Adjusting valve opening control method for natural gas high-pressure long-distance transportation pipeline

Abstract

The invention relates to the field of control and adjustment, in particular to a method for controlling the opening degree of an adjusting valve for a natural gas high-pressure long-distance pipeline, and the method comprises the steps: obtaining a sequential sequence of each factor influencing the opening degree control of the adjusting valve; acquiring a pre-constructed opening control network, and performing iterative training on the opening control network by using the data in the factor time sequence to realize the control of the opening of the regulating valve; calculating a learning rate of an input gate in the opening degree control network in a one-time iterative training process of the opening degree control network; and completing the control of the iteration training by using the learning rate of the input gate. According to the method, the learning rate of the network parameters of the input door in the opening control network is adjusted, so that the training speed of the opening control network and the prediction precision of the opening control network are improved.

Description

Technical Field

[0001] The present invention relates to the field of control and regulation, and more specifically, to a method for controlling the opening of a regulating valve used in a natural gas high-pressure long-distance pipeline. Background Art

[0002] The control of the opening of the regulating valve of the natural gas high-pressure long-distance pipeline is mainly used to adjust the flow and pressure of the fluid to ensure the stability and safety of the transmission system operation. By accurately controlling the opening, energy efficiency can be optimized, losses can be reduced, and overpressure or vacuum conditions in the pipeline can be avoided, thereby protecting the equipment and extending its service life. This control is essential to ensure the efficiency and reliability of the transmission system.

[0003] In order to achieve accurate control of the regulating valve of the natural gas high-pressure long-distance pipeline, it is necessary to predict the opening state of the regulating valve of the natural gas high-pressure long-distance pipeline according to various influencing factors. As a commonly used network for data prediction, the long short-term memory network (LSTM) can not only fit the short-term variation law, but also fit the long-term variation law of the data, so it has a higher prediction accuracy. In order to make the LSTM network have the function of regulating valve opening prediction, the LSTM network needs to be learned and trained. In the process of learning and training the LSTM network, the quality of the hyperparameter setting will affect the training effect of the LSTM network, especially the learning rate used to control the training step of the LSTM network. When it is poorly set, it will seriously affect the prediction accuracy and training efficiency of the LSTM network. Therefore, how to set a suitable learning rate becomes the research focus of the present invention.

[0004] The patent application document with publication number CN112269314A discloses a vehicle chassis decoupling control method based on LSTM neural network. The method in the patent application document only uses LSTM for predictive analysis and does not involve the setting problem of learning rate. Therefore, the method in the patent application document cannot solve the problem in the present invention. Summary of the invention

[0005] In order to solve the problem of how to set the learning rate, the present invention proposes a method for controlling the opening of a regulating valve for a natural gas high-pressure long-distance pipeline, the method comprising the following steps: Obtain the time series of each factor affecting the control of the regulating valve opening; Obtaining a pre-built opening control network, and iteratively training the opening control network using the data in the factor time series to achieve control of the opening of the regulating valve; During an iterative training process of the opening control network, the learning rate of the input gate in the opening control network is calculated. , the time corresponding to the data input into the opening control network during the iterative training is recorded as the analysis time, and the data segment consisting of a preset number of data centered at the analysis time is obtained in each factor time series sequence and recorded as each analysis segment, Represents the preset baseline learning rate, Indicates The analysis section and The correlation of the analysis segments, Indicates the number of seconds after removing the analysis time data. The first analysis segment after removing the analysis time data The correlation of the analysis segments, Indicates the number of types of factor time series, Indicates The data at the time of analysis in the analysis segment, Indicates the use of The fitting data at the analysis moment obtained from other data in the analysis segment, Indicates The correlation between the analysis segment and the output vector of the previous moment, represents the normalization function; The learning rate of the input gate is used to control the training of this iteration.

[0006] The present invention adjusts the learning rate of the input gate network parameters in each iterative training process, so that the opening control network can learn the input information at the analysis time to different degrees according to the demand for the input information at the analysis time during the prediction analysis, so that the opening control network can learn the necessary prediction information more quickly, thereby improving the prediction accuracy of the opening control network and the training efficiency of the opening control network; further, when adjusting the learning rate of the input gate network parameters, the importance of the analysis time data to the prediction analysis is accurately measured by introducing the correlation difference before and after the analysis time data is removed and the difference between the analysis time data and the fitting data, thereby achieving accurate adjustment of the learning rate; further, when adjusting the learning rate of the input gate network parameters, the correlation between the analysis segment and the input vector of the previous moment is introduced to accurately measure the situation of the opening control network learning the regular information at the analysis time, thereby judging the necessity of the opening control network to learn the input information at the analysis time, and then achieving accurate adjustment of the learning rate.

[0007] Preferably, obtaining the learning rate of the forget gate in the opening control network includes: ; Among them, the time before the analysis time is recorded as the preceding time, Indicates The autocorrelation of the data at the previous moment in the time series of the factors, Indicates The data at the previous moment in the time series of the factor The autocorrelation of the correlation of the data at the same time in the time series of the factors, Represents the learning rate of the forget gate in the opening control network.

[0008] When adjusting the learning rate of the forget gate, the present invention measures the importance of the rules of all data before the analysis moment to the prediction analysis by introducing the rules of all data before the analysis moment, thereby controlling the learning record of the previous information by the opening control network; further, when adjusting the learning rate of the forget gate, measures the demand for the rules of all previous data in the future prediction analysis by introducing the correlation between the analysis segment and the output vector of the previous moment, thereby controlling the learning record of the previous information by the opening control network.

[0009] Preferably, the method for obtaining the autocorrelation of the correlation includes: The absolute value of the Pearson correlation coefficient between each analysis segment and another analysis segment of the same period is recorded as the correlation between each data in each factor time series and the data at the same moment in another factor time series; Obtain the autocorrelation value of the correlation between all data in each factor time series and the data at the same moment in another factor time series.

[0010] The present invention reflects the association regularity between different types of factor data at previous moments through the autocorrelation of association, thereby providing a basis for accurate adjustment of the learning rate of the forget gate.

[0011] Preferably, the method for obtaining the output vector at the previous moment includes: The output vector of the output gate in the opening control network at the moment before the analysis moment is obtained and recorded as the output vector at the previous moment.

[0012] Preferably, the obtaining of a pre-built opening control network includes: Construct an LSTM network as the opening control network.

[0013] The present invention uses the LSTM network as the opening control network, so that the opening control network can not only learn short-term regular characteristics, but also learn long-term regular characteristics, so that the opening control network has stronger prediction accuracy.

[0014] Preferably, the use of the learning rate of the input gate to complete the control of the iterative training includes: The learning rate of the network parameters of the input gate and the learning rate of the network parameters of the forget gate are used to control the update amplitude of the network parameters of the input gate and the network parameters of the forget gate during this iterative training, so as to realize the gradient update of the network parameters of the input gate and the forget gate during this iterative training.

[0015] The present invention adjusts the learning rates of the forget gate and the input gate of the opening control network so that the opening control network can learn the information required for prediction and analysis more quickly, thereby improving the speed of iterative training of the opening control network and the prediction accuracy of the opening control network.

[0016] Preferably, the method of realizing the control of the opening degree of the regulating valve includes: The opening control network after iterative training is obtained, and each factor data at the current moment is newly collected into the opening control network after iterative training to obtain the opening value of the regulating valve, and the regulating valve is adjusted according to the opening value.

[0017] Preferably, the method for obtaining the fitting data includes: The least square method is used to fit the data in the analysis segment except the analysis time to obtain a fitting function, and the data at the analysis time is fitted using the fitting function and recorded as the fitting data at the analysis time.

[0018] The present invention has the following beneficial effects: The present invention adjusts the learning rate of the input gate network parameters in each iterative training process, so that the opening control network can learn the input information at the analysis time to different degrees according to the input information demand of the prediction analysis at the analysis time, so that the opening control network can learn the necessary prediction information faster, thereby improving the prediction accuracy of the opening control network and the training efficiency of the opening control network; Furthermore, when adjusting the learning rate of the input gate network parameters, the importance of the analysis time data to the prediction analysis is accurately measured by introducing the correlation difference before and after the analysis time data is removed and the difference between the analysis time data and the fitting data, thereby achieving accurate adjustment of the learning rate; Furthermore, when adjusting the learning rate of the input gate network parameters, by introducing the correlation between the analysis segment and the input vector at the previous moment, the opening control network can be accurately measured to determine the necessity of the opening control network to learn the input information at the analysis moment, thereby achieving accurate adjustment of the learning rate. BRIEF DESCRIPTION OF THE DRAWINGS

[0019] By reading the following detailed description with reference to the accompanying drawings, the above and other objects, features and advantages of the exemplary embodiments of the present invention will become readily understood. In the accompanying drawings, several embodiments of the present invention are shown in an exemplary and non-limiting manner, and the same or corresponding reference numerals represent the same or corresponding parts, wherein: Figure 1 The present invention is a flowchart of a method for controlling the opening of a regulating valve for a natural gas high-pressure long-distance pipeline according to an embodiment of the present invention. DETAILED DESCRIPTION

[0020] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative work are within the scope of protection of the present invention.

[0021] The specific embodiments of the present invention are described in detail below with reference to the accompanying drawings.

[0022] See also Figure 1 , which shows a flow chart of a method for controlling the opening of a regulating valve for a natural gas high-pressure long-distance pipeline provided by an embodiment of the present invention, the method comprising the following steps: S1: Obtain the time series of each factor that affects the control of the regulating valve opening.

[0023] Specifically, the data of each factor affecting the opening control of the regulating valve of the natural gas high-pressure long-distance pipeline at each time is obtained, and the data of all times of each factor obtained are arranged in time sequence to obtain the time series of each factor. The types of factors affecting the opening control of the regulating valve of the natural gas high-pressure long-distance pipeline include but are not limited to the following aspects: temperature, humidity, pressure, and natural gas flow.

[0024] All the factor data at each moment are manually determined to obtain the opening value of the regulating valve at each moment, and the opening value of the regulating valve at each moment is used as a label of all the factor data at the corresponding moment.

[0025] S2: Obtain a pre-constructed opening control network, and iteratively train the opening control network using the data in the factor time series to achieve control of the opening of the regulating valve.

[0026] S20: Obtain a pre-built opening control network.

[0027] Preferably, as an example, obtaining a pre-built opening control network includes: Construct an LSTM network as the opening control network.

[0028] S21: Iteratively train the opening control network using the data in the factor time series.

[0029] It should be noted that the advantage of LSTM network over traditional time series analysis network is that it can record and save historical information. In LSTM network, three gates are mainly used to control the use of information in different time periods during the prediction process. These three gates are input gate, forget gate and output gate. The forget gate is used to control which information of previous moments is stored in the cell state for subsequent use; the input gate is used to control which current input information is stored in the cell state for subsequent use; the output gate is used to control which information in the cell state is used for current prediction analysis. The cell state is a data area that comes with the LSTM network, which is used to record and store the input information of previous moments and useful information in the current input information.

[0030] It should be further explained that, since the use of regular information of previous moments and current moments is different in each moment prediction analysis, some moments use more regular information of previous moments and less regular information of current moments. At this time, more regular information of previous moments should be learned, and less regular information of current moments should be learned; some moments use less regular information of previous moments and more regular information of current moments. At this time, more regular information of current moments should be learned, and less regular information of previous moments should be learned. Therefore, the learning degree should be controlled according to the demand for regular information of previous moments and current moments in each moment prediction analysis. The LSTM network uses the same learning rate for each gate. This way of setting the learning rate cannot control the learning degree of the network for different information. Therefore, different learning rates are set for each gate according to the demand for regular information of previous moments and current moments in each moment prediction analysis.

[0031] S210: During one iterative training process of the aperture control network, calculating the learning rate of the input gate in the aperture control network.

[0032] It should be noted that the input gate is used to control which current input information is stored in the unit state. If the current input information is more important for predictive analysis, the learning rate of the input gate should be increased so that the opening control network can learn more current input information; if the current input information is not important for predictive analysis, the learning rate of the input gate should be lowered so that the opening control network can learn less current input information.

[0033] Preferably, as an example, calculating the learning rate of the input gate in the opening control network includes:

[0034] The time corresponding to the data input into the opening control network during the iterative training is recorded as the analysis time, and the data segment consisting of a preset number of data centered at the analysis time is obtained in each factor time series sequence and recorded as each analysis segment. Represents the preset baseline learning rate, Indicates The analysis section and The correlation of the analysis segments, Indicates the number of seconds after removing the analysis time data. The first analysis segment after removing the analysis time data The correlation of the analysis segments, Indicates the number of types of factor time series, Indicates The data at the time of analysis in the analysis segment, Indicates the use of The fitting data at the analysis moment obtained from other data in the analysis segment, Indicates The correlation between the analysis segment and the output vector of the previous moment, Represents a normalization function, which is used to normalize the data to between -1 and 1. This embodiment uses As the normalization function, other embodiments may set other normalization functions, and this embodiment does not make any specific limitation. Represents the learning rate of the input gate in the opening control network.

[0035] It can be understood that the preset baseline learning rate is the learning rate of the input gate set in the traditional LSTM network. The larger the value, the greater the degree of learning of the input information at the analysis moment; It reflects the change in the correlation of different types of factor data before and after the data loss at the analysis moment. The larger the value, the greater the influence of the data at the analysis moment on the correlation of different types of factor data, and the more important the data at the analysis moment. Therefore, the learning rate of the input gate should be increased so that more information at the analysis moment can be stored in the unit state for subsequent use. It reflects the learning of data rules before and after the data is missing at the analysis moment. The larger the value, the worse the learning of data rules will be if the data is missing at the analysis moment. Therefore, the data at the analysis moment is more important. Therefore, the learning rate of the input gate should be increased so that more information at the analysis moment can be stored in the unit state for subsequent use. It reflects the extent to which the regular information learned by the network contains the regular information at the analysis time. The larger the value, the more regular information the network has learned at the analysis time, and there is no need to learn too much regular information at the analysis time. Therefore, the learning rate needs to be reduced to reduce the learning of regular information at the analysis time.

[0036] It should be supplemented that the method for obtaining the correlation between different analysis segments includes: taking the absolute value of the Pearson correlation coefficient of different types of analysis segments as the correlation between different types of analysis segments.

[0037] The method for obtaining fitting data at the analysis time includes: using the least square method to fit the data in the analysis section except the analysis time to obtain a fitting function, and using the fitting function to fit the data at the analysis time to record it as the fitting data at the analysis time.

[0038] The method for obtaining the output vector includes: obtaining the output vector of the output gate in the opening control network at the previous moment before the analysis moment and recording it as the output vector at the previous moment.

[0039] S211: Obtaining the learning rate of the forget gate in the opening control network.

[0040] Optionally, as an example, obtaining a learning rate of a forget gate in the opening control network includes: The preset benchmark learning rate is used as the learning rate of the forget gate in the opening control network.

[0041] It should be noted that the forget gate is used to control which information of previous moments is stored in the unit state. If the information of previous moments is more important for prediction, the learning rate of the forget gate should be increased so that the opening control network can learn more information of previous moments. If the information of previous moments is not important for prediction, the learning rate of the forget gate should be lowered so that the opening control network can learn less information of previous moments.

[0042] Preferably, as an example, obtaining the learning rate of the forget gate in the opening control network includes: ; Among them, the time before the analysis time is recorded as the preceding time, Indicates The autocorrelation of the data at the previous moment in the time series of the factors, Indicates The data at the previous moment in the time series of the factor The autocorrelation of the correlation of the data at the same time in the time series of the factors, Represents the learning rate of the forget gate in the opening control network.

[0043] It can be seen that Reflects the The regularity of the data at the preceding moment in the time series of this factor. The larger the value, the more regularity of the data at the preceding moment in the time series of this factor. The stronger the regularity of the data at the previous moment in the time series of this factor, the more important the data at the previous moment is for prediction analysis. Therefore, the learning rate of the forget gate should be increased to allow the opening control network to record more regular information at the previous moment. It reflects the similarity between the regularity characteristics of the information before the analysis moment and the regularity characteristics at the analysis moment. The larger the value, the more similar the regularity characteristics of the information before the analysis moment are to the regularity characteristics at the analysis moment. Therefore, the information before the analysis moment has a greater effect on the prediction analysis at the analysis moment. Therefore, the learning rate of the forget gate should be increased so that the opening control network can record more regular information at previous moments. It reflects the regular characteristics of the correlation between different types of factor time series before the analysis time. The larger the value, the stronger the regularity of the correlation between different types of factor time series before the analysis time. Therefore, the learning rate of the forget gate should be increased to allow the opening control network to record more regular information of previous moments.

[0044] It should be added that the method for obtaining the autocorrelation of correlation includes: The absolute value of the Pearson correlation coefficient between each analysis segment and another analysis segment of the same period is recorded as the correlation between each data in each factor time series and the data at the same moment in another factor time series; Obtain the autocorrelation value of the correlation between all data in each factor time series and the data at the same moment in another factor time series.

[0045] It should be noted that the method for calculating the autocorrelation is a prior art and will not be described in detail here.

[0046] S212: Use the learning rate of the input gate and the learning rate of the forget gate to complete the control of this iterative training.

[0047] Preferably, as an example, the learning rate of the network parameters of the input gate and the learning rate of the network parameters of the forget gate are used to control the iterative training, including: After each factor data at the analysis time is input into the opening control network, the output data of the opening control network is obtained; according to the output data and the corresponding label, the loss value is calculated using the loss function of the opening control network.

[0048] The learning rate of the network parameters of the input gate and the forget gate is used to control the update amplitude of the network parameters of the input gate and the forget gate, and the network parameters in the opening control network are reversely updated based on the loss value.

[0049] S22: To realize the control of the opening of the regulating valve.

[0050] Preferably, as an example, to realize the control of the opening degree of the regulating valve, the method includes: The opening control network after iterative training is obtained, and each factor data at the current moment is newly collected into the opening control network after iterative training to obtain the opening value of the regulating valve, and the regulating valve is adjusted according to the opening value.

[0051] At this point, this embodiment is completed.

[0052] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc. made within the principles of the present invention should be included in the protection scope of the present invention.

Claims

1. A method for controlling the opening of a regulating valve for a natural gas high-pressure long-distance pipeline, characterized in that: include: Obtain the time series of each factor affecting the control of the regulating valve opening; Obtaining a pre-built opening control network, and iteratively training the opening control network using the data in the factor time series to achieve control of the opening of the regulating valve; During an iterative training process of the opening control network, the learning rate of the input gate in the opening control network is calculated. , the time corresponding to the data input into the opening control network during the iterative training is recorded as the analysis time, and the data segment consisting of a preset number of data centered at the analysis time is obtained in each factor time series sequence and recorded as each analysis segment, Represents the preset baseline learning rate, Indicates The analysis section and The correlation of the analysis segments, Indicates the number of seconds after removing the analysis time data. The first analysis segment after removing the analysis time data The correlation of the analysis segments, Indicates the number of types of factor time series, Indicates The data at the time of analysis in the analysis segment, Indicates the use of The fitting data at the analysis moment obtained from other data in the analysis segment, Indicates The correlation between the analysis segment and the output vector of the previous moment, represents the normalization function; The learning rate of the input gate is used to control the training of this iteration.

2. A method for controlling the opening of a regulating valve for a natural gas high-pressure long-distance pipeline according to claim 1, characterized in that: Obtaining the learning rate of the forget gate in the opening control network, including: ； Among them, the time before the analysis time is recorded as the preceding time, Indicates The autocorrelation of the data at the previous moment in the time series of the factors, Indicates The data at the previous moment in the time series of the factor The autocorrelation of the correlation of the data at the same time in the time series of the factors, Represents the learning rate of the forget gate in the opening control network.

3. A method for controlling the opening of a regulating valve for a natural gas high-pressure long-distance pipeline according to claim 2, characterized in that: The method for obtaining the autocorrelation of the correlation includes: The absolute value of the Pearson correlation coefficient between each analysis segment and another analysis segment of the same period is recorded as the correlation between each data in each factor time series and the data at the same moment in another factor time series; Obtain the autocorrelation value of the correlation between all data in each factor time series and the data at the same moment in another factor time series.

4. The method for controlling the opening degree of a regulating valve for a natural gas high-pressure long-distance pipeline according to claim 1, characterized in that: The method for obtaining the output vector at the previous moment includes: The output vector of the output gate in the opening control network at the previous moment before the analysis moment is obtained and recorded as the output vector at the previous moment.

5. The method for controlling the opening degree of a regulating valve for a natural gas high-pressure long-distance pipeline according to claim 1, characterized in that: The obtaining of a pre-built opening control network comprises: Construct an LSTM network as the opening control network.

6. A method for controlling the opening of a regulating valve for a natural gas high-pressure long-distance pipeline according to claim 2, characterized in that: The control of the iterative training is completed by using the learning rate of the input gate, including: The learning rate of the network parameters of the input gate and the learning rate of the network parameters of the forget gate are used to control the update amplitude of the network parameters of the input gate and the network parameters of the forget gate during this iterative training, so as to realize the gradient update of the network parameters of the input gate and the forget gate during this iterative training.

7. A method for controlling the opening of a regulating valve for a natural gas high-pressure long-distance pipeline according to claim 1, characterized in that: The method for realizing the control of the opening degree of the regulating valve comprises: The opening control network after iterative training is obtained, and each factor data at the current moment is newly collected into the opening control network after iterative training to obtain the opening value of the regulating valve, and the regulating valve is adjusted according to the opening value.

8. The method for controlling the opening degree of a regulating valve for a natural gas high-pressure long-distance pipeline according to claim 1, characterized in that: The method for obtaining fitting data includes: The least square method is used to fit the data in the analysis segment except the analysis time to obtain a fitting function, and the data at the analysis time is fitted using the fitting function and recorded as the fitting data at the analysis time.

Images