorrelation function is tailing, it is greatly affected by external random factors. The MA(q) model can better handle it by using the linear combination of current and past error terms based on the accumulation of error terms to represent the current value; if both are tailing, the data is affected by the linear influence of its own historical data and the accumulation of error terms. The ARMA(p,q) model combines the characteristics of AR and MA to comprehensively characterize the complex dynamic changes caused by various factors such as pressure changes and random interference in the production of tight oil reservoirs, thereby accurately selecting the model.

[0054] The selected model is trained using historical production data, so that the model can learn the inherent laws and implicit information from the data, improve the prediction ability, and finally use the trained model to predict the future estimated final recoverable reserves. This solution is data-driven and fully considers the inherent relationship of the data, thus overcoming the problems of low accuracy and poor adaptability of traditional methods.

[0055] Other features and advantages of the present disclosure will be described in the following description, and partly become apparent from the description, or be understood by implementing the present disclosure. The purpose and other advantages of the present disclosure can be realized and obtained by the structures pointed out in the description, claims and drawings. BRIEF DESCRIPTION OF THE DRAWINGS

[0056] In order to more clearly illustrate the embodiments of the present disclosure or the technical solutions in the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of the present disclosure. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative work.

[0057] Figure 1 A schematic diagram of the method of the present invention is shown;

[0058] Figure 2A single well production performance curve diagram of a tight oil reservoir according to an embodiment of the present invention is shown;

[0059] Figure 3 A diagram showing the single well prediction results of the tight oil reservoir ARIMA model according to an embodiment of the present invention;

[0060] Figure 4 A basic flow chart of the ARIMA model of an embodiment of the present invention is shown. DETAILED DESCRIPTION

[0061] In order to make the purpose, technical solution and advantages of the embodiments of the present disclosure clearer, the technical solution in the embodiments of the present disclosure will be clearly and completely described below in conjunction with the drawings in the embodiments of the present disclosure. Obviously, the described embodiments are part of the embodiments of the present disclosure, not all of the embodiments. Based on the embodiments in the present disclosure, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present disclosure.

[0062] Figure 1 A schematic diagram of the method according to the present invention is shown, and the specific implementation details of the present invention include:

[0063] 1. Step 1: For the study of tight oil reservoir blocks, collect historical production dynamic data of single wells in tight oil reservoirs, including geological data of wells that have been put into production, production dynamic data and other static and dynamic data, to achieve data input, database construction and management.

[0064] 2. Step 2: Preprocessing:

[0065] Filter and remove redundant data, completely supplement missing data, correct or delete erroneous data, and finally organize it into data that can be further processed and used.

[0066] 3. Step 3: Time series construction:

[0067] Arrange the sorted data (historical production dynamic data of single wells in tight oil reservoirs) in chronological order to construct a time series. Perform a preliminary analysis of the time series to identify its long-term trends, seasonal changes, cyclic changes, and irregular changes.

[0068] 4. Step 4: Model selection and construction:

[0069] According to the characteristics of the time series, select the appropriate time series prediction model - ARIMA model, use historical data to train the model, verify and test it, adjust the parameters multiple times to achieve the fitting accuracy, obtain a high-precision and generalization-capable prediction model, and determine the parameters and structure of the model.

[0070] 1) Introduction to ARIMA model:

[0071] Specifically, the ARIMA model contains three main components: autoregression (AR), differencing (I), and moving average (MA).

[0072] In the autoregressive (AR) part, the model uses past data points to predict future values. This model assumes that the current value has a linear relationship with the historical values. (For example, if AR(p) is used to represent a model with p autoregressive terms, then the value in period n can be represented as a linear combination of the values ​​in the previous p periods, plus an error term.)

[0073] The moving average (MA) model focuses on the accumulation of error terms. Specifically, the current value can be expressed as a linear combination of current and past error terms. (The ARMA model is formed by combining autoregression (AR) and moving average (MA), which combines the characteristics of both to model time series more comprehensively.)

[0074] In order to apply the ARIMA model, the time series needs to be differentiated to make it a stationary time series. The purpose of the difference is to eliminate the trend and seasonality in the time series and keep the mean value of the data stable. This step is the main difference between the ARIMA model and the simple ARMA model.

[0075] The essence of the ARIMA (p, d, q) model is to first perform d (d = 0, 1, dots, n) times of difference processing on the non-stationary historical data Yt to obtain a new stationary data series Xt, fit Xt to the ARMA (p, q) model, and then restore the original d times of difference to obtain the predicted data of Yt.

[0076] Among them, the general expression of ARMA(p,q) is:

[0077]

[0078] In the formula, the first half is the autoregressive part, the non-negative integer p is the autoregressive order, is the autoregressive coefficient, the latter part is the sliding average part, the non-negative integer q is the sliding average order, θ1,...,θ q is the sliding mean coefficient; Xt is the related sequence of spare parts consumption data, and εt is WN(0,σ2).

[0079] When q = 0, the model becomes an AR(p) model:

[0080]

[0081] When p = 0, the model becomes a MA(q) model:

[0082] X t =∈ t -θ1∈ t-1-...-θ q ∈ t- q,t∈Z (3)

[0083] 2) Predictive modeling process, such as Figure 4

[0084] (1) Obtain data and preprocess them. Collect data sequences, denoted as Y1, Y2, ..., Y t The run test method is used to determine whether the sequence is a stationary sequence; if it is a non-stationary sequence, the difference method is used, that is: Y′ t-i =Y t -Y t-1 , perform stationary preprocessing on the sequence, perform run test on the data after each difference, until the data obtained by difference can pass the stationary test, recorded as d times difference, and obtain the new stationary sequence X1, X2, ..., X t-d Take the first N groups (or all) of data as observation data and perform zero mean processing, that is: Get a set of preprocessed new sequences X′ t .

[0085] (2) Identification of ARMA Model

[0086] By calculating the preprocessed sequence X′ t Autocorrelation function (ACF) of and partial autocorrelation function (PACF) To perform model identification. The specific calculation formula is:

[0087]

[0088] According to the above calculation results and the model identification principle in Table 1, X′ can be determined t Model that fits.

[0089] Table 1 ARMA (p, q) model identification principles

[0090] Model AR(p) MA(q) ARMA(pq) Autocorrelation function Tailing, exponential decay or oscillation Finite length, truncated (q steps) Tailing, exponential decay or oscillation Partial autocorrelation function Finite length, truncated (p steps) Tailing, exponential decay or oscillation Tailing, exponential decay or oscillation

[0091] (3) Parameter estimation and model order determination

[0092] Parameter estimation and model order determination are important contents in establishing prediction models, and the two influence each other. Based on the above model identification, the unknown parameters of ARMA (p, q), namely the autoregressive coefficient, the moving average coefficient and the white noise variance, are estimated using the sample moment estimation method, the least squares estimation method or the maximum likelihood estimation method, and the results are as follows: The AIC and BIC criteria were used to determine the model order.

[0093] (4) Model verification

[0094] First, we need to check whether the established model can satisfy the stability and reversibility, that is, it is required that the roots of the following equations (6) and (7) are outside the unit circle. The specific formula is as follows:

[0095]

[0096] Further determine whether the residual sequence of the above model is white noise. If not, the model needs to be re-identified. If so, the software reliability prediction model is obtained through testing:

[0097]

[0098] (5) EUR forecast

[0099] According to the above prediction model, X′ is predicted by the one-step prediction method. t Make a prediction and consider the d differences performed previously to restore the predicted result of data Yt.

[0100] 5. Step 5: Use the trained model to predict the future EUR of a single well in a tight oil reservoir. Consider the impact of geological conditions, engineering technology, production time and other factors on EUR, and make necessary corrections and adjustments to the prediction results.

[0101] The following are specific examples of the present invention:

[0102] In a certain northwestern region, the exploration area covers an area of ​​7,300 square kilometers, with an estimated geological oil resource of 4.6 billion tons. It is a rare oil-rich zone and a typical tight conglomerate reservoir, and horizontal fracturing is its main development technology. Since the block was put into production in 2016, it has exposed problems such as large differences in the development effect of horizontal fracturing wells and difficulty in predicting the recoverable reserves of a single well.

[0103] The method provided by the present invention can quickly and accurately predict the recoverable reserves of the fractured horizontal wells in the study area, thereby providing a scientific basis for the efficient development of the block. Figure 1 shown.

[0104] Step S101: For the horizontal wells that have been put into production and fractured in the tight oil reservoirs in the study area, collect monthly production data, input, build a database, and manage them, as shown in Table 1. Figure 2 .

[0105] Step S102: Screen and clear redundant data, supplement missing data, correct or delete erroneous data, and organize them into oil well production sample data that can be further processed and used.

[0106] In combination with the data actually collected in the block, it is necessary to properly clean and integrate the basic sample data using the method described in step 2 of the technical solution of the present invention to establish the original sample data of the recoverable reserves of single horizontal wells in the block.

[0107] Step S103: Time series construction, arranging the sorted data in chronological order to construct a time series, as shown in Table 2.

[0108] Step S104: Model selection and construction. According to the characteristics of the time series, select the appropriate time series prediction model - ARIMA model, use historical data to train the model, verify and test it, adjust the parameters multiple times to achieve the fitting accuracy, obtain a high-precision and generalization-capable prediction model, and determine the parameters and structure of the model.

[0109] Step S105: EUR prediction, using the trained model to predict the future EUR of a single well in a tight oil reservoir. Considering the impact of geological conditions, engineering technology, production time and other factors on EUR, necessary corrections and adjustments are made to the prediction results. Figure 3 and Table 3.

[0110] Table 1 Production dynamic data table

[0111]

[0112]

[0113] Table 2 Construction of time series table

[0114]

[0115]

[0116] Table 3 Prediction results data table

[0117]

[0118]

[0119] like Figure 1 As shown, the production prediction results of this application are basically consistent with the actual situation, with small relative error and high prediction accuracy. Therefore, the technical solution of this patent can meet the production needs of EUR prediction of tight oil reservoirs.

[0120] In summary, compared with the prior art, the present invention has the following advantages:

[0121] (1) A comprehensive EUR prediction system for tight oil reservoir production wells has been established, solving the problem that existing prediction methods are easily affected by production data fluctuations. The present invention expands the technical means for predicting the production of tight oil reservoir wells and can meet the actual production needs on site.

[0122] (2) Based on production dynamic data, a high-precision production prediction model is established using actual samples through artificial intelligence technology to achieve EUR prediction for tight oil wells. The shortcomings of poor applicability and low accuracy of existing prediction methods are avoided. The production predicted by the EUR prediction method proposed in the present invention is highly consistent with the actual data and can meet on-site needs.

[0123] Based on the method of the present invention, the embodiment of the present disclosure also provides a prediction system corresponding to the above method, which includes:

[0124] Data acquisition module, used to collect historical production data of single wells in tight oil reservoirs;

[0125] The sequence building module is used to arrange the historical production data in chronological order and build a time series;

[0126] Model selection module, used to calculate the autocorrelation function and partial autocorrelation function of the time series; if the partial autocorrelation function is truncated after the pth order and the autocorrelation function shows a tail, select the AR(p) model as the time series prediction model; if the autocorrelation function is truncated after the qth order and the partial autocorrelation function shows a tail, select the MA(q) model as the time series prediction model; if both the autocorrelation function and the partial autocorrelation function show a tailing feature, select the ARMA(p,q) model as the time series prediction model;

[0127] Model training module, used to build a time series prediction model, train the time series prediction model using historical production data, and obtain a trained prediction model;

[0128] The prediction module is used to input the historical production data of a single well in a tight oil reservoir into a trained prediction model, and output the predicted value of the future estimated ultimate recoverable reserves of the single well in the tight oil reservoir.

[0129] Based on the same inventive concept as the above disclosed content, an embodiment of the present disclosure also provides a prediction device corresponding to the above method, which includes at least one processor; and a memory communicatively connected to the at least one processor; wherein the memory stores instructions that can be executed by the at least one processor, and the instructions are executed by the at least one processor to enable the at least one processor to execute the above method.

[0130] It should be noted that the electrical connection between the above-mentioned units does not necessarily mean the connection between lines. An indirect connection method can be applied to the embodiments of the present disclosure as long as the purpose of the present disclosure is achieved.

[0131] Based on the same inventive concept, the present disclosure also provides a computer storage medium on which executable instructions are stored. When the instructions are executed by a processor, the processor performs the above method.

[0132] Although the present disclosure has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the aforementioned embodiments, or make equivalent replacements for some of the technical features therein; and these modifications or replacements do not make the essence of the corresponding technical solutions deviate from the spirit and scope of the technical solutions of the embodiments of the present disclosure.

Claims

1. A method for predicting recoverable reserves of tight oil reservoirs, characterized in that: include: Collect historical production data of individual wells in tight oil reservoirs; Arrange historical production data in chronological order to construct a time series; Calculate the autocorrelation function and partial autocorrelation function of the time series; if the partial autocorrelation function is truncated after the pth order and the autocorrelation function shows a tail, select the AR model as the time series prediction model; if the autocorrelation function is truncated after the qth order and the partial autocorrelation function shows a tail, select the MA model as the time series prediction model; if both the autocorrelation function and the partial autocorrelation function show a tailing feature, select the ARMA model as the time series prediction model; where p is the autoregressive order and q is the sliding average order; Build a time series prediction model and use historical production data to train the time series prediction model to obtain a trained prediction model; The historical production data of a single well in a tight oil reservoir is input into the trained prediction model, and the predicted value of the future estimated ultimate recoverable reserves of the single well in the tight oil reservoir is output.

2. The method for predicting recoverable reserves of a tight oil reservoir according to claim 1, characterized in that: Before calculating the autocorrelation function and partial autocorrelation function of the time series, perform the following steps: The non-stationary historical data Yt in the time series is subjected to d (d = 0, 1, dots, n) times of difference processing. After each difference processing, the historical data is subjected to a run test until the differenced data passes the stationarity test, thereby obtaining a new stationary data series X1, X2, ..., X t-d ; Take the first N groups of data as observation data and perform zero mean processing, that is: Get a new sequence X′ t As a time series.

3. The method for predicting recoverable reserves of a tight oil reservoir according to claim 1, characterized in that: The expression of the ARMA model is: When q=0, the ARMA model becomes an AR model: When p = 0, the RMA model becomes the MA model: X t =∈ t -θ1∈ t -1-...-θ q ∈ t-q ,t∈Z (3) Where, the non-negative integer p is the autoregressive order, is the autoregressive coefficient; the non-negative integer q is the sliding average order, θ1,...,θ q is the sliding average coefficient; X t is the sequence related to spare parts consumption data, ε t is WN(0,σ2).

4. The method for predicting recoverable reserves of a tight oil reservoir according to claim 1, characterized in that: The constructing of the time series prediction model comprises: When constructing the ARMA model, the sample moment estimation method, the least squares estimation method or the maximum likelihood estimation method are used to estimate the autoregressive coefficient, the moving average coefficient and the white noise variance of the ARMA. The AIC and BIC criteria were used to determine the order of the ARMA model.

5. The method for predicting recoverable reserves of a tight oil reservoir according to claim 1, characterized in that: After selecting the ARMA model as the time series forecasting model, the stability of the time series forecasting model is tested, which includes: For the characteristic equation of the autoregressive part in the ARMA model, if all the roots of the characteristic equation of the autoregressive part are outside the unit circle, the model is judged to be stationary.

6. A method for predicting recoverable reserves of a tight oil reservoir according to claim 5, characterized in that: After selecting the ARMA model as the time series forecasting model, the reversibility of the time series forecasting model is tested, which includes: For the characteristic equation of the sliding average part, check whether its roots are outside the unit circle. When all roots are outside the unit circle, the model satisfies reversibility.

7. The method for predicting recoverable reserves of a tight oil reservoir according to claim 1, characterized in that: After selecting the time series prediction model, determine whether the residual sequence of the time series prediction model is white noise. If not, reselect the model.

8. A system for predicting recoverable reserves of tight oil reservoirs, characterized in that: include: Data acquisition module, used to collect historical production data of single wells in tight oil reservoirs; The sequence building module is used to arrange the historical production data in chronological order and build a time series; Model selection module, used to calculate the autocorrelation function and partial autocorrelation function of the time series; if the partial autocorrelation function is truncated after the pth order and the autocorrelation function shows a tail, select the AR model as the time series prediction model; if the autocorrelation function is truncated after the qth order and the partial autocorrelation function shows a tail, select the MA model as the time series prediction model; if both the autocorrelation function and the partial autocorrelation function show a tailing feature, select the ARMA model as the time series prediction model; where p is the autoregressive order and q is the sliding average order; Model training module, used to build a time series prediction model, train the time series prediction model using historical production data, and obtain a trained prediction model; The prediction module is used to input the historical production data of a single well in a tight oil reservoir into a trained prediction model, and output the predicted value of the future estimated ultimate recoverable reserves of the single well in the tight oil reservoir.

9. A device for predicting recoverable reserves of a tight oil reservoir, comprising: at least one processor; And a memory communicatively connected to the at least one processor; wherein the memory stores instructions executable by the at least one processor, and the instructions are executed by the at least one processor so that the at least one processor can execute the method for predicting recoverable reserves of a tight oil reservoir as described in any one of claims 1 to 7.

10. A computer storage medium having executable instructions stored thereon, wherein when the instructions are executed by a processor, the processor is enabled to implement the method according to any one of claims 1 to 7.

Images