Shale gas development index prediction method based on engineering parameters

Abstract

The application discloses a shale gas development index prediction method based on engineering parameters, comprising the following steps: obtaining historical engineering parameters corresponding to shale gas, and preprocessing the historical engineering parameters to obtain basic support data; a shale gas development index prediction model based on trend prediction is constructed by using a deep learning model, the shale gas development index prediction model is trained according to the basic support data and by using an improved optimization algorithm, and a trained shale gas development index prediction model is obtained; real-time production data corresponding to shale gas is obtained, shale gas production is predicted according to the real-time production data corresponding to shale gas and by using the trained shale gas development index prediction model, and a prediction result is obtained. The application collects historical production data corresponding to shale gas, uses a shale gas development index prediction model based on trend prediction and an improved optimization algorithm to perform deep learning on historical data of shale gas, so that the law in the historical data of shale gas is explored, and the accuracy of shale gas production prediction is improved.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of shale gas, in particular to a shale gas development index prediction method based on engineering parameters. BACKGROUND

[0002] Shale gas is a kind of unconventional natural gas mainly composed of methane, which is stored in mud shale and its interlayer with rich organic matter. Its high-quality, clean and efficient characteristics make it an effective low-carbon energy source to alleviate China's energy pressure, and it also occupies an increasingly important position in primary energy consumption.

[0003] The effective development and utilization of shale gas have important comprehensive benefits, which are manifested in the overall promotion of economy, ecology and society. Due to the low shale gas recovery rate and other factors, many shale gas development projects have been on the edge of economic feasibility for a long time, even in a loss situation, but its contribution to society and ecology cannot be ignored, so further comprehensive evaluation of such projects is needed, which has important practical significance for reasonable site selection, loss reduction and risk reduction.

[0004] In the prior art, the comprehensive benefits of shale gas are generally calculated to obtain the net profit in the entire project process. When calculating the comprehensive benefits, and under the condition that the market price does not change, the total benefits of shale gas are the yield multiplied by the market price, and the total benefits minus the cost are the comprehensive benefits. However, the prediction effect of shale gas yield in the prior art is poor, which leads to the inability to correctly predict the total benefits of shale gas, so that the management personnel cannot schedule personnel and equipment in real time according to the total benefits and profit targets, resulting in low shale gas mining benefits and a large amount of human and material resources being wasted. The prior art generally classifies historical data directly to predict shale gas yield, without considering the trend of feature changes, resulting in inaccurate prediction, and the existing technology model training is not good (such as using gradient descent method for training), which leads to poor final prediction effect. SUMMARY

[0005] The present application aims to provide a shale gas development index prediction method based on engineering parameters, which solves the technical problems existing in the prior art.

[0006] The present application is realized by the following technical scheme:

[0007] A shale gas development index prediction method based on engineering parameters, comprising:

[0008] Obtain the historical engineering parameters corresponding to the shale gas, and preprocess the historical engineering parameters to obtain the basic support data;

[0009] A shale gas development indicator prediction model based on trend prediction is constructed using a deep learning model, and the shale gas development indicator prediction model is trained based on the basic support data and an improved optimization algorithm to obtain a trained shale gas development indicator prediction model;

[0010] The real-time production data corresponding to shale gas is obtained, and the shale gas production is predicted based on the real-time production data corresponding to shale gas and the trained shale gas development indicator prediction model to obtain a prediction result.

[0011] In a possible implementation, obtaining historical engineering parameters corresponding to shale gas and preprocessing the historical engineering parameters include:

[0012] Obtaining historical equipment characteristics, historical site characteristics, and historical production of shale gas corresponding to multiple consecutive time periods, wherein each time period corresponds to a historical equipment characteristic, a historical site characteristic, and a historical production;

[0013] Cluster the historical production corresponding to all time periods to obtain multiple cluster centers, and obtain the Euclidean distance between the historical production of each time period and each cluster center. Determine the target cluster center with the shortest Euclidean distance to the historical production of the time period, and use the target cluster center to replace the historical production of the time period.

[0014] The feature matrix is ​​constructed using the historical equipment features and historical site features corresponding to N consecutive time periods. The output data is constructed using the target cluster center corresponding to the N+1th time end.

[0015] According to the characteristic matrix and the output data, training data is constructed, and a plurality of training data are repeatedly obtained to complete the preprocessing of the historical engineering parameters.

[0016] In one possible implementation, the historical production corresponding to all time periods is clustered to obtain multiple cluster centers, including:

[0017] Get the number of clusters as K and randomly generate K cluster centers;

[0018] Obtain the Euclidean distance between the historical output corresponding to each time period and the K cluster centers, determine the nearest cluster center with the shortest Euclidean distance to each historical output, and classify each historical output into the corresponding nearest cluster center;

[0019] Get the mean of all historical production corresponding to each cluster center, and determine whether there is a cluster center whose corresponding mean is different from the cluster center. If so, for the cluster center that is different from the mean, replace the cluster center with the mean and re-cluster until the mean corresponding to all cluster centers is the same as the cluster center, and get K cluster centers. Otherwise, get K cluster centers.

[0020] In a possible implementation, a shale gas development index prediction model based on trend prediction is constructed by using a deep learning model, and the method comprises the following steps:

[0021] A first feature conversion module, a feature preprocessing module, a second feature conversion module and a shale gas development index prediction module are sequentially connected to obtain a shale gas development index prediction model.

[0022] The first feature conversion module is configured to extract each column of data in the feature matrix and sequentially send the extracted data to the feature preprocessing module.

[0023] The feature preprocessing module comprises three parallel channels, and the three parallel channels are sequentially set as a conventional grey model, an unbiased grey model or a smooth modified grey model.

[0024] The second feature conversion module is configured to perform weighted processing on the three preprocessed data to obtain final preprocessed column data, and restore all the preprocessed column data to a feature matrix.

[0025] The shale gas development index prediction module is set as a deep convolutional neural network, and the restored feature matrix is used as an input of the shale gas development index prediction module to obtain a prediction result.

[0026] In a possible implementation, the shale gas development index prediction model is trained by using an intelligent optimization algorithm to obtain a trained shale gas development index prediction model, and the method comprises the following steps:

[0027] A1. Randomly initialize network parameters of the shale gas development index prediction model between an upper limit and a lower limit of the network parameters to obtain a single optimization individual, and repeatedly obtain multiple optimization individuals to obtain an optimization population.

[0028] A2. Obtain an adaptability value of each optimization individual, and determine an optimization individual with the maximum adaptability value to obtain an optimal individual.

[0029] A3. Update each optimization individual according to the optimal individual to obtain an optimization individual after initial updating.

[0030] A4. Perform local updating on the updated optimization individual, generate a new individual, replace M optimization individuals with the minimum adaptability value, and obtain an optimization population after secondary updating.

[0031] A5, based on the optimization population, determine whether to meet the iteration end condition, if yes, take the optimization individual with the maximum fitness value in the optimization population as the final network parameter of the shale gas development index prediction model, and end the training, otherwise return to step A2.

[0032] In a possible implementation, the fitness value of each optimization individual is obtained, and the optimization individual with the maximum fitness value is determined to obtain an optimal individual, including:

[0033] For each optimization individual, the fitness value thereof is obtained as:

[0034]

[0035] wherein i = 1, 2,..., I, I represents the total number of batch samples, j = 1, 2,..., J, J represents the total number of outputs of the shale gas development index prediction model, y ij represents the jth actual output of the shale gas development index prediction model after the ith sample input, represents the jth expected output of the shale gas development index prediction model after the ith sample input; i

[0036] The optimization individual with the maximum fitness value is determined to obtain an optimal individual.

[0037] In a possible implementation, each optimization individual is updated according to the optimal individual to obtain an optimization individual after initial updating, including:

[0038]

[0039]

[0040]

[0041]

[0042] wherein, represents the lth network parameter in the kth optimization individual in the optimization population at the t-1th training, represents the updated represents the update speed corresponding to the lth network parameter in the kth optimization individual, ω1 represents a first proportional weight coefficient, ω2 represents a second proportional weight coefficient, c represents a self-learning factor, f k represents the optimization frequency corresponding to the kth optimization individual, x best,l represents the lth network parameter in the optimal individual, ​denotes the update speed of the lth network parameter in the kth optimization individual at the t-1th training, c1 denotes the first speed update factor, c2 denotes the second speed update factor, p best,k,l denotes the lth network parameter of the kth optimization individual at the maximum fitness in the tth training, ω max denotes the maximum value of the proportional weight coefficient, ω min denotes the minimum value of the proportional weight coefficient, T max denotes the preset maximum training number, e denotes the natural constant, and t denotes the current training number.

[0043] In a possible implementation, the updated optimization individual is locally updated, including:

[0044] x' best,l = x best,l + εA t

[0045] A t = αA t-1

[0046] wherein x' best,l denotes the updated x best,l , ε denotes a random number between 0 and 1, A t denotes the average loudness of the optimization population at the tth training, A t-1 denotes the average loudness at the t-1th training; each optimization individual is assigned with an initial loudness before the iteration starts; and α denotes a random number between 0 and 1.

[0047] In a possible implementation, the new individual replaces the M optimization individuals with the minimum fitness values, to obtain the secondary updated optimization population, including:

[0048] determining the fitness value of each optimization individual in the primary updated optimization population, and determining M first target optimization individuals with the minimum fitness values;

[0049] for any first target optimization individual, randomly selecting a second target optimization individual from the optimization population, and performing a crossover operation on the first target optimization individual and the second target optimization individual to obtain two third target optimization individuals;

[0050] judging whether the fitness value of any third target optimization individual is greater than that of the first target optimization individual, if yes, replacing the first target optimization individual with the third target optimization individual with the maximum fitness value to obtain a new first target optimization individual, otherwise, performing a mutation operation;

[0051] for any first target optimization individual, performing a mutation on the parameters in the first target optimization individual to obtain a new first target optimization individual.

[0052] In a possible implementation, after each optimization individual is updated according to the optimal individual, the method further comprises updating the optimization frequency corresponding to the optimization individual, and the updating is as follows:

[0053]

[0054] wherein, represents the optimization frequency at the tth training, represents the initial optimization frequency of the kth optimization individual, and γ represents a frequency enhancement coefficient.

[0055] The shale gas development index prediction method based on engineering parameters provided by the application comprises the following steps: collecting historical production data corresponding to shale gas of a shale gas exploitation base, and performing deep learning on the historical data of the shale gas by using a shale gas development index prediction model based on trend prediction and an improved optimization algorithm, so as to explore the rules in the historical data of the shale gas and improve the accuracy of shale gas production prediction. BRIEF DESCRIPTION OF DRAWINGS

[0056] In order to more clearly illustrate the technical solutions of the exemplary embodiments of the application, the following will briefly introduce the drawings needed to be used in the embodiments, and it should be understood that the following drawings only show some embodiments of the application, and therefore should not be regarded as a limitation to the scope, and for those skilled in the art, other related drawings can also be obtained without creative labor. In the drawings:

[0057] Figure 1 The flowchart of the shale gas development index prediction method based on engineering parameters provided by the embodiment of the application. DETAILED DESCRIPTION

[0058] In order to make the purpose, technical solutions and advantages of the application more clear and obvious, the following will further describe the application in combination with the embodiments and drawings, and the exemplary embodiments of the application and the description thereof are only used to explain the application, and should not be regarded as a limitation to the application.

[0059] EMBODIMENT

[0060] As Figure 1 shown, the shale gas development index prediction method based on engineering parameters provided by the embodiment of the application comprises the following steps:

[0061] S1, obtaining historical engineering parameters corresponding to shale gas, and preprocessing the historical engineering parameters to obtain basic support data.

[0062] Optionally, the historical engineering parameters corresponding to the shale gas can include historical equipment characteristics and historical site characteristics, the historical equipment characteristics mainly including a type and a number of equipment used by the shale gas, and the historical site characteristics mainly including a type of land occupation, an area of each type of land occupation, and basic attributes of the shale gas.

[0063] The basic attributes of the shale gas can include shale composition, clay content, total organic carbon (TOC), organic matter maturity (Ro), porosity, net thickness of the shale, adsorbed gas content, and exploitation depth.

[0064] S2, a shale gas development index prediction model based on trend prediction is constructed by using a deep learning model, the shale gas development index prediction model is trained according to the basic support data and by using an improved optimization algorithm, and a trained shale gas development index prediction model is obtained.

[0065] S3, real-time production data corresponding to the shale gas is obtained, the shale gas production is predicted by using the trained shale gas development index prediction model according to the real-time production data corresponding to the shale gas, and a prediction result is obtained.

[0066] In a possible implementation, the historical engineering parameters corresponding to the shale gas are obtained, and the historical engineering parameters are preprocessed, including:

[0067] The historical equipment characteristics, the historical site characteristics, and the historical production of the shale gas corresponding to a plurality of continuous time periods are obtained, each time period corresponding to one historical equipment characteristic, one historical site characteristic, and one historical production.

[0068] Optionally, in the shale gas exploitation process, the geology and topography do not change suddenly, so the shale gas production is predicted by using a short-term prediction method, so that the staff can perform real-time scheduling and optimize the exploitation scheme to ensure the comprehensive benefits of the shale gas.

[0069] The main purpose of the embodiment of the present application is to predict the production of the shale gas, so that the manager can predict the total benefits according to the predicted production, and then schedule the personnel and equipment.

[0070] The historical equipment characteristics, historical site characteristics, and historical production of shale gas corresponding to multiple consecutive time periods may be data from a shale gas base currently being mined, or data from other shale gas bases.

[0071] Cluster the historical production corresponding to all time periods to obtain multiple cluster centers, and obtain the Euclidean distance between the historical production of each time period and each cluster center. Determine the target cluster center with the shortest Euclidean distance to the historical production of the time period, and use the target cluster center to replace the historical production of the time period.

[0072] The shale gas production and equipment usage vary between different mining bases, leading to variations in historical production. To reduce the amount of data, historical production is clustered. It's worth noting that the greater the number of clusters, the better the final prediction effect. This number can be set based on actual needs.

[0073] The feature matrix is ​​constructed using the historical equipment features and historical site features corresponding to N consecutive time periods. The output data is constructed using the target cluster center corresponding to the N+1th time end.

[0074] The feature matrix can be constructed using the following method: The historical equipment features and historical site features for each time period are combined into a feature vector, and the feature vectors for multiple consecutive time periods are combined into a feature matrix. It is worth noting that, to facilitate data processing, the various types of features in this embodiment are arranged in a fixed order and are all represented numerically.

[0075] Optionally, in order to reduce the amount of subsequent data processing, the elements in the feature matrix may be normalized to make the data easier to process and speed up the recognition process.

[0076] According to the characteristic matrix and the output data, training data is constructed, and a plurality of training data are repeatedly obtained to complete the preprocessing of the historical engineering parameters.

[0077] In one possible implementation, the historical production corresponding to all time periods is clustered to obtain multiple cluster centers, including:

[0078] The number of clusters is K, and K cluster centers are randomly generated.

[0079] Obtain the Euclidean distance between the historical output corresponding to each time period and the K cluster centers, determine the nearest cluster center with the shortest Euclidean distance to each historical output, and classify each historical output into the corresponding nearest cluster center.

[0080] The mean value of all historical production corresponding to each cluster center is obtained, and it is determined whether the mean value corresponding to the cluster center is different from the cluster center, if yes, for the cluster center which is different from the mean value, the mean value is used to replace the cluster center, and clustering is performed again until the mean value corresponding to all cluster centers is the same as the cluster center, K cluster centers are obtained, otherwise K cluster centers are obtained.

[0081] Optionally, if the historical production is one-dimensional data, the Euclidean distance between two data is the difference between the two data.

[0082] In a possible implementation, a shale gas development index prediction model based on trend prediction is constructed by using a deep learning model, comprising:

[0083] A first feature conversion module, a feature preprocessing module, a second feature conversion module and a shale gas development index prediction module connected in sequence are constructed to obtain the shale gas development index prediction model.

[0084] The first feature conversion module is used to extract each column of data of the feature matrix, and the extracted each column of data is sequentially sent into the feature preprocessing module.

[0085] The feature preprocessing module comprises three parallel channels, and the three parallel channels are sequentially set as a conventional grey model, an unbiased grey model or a smooth modified grey model, and each column of data extracted is preprocessed by the conventional grey model, the unbiased grey model or the smooth modified grey model to obtain three preprocessed column data.

[0086] The grey prediction refers to the prediction of the development and change of the characteristic value of the system behavior, so the three parallel grey models are used to predict the characteristic value in the present application, so that the characteristic value in the short term can be predicted.

[0087] The second feature conversion module is used to perform weighted processing on the three preprocessed data to obtain final preprocessed column data, and all the preprocessed column data are restored to a feature matrix.

[0088] By performing weighted processing on the three preprocessed data, the predicted characteristic value can be more accurately obtained, and the predicted characteristic value is used as the basis, so that the predicted shale gas production can be determined.

[0089] The shale gas development index prediction module is set as a deep convolutional neural network, and the restored feature matrix is used as the input of the shale gas development index prediction module to obtain a prediction result.

[0090] Optionally, the feature matrix constructed in the present application is equivalent to a simple image, so a deep convolutional neural network can be used to identify the feature matrix, and in addition to the deep convolutional neural network used in the present embodiment, other classification neural networks can also be used to explore the potential relationship between the feature matrix and the shale gas production.

[0091] The embodiment of the present application predicts the production, so as to obtain the overall benefit of shale gas development. Although the production module is fixed, the gas storage and the use of equipment can change, so the gas storage and the use of equipment in the short term can be predicted based on historical data, so as to introduce a grey model for small sample prediction, which can guarantee that the required parameters are predicted, and too many samples are not required. However, the grey module is easily affected by random factors, and the long-term prediction effect is not good, so the present application introduces an unbiased grey module and a grey model improved by a moving average method for prediction, which can effectively eliminate the influence of random factors and solve the problem of limited prediction length. At the same time, through the predicted data, the production can be accurately estimated.

[0092] In a possible implementation, a smart optimization algorithm is used to train the shale gas development index prediction model to obtain a trained shale gas development index prediction model, including:

[0093] A1, randomly initializing the network parameters of the shale gas development index prediction model between the upper limit and the lower limit of the network parameters to obtain a single optimization individual, and repeatedly obtaining a plurality of optimization individuals to obtain an optimization population.

[0094] Optionally, in order to better find the optimal network parameters of the shale gas development index prediction model, a chaotic sequence can be used to initialize the network parameters of the shale gas development index prediction model, so that the optimization individuals are more uniformly distributed in the solution space. It should be noted that the network parameters of the shale gas development index prediction model can include weights and biases.

[0095] A2, obtaining the fitness value of each optimization individual, and determining the optimization individual with the maximum fitness value to obtain the best individual.

[0096] A3, updating each optimization individual according to the best individual to obtain the optimization individual after the first update.

[0097] A4, locally updating the updated optimization individual and generating a new individual, replacing the M optimization individuals with the minimum fitness value to obtain the optimization population after the second update.

[0098] A5, determine whether the iteration end condition is met based on the optimization population, if yes, take the optimization individual with the maximum fitness value in the optimization population as the final network parameter of the shale gas development index prediction model, and end the training, otherwise return to step A2.

[0099] In order to solve the problems of slow convergence speed, low search precision and weak optimization capacity of the bat optimization algorithm, the bat optimization algorithm is improved in the embodiment, and the intelligent optimization algorithm is constructed based on the bat optimization algorithm, and the intelligent optimization algorithm is specifically as follows.

[0100] In a possible implementation, the fitness value of each optimization individual is obtained, and the optimization individual with the maximum fitness value is determined to obtain the best individual, including:

[0101] For each optimization individual, the fitness value thereof is obtained as:

[0102]

[0103] wherein, i=1, 2,..., I, I represents the total number of batch samples, j=1, 2,..., J, J represents the total number of outputs of the shale gas development index prediction model, y ij represents the jth actual output of the shale gas development index prediction model after the ith sample input, represents the jth expected output of the shale gas development index prediction model after the ith sample input. i The optimization individual with the maximum fitness value is determined to obtain the best individual.

[0104] In a possible implementation, each optimization individual is updated according to the best individual to obtain the optimization individual after the initial update, including:

[0105]

[0106]

[0107]

[0108]

[0109]

[0110] wherein, represents the lth network parameter in the kth optimization individual in the optimization population at the t-1th training, represents the updated represents the update speed corresponding to the lth network parameter in the kth optimization individual, ω1 represents the first proportional weight coefficient, ω2 represents the second proportional weight coefficient, c represents the self-learning factor, f k ​represents the optimization frequency corresponding to the kth optimization individual, x best,l represents the lth network parameter in the best individual, represents the update speed of the lth network parameter in the kth optimization individual at the t-1th training, c1 represents the first speed update factor, c2 represents the second speed update factor, p best,k,l represents the lth network parameter of the kth optimization individual when the fitness is maximum in the tth training process, ω max represents the maximum value of the proportional weight coefficient, ω min represents the minimum value of the proportional weight coefficient, T max represents the preset maximum training number, e represents the natural constant, and t represents the current training number.

[0111] Optionally, the self-learning factor c can be:

[0112]

[0113] wherein c max represents the first preset update value, c min represents the second preset update value, a represents a constant between 0 and 1, and π represents the circular constant.

[0114] For the update process of the optimization individual, the first proportional weight coefficient and the second proportional weight coefficient are set in the embodiment. In the early iteration, the second proportional weight coefficient is greater than the first proportional weight coefficient, so that the optimization individual can quickly approach the best individual, which is beneficial to the global search and accelerates the convergence speed. In the later iteration, the second proportional weight coefficient is less than the first proportional weight coefficient, and the positions of the optimization individual and the best individual are relatively close, which is beneficial to the local search, balances the local search and the global search, and effectively avoids the premature phenomenon of the algorithm. Through the improvement of the update speed, the algorithm has stronger search ability and convergence speed, so as to further improve the optimization ability of the intelligent optimization algorithm and train the shale gas development index prediction model with higher accuracy.

[0115] In a possible implementation, the updated optimization individual is locally updated, including:

[0116] x' best,l =x best,l +εA t

[0117] A t =αA t-1

[0118] wherein x' best,l represents the updated x best,l , ε represents a random number between 0 and 1, and A tA t represents the average loudness of the optimization population at the t-th training t-1 A t-1 represents the average loudness at the (t-1)-th training. Each optimization individual is assigned with an initial loudness before the iteration starts. a represents a random number between 0 and 1.

[0119] When the optimization population gradually approaches the global optimal solution, the algorithm adopts a local search strategy to more accurately find the global optimal solution.

[0120] In a possible implementation, the new individual replaces the M optimization individuals with the minimum fitness values to obtain the second updated optimization population, including: determining the fitness value of each optimization individual in the first updated optimization population, and determining M first target optimization individuals with the minimum fitness values. For any first target optimization individual, a second target optimization individual is randomly selected from the optimization population, and the first target optimization individual and the second target optimization individual are crossed to obtain two third target optimization individuals. It is judged whether the fitness value of the third target optimization individual is greater than the first target optimization individual, if yes, the third target optimization individual with the maximum fitness value is used to replace the first target optimization individual to obtain a new first target optimization individual, otherwise, a mutation operation is performed. For any first target optimization individual, the parameters in the first target optimization individual are mutated to obtain a new first target optimization individual.

[0121] In each iteration process, the optimization individual with the smaller fitness value can be identified as having less value, and therefore a new individual can be generated to update it to ensure the optimization ability of the entire optimization population. To avoid the population from becoming worse, the pre-mutation individual and the post-mutation individual can be compared in the mutation operation and the crossover operation, and the individual with the greater fitness value is retained. In addition, the pre-crossover individual and the post-mutation individual can be compared, and the individual with the greater fitness value is retained.

[0122] Optionally, the crossover operation can be to exchange part of the parameters in the two individuals in a one-to-one manner to obtain two new individuals. In this embodiment, the number of selected parameters is between 20% and 50% of the total number of parameters, and the parameters are selected in a random manner.

[0123] Optionally, the mutation operation can be: a decision factor between 0 and 1 is randomly generated, and it is judged whether the decision factor is greater than a preset decision probability, if yes, a first mutation operation is performed on the optimization individual, otherwise, a second mutation operation is performed on the optimization individual, and the decision probability is a constant between 0 and 1.

[0124] The first mutation operation can be: X * =X+w(X max -X), where X represents the individual to be mutated, X* represents a variation individual, X max represents an upper limit of the optimization individual, w represents a variation inertia weight.

[0125] The second variation operation can be: X * = X + w (X - X min ), wherein X min represents a lower limit of the optimization individual.

[0126] At the end of each training, the fitness values before and after the crossover and variation operations are performed on the optimization individual, and the better individual is selected to enter the next iteration, ensuring that the individual remaining after each iteration is an elite individual. At the same time, an adaptive decision probability is set in the variation process, which reduces the influence on the early trajectory of the algorithm optimization, increases the variation probability in the later stage of the algorithm, increases the probability of jumping out of the local extreme value in the later stage of the optimization, and prevents premature convergence into a local extreme value and difficulty in searching for a global optimal solution.

[0127] In a possible implementation, after each optimization individual is updated according to the best individual, the method further includes updating the optimization frequency corresponding to the optimization individual, and the updating is:

[0128]

[0129] wherein, represents the optimization frequency at the tth training, represents the initial optimization frequency of the kth optimization individual, and γ represents a frequency enhancement coefficient.

[0130] The optimization algorithm provided by the application can effectively avoid the problem of falling into a local optimum in the training process of the prior art, and effectively improves the training speed.

[0131] The shale gas development index prediction method based on engineering parameters provided by the application collects historical production data corresponding to shale gas for a certain shale gas exploitation base, and uses a shale gas development index prediction model based on trend prediction and an improved optimization algorithm to perform deep learning on the historical data of the shale gas, so as to explore the rules in the historical data of the shale gas and improve the accuracy of shale gas production prediction.

[0132] The above specific embodiments further specifically describe the purposes, technical solutions and beneficial effects of the application. It should be understood that the above description is only a specific embodiment of the application and is not used to limit the protection scope of the application. Any modification, equivalent replacement, improvement, etc. made within the spirit and principles of the application should be included in the protection scope of the application.

Claims

1. A shale gas development index prediction method based on engineering parameters, characterized in that: include: Obtaining historical engineering parameters corresponding to shale gas, and preprocessing the historical engineering parameters to obtain basic supporting data; A shale gas development indicator prediction model based on trend prediction is constructed using a deep learning model, and the shale gas development indicator prediction model is trained based on the basic support data and an improved optimization algorithm to obtain a trained shale gas development indicator prediction model; Obtaining real-time production data corresponding to shale gas, and using the trained shale gas development indicator prediction model to predict shale gas production based on the real-time production data to obtain prediction results; The pretreatment includes: Obtain historical equipment characteristics, historical site characteristics, and historical production of shale gas corresponding to multiple consecutive time periods, where each time period corresponds to a historical equipment characteristic, a historical site characteristic, and a historical production; Cluster the historical production corresponding to all time periods to obtain multiple cluster centers, and obtain the Euclidean distance between the historical production of each time period and each cluster center. Determine the target cluster center with the shortest Euclidean distance to the historical production of the time period, and use the target cluster center to replace the historical production of the time period. The feature matrix is ​​constructed using the historical equipment features and historical site features corresponding to N consecutive time periods. The output data is constructed using the target cluster center corresponding to the N+1th time period. According to the characteristic matrix and the output data, training data is constructed, and a plurality of training data are repeatedly obtained to complete the preprocessing of historical engineering parameters; A deep learning model is used to build a shale gas development indicator prediction model based on trend prediction, including: Constructing a first feature conversion module, a feature preprocessing module, a second feature conversion module, and a shale gas development index prediction module that are connected in sequence to obtain a shale gas development index prediction model; The first feature conversion module is used to extract each column of data from the feature matrix and sequentially send each column of extracted data to the feature preprocessing module; The feature preprocessing module includes three parallel channels, and the three parallel channels are sequentially set to a conventional gray model, an unbiased gray model, or a smoothly modified gray model, and each column of extracted data is preprocessed by the conventional gray model, the unbiased gray model, or the smoothly modified gray model to obtain three preprocessed column data; The second feature conversion module is used to perform weighted processing on the three preprocessed data to obtain final preprocessed column data, and restore all the preprocessed column data into a feature matrix; The shale gas development index prediction module is set as a deep convolutional neural network, and the restored feature matrix is ​​used as the input of the shale gas development index prediction module to obtain a prediction result; The shale gas development index prediction model is trained using an improved optimization algorithm to obtain a trained shale gas development index prediction model, including: A1. Randomly initialize the network parameters of the shale gas development indicator prediction model between the upper and lower limits of the network parameters to obtain a single optimal individual, and repeatedly obtain multiple optimal individuals to obtain an optimal population; A2. Obtain the fitness value of each optimization individual, and determine the optimization individual with the largest fitness value to obtain the best individual; A3. Update each optimal individual according to the best individual to obtain an optimal individual after initial update; A4. Perform a local update on the updated optimal individuals and generate new individuals to replace the M optimal individuals with the smallest fitness values ​​to obtain the optimal population after the second update. A5. Based on the optimized population, determine whether the iteration end condition is met. If so, use the optimized individual with the largest fitness value in the optimized population as the final network parameter of the shale gas development index prediction model and end the training. Otherwise, return to step A2.

2. The shale gas development index prediction method based on engineering parameters according to claim 1, characterized in that: Cluster the historical production corresponding to all time periods to obtain multiple cluster centers, including: Get the number of clusters as K and randomly generate K cluster centers; Obtain the Euclidean distance between the historical output corresponding to each time period and the K cluster centers, determine the nearest cluster center with the shortest Euclidean distance to each historical output, and classify each historical output into the corresponding nearest cluster center; Get the mean of all historical production corresponding to each cluster center, and determine whether there is a cluster center whose corresponding mean is different from the cluster center. If so, for the cluster center that is different from the mean, replace the cluster center with the mean and re-cluster until the mean corresponding to all cluster centers is the same as the cluster center, and get K cluster centers. Otherwise, get K cluster centers.

3. The shale gas development index prediction method based on engineering parameters according to claim 1, characterized in that: Obtain the fitness value of each optimization individual, and determine the optimization individual with the largest fitness value to obtain the best individual, including: For each optimal individual, obtain its fitness value as: Where i = 1, 2, ..., U, I represents the total number of batch samples, j = 1, 2, ..., J, J represents the total number of outputs of the shale gas development index prediction model, and y ij represents the jth actual output of the shale gas development index prediction model after the i-th sample input, represents the jth expected output of the shale gas development index prediction model after the i-th sample input; Determine the optimal individual with the largest fitness value and obtain the best individual.

4. The shale gas development index prediction method based on engineering parameters according to claim 3, characterized in that: Each optimal individual is updated according to the best individual to obtain the optimal individual after initial update, including: in, represents the lth network parameter of the kth optimizing individual in the optimizing population during the t-1th training. Indicates updated represents the update speed of the lth network parameter in the kth optimization individual, ω1 represents the first proportional weight coefficient, ω2 represents the second proportional weight coefficient, c represents the self-learning factor, f k represents the optimization frequency corresponding to the kth optimization individual, x best,l represents the lth network parameter in the best individual, Indicates the update speed corresponding to the lth network parameter in the kth optimization individual during the t-1th training, c1 represents the first speed update factor, c2 represents the second speed update factor, and p best,k,l represents the lth network parameter when the kth optimization individual has the maximum fitness during the tth training process, ω max Represents the maximum value of the proportional weight coefficient, ω min Indicates the minimum value of the proportional weight coefficient, T max Represents the preset maximum number of training times, e represents a natural constant, and t represents the current number of training times.

5. The shale gas development index prediction method based on engineering parameters according to claim 4 is characterized in that: Perform local updates on the updated optimal individuals, including: x' best,l =x best,l +εA t A t =αA t-1 Where x′ best,l Represents the updated x best,l , ε represents a random number between (0,1), A t Indicates the average loudness of the optimal population during the t-th training, A t-1 Represents the average loudness during the t-1th training; each optimal individual is assigned an initial loudness before the start of the iteration; α represents a random number between (0,1).

6. The shale gas development index prediction method based on engineering parameters according to claim 4, characterized in that: Generate new individuals to replace the M optimal individuals with the smallest fitness value, and obtain the optimal population after secondary update, including: Determine the fitness value of each optimizing individual in the optimizing population after the initial update, and determine the M first target optimizing individuals with the smallest fitness value; For any first-target optimizing individual, randomly select the second-target optimizing individual from the optimizing population, and perform a cross operation on the first-target optimizing individual and the second-target optimizing individual to obtain two third-target optimizing individuals; Determine whether there is a third target optimization individual with a fitness value greater than the first target optimization individual. If so, replace the first target optimization individual with the third target optimization individual with the largest fitness value to obtain a new first target optimization individual. Otherwise, perform a mutation operation. For any first-target optimizing individual, the parameters in the first-target optimizing individual are mutated to obtain a new first-target optimizing individual.

7. The shale gas development index prediction method based on engineering parameters according to claim 4, characterized in that: After each optimizing individual is updated according to the best individual, the method further includes: updating the optimizing frequency corresponding to the optimizing individual, and the updating is: in, represents the optimization frequency during the t-th training, represents the initial optimization frequency of the kth optimization individual, and γ represents the frequency enhancement coefficient.

Images