A pipe network load control method and system based on deep learning

Abstract

The application relates to the technical field of natural gas pipe networks, and discloses a pipe network load control method and system based on deep learning, which acquires position information of a to-be-controlled pipe network, predeploys detection equipment in the to-be-controlled pipe network based on preset conditions, and collects initial real-time data of the to-be-controlled pipe network according to the detection equipment; the initial real-time data is subjected to data processing, and target real-time data is obtained; historical data corresponding to the target real-time data is extracted, and a load prediction model is constructed according to the historical data and the target real-time data; the target real-time data is input into the load prediction model, a load prediction value of the to-be-controlled pipe network is obtained, and the load of the to-be-controlled pipe network is managed and controlled based on the load prediction value; the application can realize intelligent control of the load of the pipe network; the pipe network load control method based on deep learning can improve the operation efficiency of the pipe network and the accuracy of load control, and helps operation personnel to realize intelligent management and optimized control of the pipe network.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of natural gas pipeline network control, in particular to a pipeline network load control method and system based on deep learning. BACKGROUND

[0002] Natural gas pipeline network refers to the pipeline that transports natural gas (including associated gas produced in oil fields) from the mining site or processing plant to the city gas distribution center or industrial enterprise user, also known as gas pipeline. Using natural gas pipeline to transport natural gas is the main way of transporting natural gas on land. In the total length of the world pipeline, natural gas pipeline accounts for about half. Natural gas pipeline network has the advantages of low transportation cost, small land occupation, fast construction, large oil and gas transportation capacity, high safety performance, small transportation loss, small leakage risk, small environmental pollution, small influence of bad weather, small equipment maintenance amount, easy management, and easy realization of remote centralized monitoring. However, with the large-scale use of natural gas pipeline network, how to further improve the accuracy of natural gas pipeline network load control has become a technical problem to be solved.

[0003] The current natural gas pipeline network load control method has the following shortcomings:

[0004] 1) Lack of intelligence and adaptability: the existing natural gas pipeline network load control method lacks intelligence and adaptability in load control, and cannot fully utilize advanced artificial intelligence, big data and Internet of Things technology. This makes it difficult for pipeline network control to adapt to complex load changes and environmental changes, and unable to achieve efficient, flexible and intelligent load control.

[0005] 2) Lack of global optimization: the existing natural gas pipeline network load control method is usually based on local pipeline network load adjustment, lacking global optimization of the entire pipeline network system. This may lead to local load overload or imbalance, affecting the stability and safety of the pipeline network.

[0006] 3) Limited control accuracy: the existing natural gas pipeline network load control method may have certain limitations in the accuracy of load control. Due to the existence of many complex factors and uncertainties in the pipeline network, as well as the limitations of related measurement and control equipment, it may not be possible to achieve accurate control and prediction of load.

[0007] 4) High cost and complexity: some existing natural gas pipeline network load control methods may require a large amount of cost and complex equipment to achieve pipeline network load control. This may cause economic burden to operators and managers, especially for small-scale pipeline networks and resource-limited areas.

[0008] Therefore, how to provide an intelligent pipeline network load control method and system based on deep learning is a technical problem to be solved at present. SUMMARY

[0009] The embodiment of the present application provides a kind of based on deep learning's pipe network load control method and system, to solve the technical problem that natural gas pipe network load cannot be intelligently controlled in prior art, cannot improve pipe network operation efficiency and natural gas pipe network load control accuracy.

[0010] To achieve the above object, the present application provides a kind of based on deep learning's pipe network load control method, the method comprises:

[0011] The position information of the pipe network to be controlled is acquired, the detection equipment is pre-deployed in the pipe network to be controlled based on the preset condition, and the initial real-time data of the pipe network to be controlled is collected according to the pre-deployed detection equipment;

[0012] The initial real-time data is processed according to the preset rule, and the target real-time data is obtained, wherein the data processing includes data cleaning, data exception processing and data normalization processing;

[0013] The historical data corresponding to the target real-time data is extracted, and the load prediction model is constructed according to the historical data and the target real-time data;

[0014] The target real-time data is input into the load prediction model, the load prediction value of the pipe network to be controlled is obtained, and the load of the pipe network to be controlled is managed and controlled based on the load prediction value.

[0015] In one embodiment, when the initial real-time data is cleaned according to the preset rule, it comprises:

[0016] The attribute information of all initial real-time data is acquired, all initial real-time data is classified according to the attribute information, and a plurality of data sets are obtained;

[0017] A plurality of judgment data are extracted from the initial real-time data of each data set, respectively;

[0018] A plurality of preset data cleaning rules are acquired, and the judgment data are cleaned based on a plurality of preset data cleaning rules;

[0019] The cleaning success rate of judgment data in each data set is calculated, respectively;

[0020] When the cleaning success rate is greater than or equal to the preset cleaning success rate, the corresponding preset data cleaning rule is determined as the data cleaning rule of the corresponding data set;

[0021] When the cleaning success rate is less than the preset cleaning success rate, the data cleaning rule of the corresponding data set is re-determined;

[0022] Based on the determined data cleaning rules, the data set is cleaned, and a plurality of data cleaning sets are obtained.

[0023] In one embodiment, when the initial real-time data is normalized according to the preset rule, it includes:

[0024] Determine whether the initial real-time data in each data cleaning set conforms to the normal distribution, if so, calculate the average value and the standard value of the initial real-time data in each data cleaning set;

[0025] According to the average value and the standard value of the initial real-time data in each data cleaning set, the initial real-time data is normalized;

[0026] According to the following formula, the initial real-time data in each data cleaning set is normalized:

[0027] , i = 1, 2, 3,..., n;

[0028] ;

[0029] ;

[0030] Wherein, Ax1 represents the normalized initial real-time data, the value is between 0-1; M is the standard deviation of the initial real-time data in the data cleaning set; is the average value of the initial real-time data in the data cleaning set; Ai is the original initial real-time data; n is the number of initial real-time data in the data cleaning set.

[0031] In one embodiment, when the historical data corresponding to the target real-time data is extracted, and the load prediction model is constructed according to the historical data and the target real-time data, it includes:

[0032] According to the historical data and the target real-time data, a data set is constructed;

[0033] According to the preset proportion, the training subset and the test subset are obtained by sampling the data set;

[0034] The preselected network model is obtained, and the network model is iteratively trained according to the training subset, the model after iterative training is evaluated according to the test subset, and the load prediction model is obtained, wherein,

[0035] If the evaluation value of the current iterative training network model is less than the evaluation value of the previous iterative training network model, the amplitude of the change of the network model in the gradient direction is reduced, and the iterative training is continued until the preset iteration number is reached;

[0036] If the evaluation value of the network model after the current iteration training is greater than or equal to the evaluation value of the network model after the previous iteration training, the iteration training is stopped.

[0037] In one of the embodiments, when managing and controlling the load of the pipe network to be controlled based on the load prediction value, the method comprises:

[0038] Obtaining the pipe network pressure P, the natural gas flow V and the valve opening K of the pipe network to be controlled;

[0039] According to the load prediction value W of the pipe network to be controlled, the pipe network pressure P, the natural gas flow V and the valve opening K are adjusted and controlled to obtain the pipe network adjustment pressure, the natural gas adjustment flow and the valve adjustment opening;

[0040] The load prediction value matrix B of the pipe network to be controlled is preset, and B is set as (B1, B2, B3, B4), wherein B1 is a first preset load prediction value, B2 is a second preset load prediction value, B3 is a third preset load prediction value, and B4 is a fourth preset load prediction value, and B1

[0041] The pipe network pressure correction coefficient matrix h of the pipe network to be controlled is preset, and h is set as (h1, h2, h3, h4, h5), wherein h1 is a first preset pipe network pressure correction coefficient, h2 is a second preset pipe network pressure correction coefficient, h3 is a third preset pipe network pressure correction coefficient, h4 is a fourth preset pipe network pressure correction coefficient, and h5 is a fifth preset pipe network pressure correction coefficient, and 0.8

[0042] The natural gas flow correction coefficient matrix y of the pipe network to be controlled is preset, and y is set as (y1, y2, y3, y4, y5), wherein y1 is a first preset natural gas flow correction coefficient, y2 is a second preset natural gas flow correction coefficient, y3 is a third preset natural gas flow correction coefficient, y4 is a fourth preset natural gas flow correction coefficient, and y5 is a fifth preset natural gas flow correction coefficient, and 0.8

[0043] The valve opening correction coefficient matrix e of the pipe network to be controlled is preset, and e is set as (e1, e2, e3, e4, e5), wherein e1 is a first preset valve opening correction coefficient, e2 is a second preset valve opening correction coefficient, e3 is a third preset valve opening correction coefficient, e4 is a fourth preset valve opening correction coefficient, and e5 is a fifth preset valve opening correction coefficient, and 0.8

[0044] According to the load prediction value W of the pipe network to be controlled and each preset load prediction value, the pipe network pressure P, the natural gas flow V and the valve opening K are adjusted and controlled respectively:

[0045] When W<B1, the first preset pipe network pressure correction coefficient h1 is selected to adjust the pipe network pressure P, the adjusted pipe network pressure being P*h1, the first preset natural gas flow correction coefficient y1 is selected to adjust the natural gas flow V, the adjusted natural gas flow being V*y1, and the first preset valve opening correction coefficient e1 is selected to adjust the valve opening K, the adjusted valve opening being K*e1.

[0046] When B1≤W<B2, the second preset pipe network pressure correction coefficient h2 is selected to adjust the pipe network pressure P, the adjusted pipe network pressure being P*h2, the second preset natural gas flow correction coefficient y2 is selected to adjust the natural gas flow V, the adjusted natural gas flow being V*y2, and the second preset valve opening correction coefficient e2 is selected to adjust the valve opening K, the adjusted valve opening being K*e2.

[0047] When B2≤W<B3, the third preset pipe network pressure correction coefficient h3 is selected to adjust the pipe network pressure P, the adjusted pipe network pressure being P*h3, the third preset natural gas flow correction coefficient y3 is selected to adjust the natural gas flow V, the adjusted natural gas flow being V*y3, and the third preset valve opening correction coefficient e3 is selected to adjust the valve opening K, the adjusted valve opening being K*e3.

[0048] When B3≤W<B4, the fourth preset pipe network pressure correction coefficient h4 is selected to adjust the pipe network pressure P, the adjusted pipe network pressure being P*h4, the fourth preset natural gas flow correction coefficient y4 is selected to adjust the natural gas flow V, the adjusted natural gas flow being V*y4, and the fourth preset valve opening correction coefficient e4 is selected to adjust the valve opening K, the adjusted valve opening being K*e4.

[0049] When B4≤W, the fifth preset pipe network pressure correction coefficient h5 is selected to adjust the pipe network pressure P, the adjusted pipe network pressure being P*h5, the fifth preset natural gas flow correction coefficient y5 is selected to adjust the natural gas flow V, the adjusted natural gas flow being V*y5, and the fifth preset valve opening correction coefficient e5 is selected to adjust the valve opening K, the adjusted valve opening being K*e5.

[0050] To achieve the above-mentioned purpose, the application provides a pipe network load control system based on deep learning, which comprises:

[0051] a data collection module, configured to acquire position information of a to-be-controlled pipe network, predeploy detection devices in the to-be-controlled pipe network based on preset conditions, and collect initial real-time data of the to-be-controlled pipe network according to the predeployed detection devices;

[0052] a data processing module, configured to perform data processing on the initial real-time data according to preset rules, and obtain target real-time data, wherein the data processing includes data cleaning, data exception processing and data normalization processing;

[0053] a model construction module, configured to extract historical data corresponding to the target real-time data, and construct a load prediction model according to the historical data and the target real-time data;

[0054] a load control module, configured to input the target real-time data into the load prediction model, obtain a load prediction value of the to-be-controlled pipe network, and manage and control the load of the to-be-controlled pipe network based on the load prediction value.

[0055] In one of the embodiments, the data processing module is specifically configured to:

[0056] The data processing module is configured to acquire attribute information of all the initial real-time data, classify all the initial real-time data according to the attribute information, and obtain a plurality of data sets;

[0057] The data processing module is configured to extract a plurality of judgment data from the initial real-time data of each data set, respectively;

[0058] The data processing module is configured to acquire a plurality of preset data cleaning rules stored in advance, and perform data cleaning on the judgment data based on the plurality of preset data cleaning rules;

[0059] The data processing module is configured to calculate a cleaning success rate of the judgment data in each data set, respectively;

[0060] The data processing module is configured to determine, when the cleaning success rate is greater than or equal to a preset cleaning success rate, a corresponding preset data cleaning rule as a data cleaning rule of a corresponding data set;

[0061] The data processing module is configured to determine, when the cleaning success rate is less than the preset cleaning success rate, a data cleaning rule of the corresponding data set again;

[0062] The data processing module is configured to perform data cleaning on the data set based on the determined data cleaning rule, and obtain a plurality of data cleaning sets.

[0063] In one of the embodiments, the data processing module is specifically configured to:

[0064] The data processing module is configured to determine whether the initial real-time data in each data cleaning set conforms to a normal distribution, and if so, calculate a mean value and a standard value of the initial real-time data in each data cleaning set;

[0065] The data processing module is configured to normalize the initial real-time data in each data cleaning set according to the mean value and the standard value of the initial real-time data in each data cleaning set;

[0066] The data processing module is configured to normalize the initial real-time data in each data cleaning set according to the following formula:

[0067] , i = 1, 2, 3,..., n;

[0068] ;

[0069] ;

[0070] wherein, Ax1 represents the initial real-time data after normalization, the value of which is between 0 and 1; M is a standard deviation of the initial real-time data in the data cleaning set, is a mean value of the initial real-time data in the data cleaning set; Ai is the original initial real-time data; and n is a quantity of the initial real-time data in the data cleaning set.

[0071] In one of the embodiments, the model construction module is specifically configured to:

[0072] The model construction module is configured to construct a data set according to the historical data and the target real-time data;

[0073] The model construction module is configured to sample the data set according to a preset proportion to obtain a training subset and a test subset;

[0074] The model construction module is configured to obtain a preselected network model, and iteratively train the network model according to the training subset, evaluate the model after iterative training according to the test subset, and obtain the load prediction model, wherein,

[0075] The model construction module is configured to, if an evaluation value of the network model after current iterative training is less than an evaluation value of the network model after previous iterative training, reduce a magnitude of change of the network model in a gradient direction, and continue the iterative training until a preset number of iterations is reached.

[0076] The model construction module is configured to, if the evaluation value of the network model after current iterative training is greater than or equal to the evaluation value of the network model after previous iterative training, stop the iterative training.

[0077] In one of the embodiments, the load control module is specifically used for:

[0078] The load control module is used to obtain the pipe network pressure P, the natural gas flow V and the valve opening K of the pipe network to be controlled;

[0079] The load control module is used to adjust and control the pipe network pressure P, the natural gas flow V and the valve opening K according to the load prediction value W of the pipe network to be controlled, respectively, to obtain the pipe network adjustment pressure, the natural gas adjustment flow and the valve adjustment opening;

[0080] The load control module is used to preset the load prediction value matrix B of the pipe network to be controlled, and set B as (B1, B2, B3, B4), wherein B1 is the first preset load prediction value, B2 is the second preset load prediction value, B3 is the third preset load prediction value, and B4 is the fourth preset load prediction value, and B1 < B2 < B3 < B4;

[0081] The load control module is used to preset the pipe network pressure correction coefficient matrix h of the pipe network to be controlled, and set h as (h1, h2, h3, h4, h5), wherein h1 is the first preset pipe network pressure correction coefficient, h2 is the second preset pipe network pressure correction coefficient, h3 is the third preset pipe network pressure correction coefficient, h4 is the fourth preset pipe network pressure correction coefficient, and h5 is the fifth preset pipe network pressure correction coefficient, and 0.8 < h1 < h2 < h3 < h4 < h5 < 1.2;

[0082] The load control module is used to preset the natural gas flow correction coefficient matrix y of the pipe network to be controlled, and set y as (y1, y2, y3, y4, y5), wherein y1 is the first preset natural gas flow correction coefficient, y2 is the second preset natural gas flow correction coefficient, y3 is the third preset natural gas flow correction coefficient, y4 is the fourth preset natural gas flow correction coefficient, and y5 is the fifth preset natural gas flow correction coefficient, and 0.8 < y1 < y2 < y3 < y4 < y5 < 1.2;

[0083] The load control module is used to preset the valve opening correction coefficient matrix e of the pipe network to be controlled, and set e as (e1, e2, e3, e4, e5), wherein e1 is the first preset valve opening correction coefficient, e2 is the second preset valve opening correction coefficient, e3 is the third preset valve opening correction coefficient, e4 is the fourth preset valve opening correction coefficient, and e5 is the fifth preset valve opening correction coefficient, and 0.8 < e1 < e2 < e3 < e4 < h5 < 1.2;

[0084] The load control module is used to adjust and control the pipe network pressure P, the natural gas flow V and the valve opening K according to the load prediction value W of the pipe network to be controlled and each preset load prediction value, respectively:

[0085] When W

[0086] When B1

[0087] When B2

[0088] When B3

[0089] When B4

[0090] The application provides a pipe network load control method and system based on deep learning, which has the following beneficial effects compared with the prior art:

[0091] The application discloses a kind of based on deep learning's pipe network load control method and system, obtain the position information of the pipe network to be controlled, based on preset condition in the pipe network to be controlled in advance deployment detection equipment, and according to detection equipment acquisition initial real-time data of the pipe network to be controlled;Initial real-time data is processed, and target real-time data is obtained, historical data corresponding to target real-time data is extracted, and load prediction model is constructed according to historical data and target real-time data;Target real-time data is input into load prediction model, and the load prediction value of the pipe network to be controlled is obtained, and based on load prediction value, the load of the pipe network to be controlled is managed and controlled, the present application can realize the intelligent control of pipe network load, based on deep learning's pipe network load control method can improve the operation efficiency of pipe network and the accuracy of load control, help operator realizes the intelligent management and optimization control of pipe network. BRIEF DESCRIPTION OF DRAWINGS

[0092] Figure 1 The flow diagram of the pipe network load control method based on deep learning in the embodiment of the application is shown.

[0093] Figure 2 The structure diagram of the pipe network load control system based on deep learning in the embodiment of the application is shown. DETAILED DESCRIPTION

[0094] The specific embodiments of the present application will be further described in detail below in combination with the drawings and embodiments. The following embodiments are used to illustrate the present application, but not to limit the scope of the present application.

[0095] In the description of the present application, it should be understood that the terms "center", "upper", "lower", "front", "back", "left", "right", "vertical", "horizontal", "top", "bottom", "inner", "outer" and the like indicate the orientation or positional relationship based on the orientation or positional relationship shown in the drawings, and are only for the convenience of describing the present application and simplifying the description, and do not indicate or imply that the device or element referred to must have a particular orientation, be constructed and operated in a particular orientation, and therefore cannot be understood as a limitation on the present application.

[0096] The terms "first", "second" are only for descriptive purposes, and cannot be understood as indicating or implying relative importance or implicitly indicating the number of the technical features indicated. Therefore, the features defined with "first", "second" can explicitly or implicitly include one or more features. In the description of the present application, unless otherwise specified, the meaning of "multiple" is two or more.

[0097] In the description of the present application, it should be noted that unless otherwise explicitly specified and limited, the terms "mounting", "connection", "connecting" should be understood in a broad sense, for example, can be fixed connection, can also be detachable connection, or integral connection; can be mechanical connection, can also be electrical connection; can be direct connection, can also be indirect connection through an intermediate medium, can be internal communication of two elements. For those skilled in the art, the specific meaning of the above terms in the present application can be understood according to the specific circumstances.

[0098] The following is a description of the preferred embodiments of the present application in conjunction with the accompanying drawings.

[0099] As Figure 1 shown, the embodiments of the present application disclose a pipe network load control method based on deep learning, the method comprising:

[0100] S110: acquiring position information of a to-be-controlled pipe network, pre-deploying a detection device in the to-be-controlled pipe network based on a preset condition, and collecting initial real-time data of the to-be-controlled pipe network according to the pre-deployed detection device;

[0101] In the present embodiment, the preset condition refers to the type of data to be detected, such as pipe network pressure, pipe network temperature, pipe network flow, etc.

[0102] In the present embodiment, the detection device corresponds to the type of data to be detected, such as pressure detection device, temperature detection device, flow detection device, etc.

[0103] S120: performing data processing on the initial real-time data according to a preset rule, and obtaining target real-time data, wherein the data processing includes data cleaning, data anomaly processing and data normalization processing;

[0104] In some embodiments of the present application, when the initial real-time data is cleaned according to the preset rule, it includes:

[0105] Acquiring attribute information of all initial real-time data, classifying all initial real-time data according to the attribute information, and obtaining a plurality of data sets;

[0106] Respectively extracting a plurality of judgment data from the initial real-time data of each data set;

[0107] Obtaining a plurality of preset data cleaning rules pre-stored, and cleaning the judgment data based on the plurality of preset data cleaning rules;

[0108] Respectively calculating the cleaning success rate of the judgment data in each data set;

[0109] When the cleaning success rate is greater than or equal to a preset cleaning success rate, a corresponding preset data cleaning rule is determined as a data cleaning rule of the corresponding data set;

[0110] When the cleaning success rate is less than the preset cleaning success rate, the data cleaning rule of the corresponding data set is re-determined;

[0111] The data set is cleaned based on the determined data cleaning rule, and a plurality of data cleaning sets are obtained.

[0112] In this embodiment, the attribute information of the initial real-time data refers to the data type of the initial real-time data, such as pipe network pressure, pipe network temperature, pipe network flow, etc.

[0113] In this embodiment, each data set includes a plurality of initial real-time data.

[0114] In this embodiment, the judgment data is randomly extracted from the data set, and the number of data that can be extracted can be determined according to the number of data sets, such as 10 or 15.

[0115] In this embodiment, when the judgment data is cleaned, the obvious error data in the judgment data is mainly cleaned. For example, when the temperature data set is cleaned, the pressure data appears, and the pressure data is an obvious error data.

[0116] In this embodiment, when the cleaning success rate of the judgment data in each data set is calculated, if the judgment data is 10, and there is still an obvious error data that has not been successfully cleaned after data cleaning, the cleaning success rate at this time is 90%.

[0117] In this embodiment, the preset cleaning success rate can be set according to actual needs, which is not limited here.

[0118] In this embodiment, each data set corresponds to a data cleaning rule.

[0119] The beneficial effects of the above technical solutions are: by determining the data cleaning rule, the initial real-time data can be effectively cleaned, the accuracy and reliability of the data are ensured, and the foundation for subsequent control of the pipe network load is laid.

[0120] In some embodiments of the present application, when the initial real-time data is processed according to the preset rule, a statistical method or a machine learning algorithm is used to detect and process abnormal values, for example, an outlier detection algorithm (such as a statistical-based Z-score method) is used to identify and process abnormal data.

[0121] In some embodiments of the present application, when the initial real-time data is normalized according to the preset rule, it includes:

[0122] determining whether the initial real-time data in each data cleaning set conforms to a normal distribution, and if so, calculating the mean value and standard value of the initial real-time data in each data cleaning set;

[0123] normalizing the initial real-time data in each data cleaning set according to the mean value and standard value of the initial real-time data in each data cleaning set;

[0124] normalizing the initial real-time data in each data cleaning set according to the following formula:

[0125] , i = 1, 2, 3,..., n;

[0126] ;

[0127] ;

[0128] wherein, Ax1 represents the normalized initial real-time data, the value of which is between 0 and 1; M is the standard deviation of the initial real-time data in the data cleaning set; is the mean value of the initial real-time data in the data cleaning set; Ai is the original initial real-time data; and n is the number of initial real-time data in the data cleaning set.

[0129] The beneficial effects of the above technical solution are that the original initial real-time data is normalized to between 0 and 1, thereby improving the data analysis efficiency, making the data of different characteristics have the same dimension, and helping to improve the training and prediction effect of the load prediction model.

[0130] S130: extracting historical data corresponding to the target real-time data, and constructing a load prediction model according to the historical data and the target real-time data;

[0131] In some embodiments of the present application, when extracting historical data corresponding to the target real-time data and constructing a load prediction model according to the historical data and the target real-time data, it includes:

[0132] constructing a data set according to the historical data and the target real-time data;

[0133] sampling the data set according to a preset proportion to obtain a training subset and a test subset;

[0134] obtaining a preselected network model, and iteratively training the network model according to the training subset, evaluating the model after iterative training according to the test subset, and obtaining the load prediction model, wherein,

[0135] If the evaluation value of the network model after the current iteration training is less than the evaluation value of the network model after the previous iteration training, the magnitude of the change of the network model in the gradient direction is reduced, and the iteration training is continued until a preset iteration number is reached.

[0136] If the evaluation value of the network model after the current iteration training is greater than or equal to the evaluation value of the network model after the previous iteration training, the iteration training is stopped.

[0137] In this embodiment, the preset proportion can be set according to actual conditions, the training subset is used for training the network model, and the test subset is used for testing and evaluating the model after iteration training.

[0138] In this embodiment, the preselected network model can be a recurrent neural network (RNN) or a long short-term memory network (LSTM).

[0139] In this embodiment, the training subset contains historical data, and the test subset contains real-time data and future data.

[0140] In this embodiment, the preset iteration number can be set to 10 or 15, which is not limited here.

[0141] In this embodiment, reducing the magnitude of the change of the network model in the gradient direction means accelerating and stabilizing the convergence of the algorithm model by reasonable step setting.

[0142] The beneficial effects of the above technical solutions are: the present application can accurately predict the load change trend of the pipe network in a future period of time according to historical data and target real-time data. This helps the operation personnel to make reasonable decisions and optimize the operation efficiency of the pipe network. At the same time, the present application can train a more general model through iteration training, thereby improving the prediction accuracy and control effect.

[0143] S140: inputting the target real-time data into the load prediction model to obtain a load prediction value of the pipe network to be controlled, and managing and controlling the load of the pipe network to be controlled based on the load prediction value.

[0144] In some embodiments of the present application, when the load of the pipe network to be controlled is managed and controlled based on the load prediction value, it includes:

[0145] obtaining the pipe network pressure P, the natural gas flow V and the valve opening K of the pipe network to be controlled;

[0146] adjusting and controlling the pipe network pressure P, the natural gas flow V and the valve opening K according to the load prediction value W of the pipe network to be controlled, to obtain the pipe network adjustment pressure, the natural gas adjustment flow and the valve adjustment opening;

[0147] A load prediction value matrix B of the pipe network to be controlled is preset, and B is set as (B1, B2, B3, B4), wherein B1 is a first preset load prediction value, B2 is a second preset load prediction value, B3 is a third preset load prediction value, and B4 is a fourth preset load prediction value, and B1

[0148] A pipe network pressure correction coefficient matrix h of the pipe network to be controlled is preset, and h is set as (h1, h2, h3, h4, h5), wherein h1 is a first preset pipe network pressure correction coefficient, h2 is a second preset pipe network pressure correction coefficient, h3 is a third preset pipe network pressure correction coefficient, h4 is a fourth preset pipe network pressure correction coefficient, and h5 is a fifth preset pipe network pressure correction coefficient, and 0.8

[0149] A natural gas flow correction coefficient matrix y of the pipe network to be controlled is preset, and y is set as (y1, y2, y3, y4, y5), wherein y1 is a first preset natural gas flow correction coefficient, y2 is a second preset natural gas flow correction coefficient, y3 is a third preset natural gas flow correction coefficient, y4 is a fourth preset natural gas flow correction coefficient, and y5 is a fifth preset natural gas flow correction coefficient, and 0.8

[0150] A valve opening correction coefficient matrix e of the pipe network to be controlled is preset, and e is set as (e1, e2, e3, e4, e5), wherein e1 is a first preset valve opening correction coefficient, e2 is a second preset valve opening correction coefficient, e3 is a third preset valve opening correction coefficient, e4 is a fourth preset valve opening correction coefficient, and e5 is a fifth preset valve opening correction coefficient, and 0.8

[0151] The pipe network pressure P, the natural gas flow V, and the valve opening K are adjusted and controlled according to the load prediction value W of the pipe network to be controlled and each preset load prediction value, respectively:

[0152] When W

[0153] When B1≤W<B2, the second preset pipe network pressure correction coefficient h2 is selected to adjust the pipe network pressure P, the adjusted pipe network pressure being P*h2, the second preset natural gas flow correction coefficient y2 is selected to adjust the natural gas flow V, the adjusted natural gas flow being V*y2, and the second preset valve opening correction coefficient e2 is selected to adjust the valve opening K, the adjusted valve opening being K*e2.

[0154] When B2≤W<B3, the third preset pipe network pressure correction coefficient h3 is selected to adjust the pipe network pressure P, the adjusted pipe network pressure being P*h3, the third preset natural gas flow correction coefficient y3 is selected to adjust the natural gas flow V, the adjusted natural gas flow being V*y3, and the third preset valve opening correction coefficient e3 is selected to adjust the valve opening K, the adjusted valve opening being K*e3.

[0155] When B3≤W<B4, the fourth preset pipe network pressure correction coefficient h4 is selected to adjust the pipe network pressure P, the adjusted pipe network pressure being P*h4, the fourth preset natural gas flow correction coefficient y4 is selected to adjust the natural gas flow V, the adjusted natural gas flow being V*y4, and the fourth preset valve opening correction coefficient e4 is selected to adjust the valve opening K, the adjusted valve opening being K*e4.

[0156] When B4≤W, the fifth preset pipe network pressure correction coefficient h5 is selected to adjust the pipe network pressure P, the adjusted pipe network pressure being P*h5, the fifth preset natural gas flow correction coefficient y5 is selected to adjust the natural gas flow V, the adjusted natural gas flow being V*y5, and the fifth preset valve opening correction coefficient e5 is selected to adjust the valve opening K, the adjusted valve opening being K*e5.

[0157] The beneficial effects of the above technical solution are: according to the load prediction value W of the pipe network to be controlled and each preset load prediction value, the pipe network pressure P, the natural gas flow V and the valve opening K are adjusted and controlled, the optimization control of the pipe network is realized, the load control accuracy is improved, and the dynamic adjustment of the pipe network pressure, the natural gas flow and the valve opening is realized.

[0158] In order to further illustrate the technical idea of the present application, the technical solution of the present application will be described in combination with a specific application scenario.

[0159] Correspondingly, as shown in Figure 2 The present application also provides a pipe network load control system based on deep learning, which comprises:

[0160] The data collection module is configured to acquire position information of a to-be-controlled pipe network, pre-deploy a detection device in the to-be-controlled pipe network based on a preset condition, and collect initial real-time data of the to-be-controlled pipe network according to the pre-deployed detection device.

[0161] The data processing module is configured to perform data processing on the initial real-time data according to a preset rule, and obtain target real-time data, wherein the data processing includes data cleaning, data exception processing and data normalization processing.

[0162] The model construction module is configured to extract historical data corresponding to the target real-time data, and construct a load prediction model according to the historical data and the target real-time data.

[0163] The load control module is configured to input the target real-time data into the load prediction model, obtain a load prediction value of the to-be-controlled pipe network, and manage and control the load of the to-be-controlled pipe network based on the load prediction value.

[0164] In some embodiments of the present application, the data processing module is specifically configured to:

[0165] The data processing module is configured to acquire attribute information of all initial real-time data, classify all initial real-time data according to the attribute information, and obtain a plurality of data sets.

[0166] The data processing module is configured to extract a plurality of judgment data from the initial real-time data of each data set, respectively.

[0167] The data processing module is configured to acquire a plurality of preset data cleaning rules pre-stored, and perform data cleaning on the judgment data based on the plurality of preset data cleaning rules.

[0168] The data processing module is configured to calculate a cleaning success rate of the judgment data in each data set, respectively.

[0169] The data processing module is configured to determine the corresponding preset data cleaning rule as a data cleaning rule of the corresponding data set when the cleaning success rate is greater than or equal to a preset cleaning success rate.

[0170] The data processing module is configured to re-determine the data cleaning rule of the corresponding data set when the cleaning success rate is less than the preset cleaning success rate.

[0171] The data processing module is configured to perform data cleaning on the data set based on the determined data cleaning rule, and obtain a plurality of data cleaning sets.

[0172] In some embodiments of the present application, the data processing module is specifically configured to:

[0173] The data processing module is configured to determine whether the initial real-time data in each data cleaning set conforms to a normal distribution, and if so, calculate a mean value and a standard value of the initial real-time data in each data cleaning set;

[0174] The data processing module is configured to normalize the initial real-time data in each data cleaning set according to the mean value and the standard value of the initial real-time data in each data cleaning set;

[0175] The data processing module is configured to normalize the initial real-time data in each data cleaning set according to the following formula:

[0176] , i = 1, 2, 3,..., n;

[0177] ;

[0178] ;

[0179] wherein, Ax1 represents the initial real-time data after normalization, the value of which is between 0 and 1; M is a standard deviation of the initial real-time data in the data cleaning set; is a mean value of the initial real-time data in the data cleaning set; Ai is the original initial real-time data; and n is a quantity of the initial real-time data in the data cleaning set.

[0180] In some embodiments of the present application, the model construction module is specifically configured to:

[0181] The model construction module is configured to construct a data set according to the historical data and the target real-time data;

[0182] The model construction module is configured to sample the data set according to a preset proportion to obtain a training subset and a test subset;

[0183] The model construction module is configured to obtain a preselected network model, and iteratively train the network model according to the training subset, evaluate the model after iterative training according to the test subset, and obtain the load prediction model, wherein,

[0184] The model construction module is configured to, if an evaluation value of the network model after current iterative training is less than an evaluation value of the network model after previous iterative training, reduce a magnitude of change of the network model in a gradient direction, and continue iterative training until a preset number of iterations is reached.

[0185] The model construction module is configured to, if the evaluation value of the network model after current iterative training is greater than or equal to the evaluation value of the network model after previous iterative training, stop iterative training.

[0186] In some embodiments of the present application, the load control module is specifically used for:

[0187] The load control module is used to obtain the pipe network pressure P, the natural gas flow V and the valve opening K of the pipe network to be controlled;

[0188] The load control module is used to adjust and control the pipe network pressure P, the natural gas flow V and the valve opening K according to the load prediction value W of the pipe network to be controlled, to obtain the pipe network adjustment pressure, the natural gas adjustment flow and the valve adjustment opening;

[0189] The load control module is used to preset a load prediction value matrix B of the pipe network to be controlled, and set B as (B1, B2, B3, B4), wherein B1 is a first preset load prediction value, B2 is a second preset load prediction value, B3 is a third preset load prediction value, and B4 is a fourth preset load prediction value, and B1

[0190] The load control module is used to preset a pipe network pressure correction coefficient matrix h of the pipe network to be controlled, and set h as (h1, h2, h3, h4, h5), wherein h1 is a first preset pipe network pressure correction coefficient, h2 is a second preset pipe network pressure correction coefficient, h3 is a third preset pipe network pressure correction coefficient, h4 is a fourth preset pipe network pressure correction coefficient, and h5 is a fifth preset pipe network pressure correction coefficient, and 0.8

[0191] The load control module is used to preset a natural gas flow correction coefficient matrix y of the pipe network to be controlled, and set y as (y1, y2, y3, y4, y5), wherein y1 is a first preset natural gas flow correction coefficient, y2 is a second preset natural gas flow correction coefficient, y3 is a third preset natural gas flow correction coefficient, y4 is a fourth preset natural gas flow correction coefficient, and y5 is a fifth preset natural gas flow correction coefficient, and 0.8

[0192] The load control module is used to preset a valve opening correction coefficient matrix e of the pipe network to be controlled, and set e as (e1, e2, e3, e4, e5), wherein e1 is a first preset valve opening correction coefficient, e2 is a second preset valve opening correction coefficient, e3 is a third preset valve opening correction coefficient, e4 is a fourth preset valve opening correction coefficient, and e5 is a fifth preset valve opening correction coefficient, and 0.8

[0193] The load control module is used to adjust and control the pipe network pressure P, the natural gas flow V and the valve opening K according to the load prediction value W of the pipe network to be controlled and each preset load prediction value, respectively:

[0194] When W < B1, the first preset pipe network pressure correction coefficient h1 is selected to adjust the pipe network pressure P, the adjusted pipe network pressure is P*h1, the first preset natural gas flow correction coefficient y1 is selected to adjust the natural gas flow V, the adjusted natural gas flow is V*y1, and the first preset valve opening correction coefficient e1 is selected to adjust the valve opening K, the adjusted valve opening is K*e1.

[0195] When B1≤W < B2, the second preset pipe network pressure correction coefficient h2 is selected to adjust the pipe network pressure P, the adjusted pipe network pressure is P*h2, the second preset natural gas flow correction coefficient y2 is selected to adjust the natural gas flow V, the adjusted natural gas flow is V*y2, and the second preset valve opening correction coefficient e2 is selected to adjust the valve opening K, the adjusted valve opening is K*e2.

[0196] When B2≤W < B3, the third preset pipe network pressure correction coefficient h3 is selected to adjust the pipe network pressure P, the adjusted pipe network pressure is P*h3, the third preset natural gas flow correction coefficient y3 is selected to adjust the natural gas flow V, the adjusted natural gas flow is V*y3, and the third preset valve opening correction coefficient e3 is selected to adjust the valve opening K, the adjusted valve opening is K*e3.

[0197] When B3≤W < B4, the fourth preset pipe network pressure correction coefficient h4 is selected to adjust the pipe network pressure P, the adjusted pipe network pressure is P*h4, the fourth preset natural gas flow correction coefficient y4 is selected to adjust the natural gas flow V, the adjusted natural gas flow is V*y4, and the fourth preset valve opening correction coefficient e4 is selected to adjust the valve opening K, the adjusted valve opening is K*e4.

[0198] When B4≤W, the fifth preset pipe network pressure correction coefficient h5 is selected to adjust the pipe network pressure P, the adjusted pipe network pressure is P*h5, the fifth preset natural gas flow correction coefficient y5 is selected to adjust the natural gas flow V, the adjusted natural gas flow is V*y5, and the fifth preset valve opening correction coefficient e5 is selected to adjust the valve opening K, the adjusted valve opening is K*e5.

[0199] In the description of the above embodiments, specific features, structures, materials or characteristics can be combined in any one or more embodiments or examples in a suitable manner.

[0200] Although the present application has been described with reference to the above embodiments, various modifications can be made to the application and equivalents thereof without departing from the scope of the application. In particular, features of the disclosed embodiments can be used in any combination without departing from the scope of the application, and the description of the various embodiments does not imply that the features are not combinable unless the context clearly dictates otherwise. Thus, the present application is not limited to the specific embodiments disclosed herein but encompasses all alternatives falling within the scope of the claims.

[0201] It is to be understood that the above description is intended to be illustrative and not restrictive. Many other embodiments will be apparent to those of skill in the art upon reading the above description. The scope of the application should, therefore, be determined not with reference to the above description, but should instead be determined with reference to the appended claims, along with their full scope of equivalents.

Claims

1. A pipe network load control method based on deep learning, characterized by, The method comprises: obtaining position information of a to-be-controlled pipe network, pre-deploying a detection device in the to-be-controlled pipe network based on a preset condition, and collecting initial real-time data of the to-be-controlled pipe network according to the pre-deployed detection device; performing data processing on the initial real-time data according to a preset rule to obtain target real-time data, wherein the data processing comprises data cleaning, data anomaly processing and data normalization processing; extracting historical data corresponding to the target real-time data, and constructing a load prediction model according to the historical data and the target real-time data; inputting the target real-time data into the load prediction model to obtain a load prediction value of the to-be-controlled pipe network, and managing and controlling the load of the to-be-controlled pipe network based on the load prediction value; when performing data cleaning on the initial real-time data according to a preset rule, comprising: obtaining attribute information of all initial real-time data, classifying all initial real-time data according to the attribute information to obtain a plurality of data sets; extracting a plurality of judgment data from the initial real-time data of each data set respectively; obtaining a plurality of preset data cleaning rules stored in advance, and cleaning the judgment data based on the plurality of preset data cleaning rules; calculating the cleaning success rate of the judgment data in each data set respectively; when the cleaning success rate is greater than or equal to a preset cleaning success rate, the corresponding preset data cleaning rule is determined as the data cleaning rule of the corresponding data set; when the cleaning success rate is less than the preset cleaning success rate, the data cleaning rule of the corresponding data set is re-determined; cleaning the data set based on the determined data cleaning rule to obtain a plurality of data cleaning sets; when extracting historical data corresponding to the target real-time data and constructing a load prediction model according to the historical data and the target real-time data, comprising: constructing a data set according to the historical data and the target real-time data; sampling the data set according to a preset proportion to obtain a training subset and a test subset; obtaining a pre-selected network model, and iteratively training the network model according to the training subset, evaluating the iteratively trained model according to the test subset, and obtaining the load prediction model, wherein, if the evaluation value of the current iteratively trained network model is less than the evaluation value of the previously iteratively trained network model, the amplitude of the change of the network model in the gradient direction is reduced, and the iterative training is continued until a preset number of iterations is reached; if the evaluation value of the current iteratively trained network model is greater than or equal to the evaluation value of the previously iteratively trained network model, the iterative training is stopped; when managing and controlling the load of the to-be-controlled pipe network based on the load prediction value, comprising: obtaining the pipe network pressure P, the natural gas flow V and the valve opening K of the to-be-controlled pipe network; adjusting and controlling the pipe network pressure P, the natural gas flow V and the valve opening K according to the load prediction value W of the to-be-controlled pipe network to obtain a pipe network adjustment pressure, a natural gas adjustment flow and a valve adjustment opening. The preset load prediction value matrix B of the pipe network to be controlled is set as (B1, B2, B3, B4), wherein B1 is a first preset load prediction value, B2 is a second preset load prediction value, B3 is a third preset load prediction value, and B4 is a fourth preset load prediction value, and B1 The preset pipe network pressure correction coefficient matrix h of the pipe network to be controlled is set as (h1, h2, h3, h4, h5), wherein h1 is a first preset pipe network pressure correction coefficient, h2 is a second preset pipe network pressure correction coefficient, h3 is a third preset pipe network pressure correction coefficient, h4 is a fourth preset pipe network pressure correction coefficient, and h5 is a fifth preset pipe network pressure correction coefficient, and 0.8 The preset natural gas flow correction coefficient matrix y of the pipe network to be controlled is set as (y1, y2, y3, y4, y5), wherein y1 is a first preset natural gas flow correction coefficient, y2 is a second preset natural gas flow correction coefficient, y3 is a third preset natural gas flow correction coefficient, y4 is a fourth preset natural gas flow correction coefficient, and y5 is a fifth preset natural gas flow correction coefficient, and 0.8 The preset valve opening correction coefficient matrix e of the pipe network to be controlled is set as (e1, e2, e3, e4, e5), wherein e1 is a first preset valve opening correction coefficient, e2 is a second preset valve opening correction coefficient, e3 is a third preset valve opening correction coefficient, e4 is a fourth preset valve opening correction coefficient, and e5 is a fifth preset valve opening correction coefficient, and 0.8 According to the load prediction value W of the pipe network to be controlled and each preset load prediction value, the pipe network pressure P, the natural gas flow V, and the valve opening K are adjusted and controlled: When W When B1 When B1 When B2≤W When B3≤W When B4≤W 2. The deep learning-based pipe network load control method of claim 1, wherein, According to the preset rule, the initial real-time data is normalized, comprising: determining whether the initial real-time data in each data cleaning set conforms to normal distribution, if so, calculating the average value and standard value of the initial real-time data in each data cleaning set; normalizing the initial real-time data according to the average value and standard value of the initial real-time data in each data cleaning set; normalizing the initial real-time data in each data cleaning set according to the following formula: , i = 1, 2, 3,..., n; ； ； Wherein, Ax1 represents the normalized initial real-time data, the value is between 0-1; M is the standard deviation of the initial real-time data in the data cleaning set; is the average value of the initial real-time data in the data cleaning set; Ai is the original initial real-time data; n is the number of the initial real-time data in the data cleaning set.

3. A deep learning-based pipe network load control system, characterized by, The system comprises: a data acquisition module, configured to acquire position information of a to-be-controlled pipe network, pre-deploy a detection device in the to-be-controlled pipe network based on a preset condition, and collect initial real-time data of the to-be-controlled pipe network according to the pre-deployed detection device; a data processing module, configured to perform data processing on the initial real-time data according to a preset rule, and obtain target real-time data, wherein the data processing comprises data cleaning, data anomaly processing and data normalization processing; a model construction module, configured to extract historical data corresponding to the target real-time data, and construct a load prediction model according to the historical data and the target real-time data; a load control module, configured to input the target real-time data into the load prediction model, obtain a load prediction value of the to-be-controlled pipe network, and manage and control the load of the to-be-controlled pipe network based on the load prediction value; The data processing module is specifically configured to: The data processing module is configured to acquire attribute information of all initial real-time data, classify all initial real-time data according to the attribute information, and obtain a plurality of data sets; The data processing module is configured to extract a plurality of judgment data from the initial real-time data of each data set, respectively. The data processing module is configured to acquire a plurality of pre-stored preset data cleaning rules, and clean the judgment data based on the plurality of preset data cleaning rules; The data processing module is configured to calculate a cleaning success rate of the judgment data in each data set respectively; The data processing module is configured to determine the corresponding preset data cleaning rule as the data cleaning rule of the corresponding data set when the cleaning success rate is greater than or equal to a preset cleaning success rate; The data processing module is configured to re-determine the data cleaning rule of the corresponding data set when the cleaning success rate is less than the preset cleaning success rate; The data processing module is configured to clean the data set based on the determined data cleaning rule, and obtain a plurality of data cleaning sets; The model construction module is specifically configured to: The model construction module is configured to construct a data set according to the historical data and the target real-time data; The model construction module is configured to sample the data set according to a preset proportion to obtain a training subset and a test subset; The model construction module is configured to acquire a pre-selected network model, iteratively train the network model according to the training subset, evaluate the iteratively trained model according to the test subset, and obtain the load prediction model, wherein, The model construction module is configured to reduce the amplitude of the change of the network model in the gradient direction if the evaluation value of the current iteratively trained network model is less than the evaluation value of the network model of the previous iteration, and continue the iterative training until a preset number of iterations is reached; The model construction module is configured to stop the iterative training if the evaluation value of the current iteratively trained network model is greater than or equal to the evaluation value of the network model of the previous iteration; The load control module is specifically configured to: The load control module is configured to acquire a pipe network pressure P, a natural gas flow V, and a valve opening K of the pipe network to be controlled; The load control module is configured to adjust and control the pipe network pressure P, the natural gas flow V, and the valve opening K according to the load prediction value W of the pipe network to be controlled, to obtain a pipe network adjustment pressure, a natural gas adjustment flow, and a valve adjustment opening; The load control module is configured to preset a load prediction value matrix B of the pipe network to be controlled, and set B as (B1, B2, B3, B4), wherein B1 is a first preset load prediction value, B2 is a second preset load prediction value, B3 is a third preset load prediction value, and B4 is a fourth preset load prediction value, and B1 The load control module is configured to preset a pipe network pressure correction coefficient matrix h of the pipe network to be controlled, and set h as (h1, h2, h3, h4, h5), wherein h1 is a first preset pipe network pressure correction coefficient, h2 is a second preset pipe network pressure correction coefficient, h3 is a third preset pipe network pressure correction coefficient, h4 is a fourth preset pipe network pressure correction coefficient, and h5 is a fifth preset pipe network pressure correction coefficient, and 0.8 The load control module is configured to preset a natural gas flow correction coefficient matrix y of a pipe network to be controlled, and set y as (y1, y2, y3, y4, y5), wherein y1 is a first preset natural gas flow correction coefficient, y2 is a second preset natural gas flow correction coefficient, y3 is a third preset natural gas flow correction coefficient, y4 is a fourth preset natural gas flow correction coefficient, and y5 is a fifth preset natural gas flow correction coefficient, and 0.8 The load control module is configured to preset a valve opening correction coefficient matrix e of the pipe network to be controlled, and set e as (e1, e2, e3, e4, e5), wherein e1 is a first preset valve opening correction coefficient, e2 is a second preset valve opening correction coefficient, e3 is a third preset valve opening correction coefficient, e4 is a fourth preset valve opening correction coefficient, and e5 is a fifth preset valve opening correction coefficient, and 0.8 The load control module is configured to adjust and control the pipe network pressure P, the natural gas flow V, and the valve opening K according to a load prediction value W of the pipe network to be controlled and each preset load prediction value, respectively. When W When B1 When B2 When B3 When B3 When B4≤W, the fifth preset pipe network pressure correction coefficient h5 is selected to adjust the pipe network pressure P, the adjusted pipe network pressure is P*h5, the fifth preset natural gas flow correction coefficient y5 is selected to adjust the natural gas flow V, the adjusted natural gas flow is V*y5, and the fifth preset valve opening correction coefficient e5 is selected to adjust the valve opening K, and the adjusted valve opening is K*e5.

4. The deep learning-based pipe network load control system of claim 3, wherein, The data processing module is specifically used for: The data processing module is used for judging whether the initial real-time data in each data cleaning set conforms to a normal distribution, and if so, calculating the average value and the standard value of the initial real-time data in each data cleaning set; The data processing module is used for normalizing the initial real-time data in each data cleaning set according to the average value and the standard value of the initial real-time data in each data cleaning set; The data processing module is used for normalizing the initial real-time data in each data cleaning set according to the following formula: i = 1, 2, 3,..., n; ； ； Wherein, Ax1 represents the normalized initial real-time data, the value is between 0-1; M is the standard deviation of the initial real-time data in the data cleaning set; is the average value of the initial real-time data in the data cleaning set; Ai is the original initial real-time data; n is the number of the initial real-time data in the data cleaning set.

Images