A method and system for automatically adjusting gas flow of a gas generator

Abstract

The present application provides a method and system for automatically regulating the gas flow of a gas generator, relating to the technical field of adaptive regulation. The method comprises: collecting historical data and real-time data, integrating the historical data into a training set; establishing a control model, wherein the control model comprises a BI-LSTM sub-model; training the BI-LSTM sub-model using the training set to obtain a trained BI-LSTM sub-model; inputting a target gas flow and a current pressure of the gas generator into the trained BI-LSTM sub-model to obtain a target valve opening, and adjusting the current valve opening to the target valve opening. The present application trains the BI-LSTM sub-model using historical data, considers the influence of multiple factors during the training process, and improves the control accuracy of the valve.

Description

Technical Field

[0001] The present application relates to the technical field of adaptive regulation, and in particular to a method and system for automatically regulating the gas flow of a gas generator. Background Art

[0002] Gas generators are critical equipment in industrial production, energy conversion, and numerous scientific research fields. Their performance stability and efficiency are directly related to the operational effectiveness and cost-effectiveness of the entire system. Precise control of gas flow is a key element in the efficient and safe operation of gas generators. Traditional gas flow control methods rely on empirical formulas, PID (proportional-integral-derivative) controllers, or simple open-loop control systems. These control methods often fail to achieve ideal control results when faced with complex and changing operating conditions.

[0003] Traditional control methods are usually based on fixed mathematical models or empirical parameters, but the actual operating conditions of the gas generator are affected by many factors, such as changes in gas generator pressure and valve aging. These factors will lead to deviations between model parameters and actual conditions. In addition, changes in gas flow are often time-varying and nonlinear. Traditional linear control strategies are difficult to effectively cope with this complex change pattern, which seriously affects the control accuracy of the valve. Summary of the Invention

[0004] In order to improve the control accuracy of the valve, the present application provides a method and system for automatically adjusting the gas flow of a gas generator.

[0005] In a first aspect, the present application provides a method for automatically adjusting the gas flow of a gas generator, which adopts the following technical solution:

[0006] A method for automatically adjusting the gas flow of a gas generator comprises the following steps:

[0007] Data collection: including the first collection and the second collection;

[0008] First, collect historical gas flow, historical pressure data of the gas generator, and historical valve opening, and integrate the historical gas flow, historical pressure data of the gas generator, and historical valve opening into a training set;

[0009] Second collection: collect target gas flow and current pressure data of the gas generator;

[0010] First modeling: establishing a control model, wherein the control model includes a BI-LSTM sub-model;

[0011] Model training: Use the training set to train the BI-LSTM sub-model to obtain the trained BI-LSTM sub-model;

[0012] First prediction: Input the target gas flow rate and the current gas generator pressure into the trained BI-LSTM sub-model to obtain the target valve opening;

[0013] Adjustment: Adjust the current valve opening to the target valve opening.

[0014] By adopting the above-mentioned technical solution, this application collects historical gas flow, historical pressure data of the gas generator, and historical valve opening, providing a rich data set for the model training step. This data covers the performance of the gas generator under different operating conditions, helping the control model learn the complex relationship between gas flow, pressure, and valve opening. The control model of this application includes a BI-LSTM sub-model, which can capture long-term dependencies in time series data while considering past and future information. It is suitable for predicting problems with time series characteristics such as gas flow. This application also uses a training set to train the BI-LSTM sub-model, enabling it to learn the mapping relationship between gas flow, pressure, and valve opening. Through training, the BI-LSTM sub-model can more accurately predict the target valve opening for a given target gas flow and current pressure, can adapt to different operating conditions and gas generator characteristics, and has strong generalization capabilities. By combining data-driven and BI-LSTM models, this application can achieve precise control of gas flow and improve the operating efficiency and stability of the gas generator.

[0015] Optionally, the control model further includes a BP neural network sub-model, and the model training step further includes:

[0016] The third collection: collect the weight coefficients of the BI-LSTM sub-model during each training process, which is recorded as the first data;

[0017] Fourth collection: collecting controller parameters corresponding to the historical gas flow during each training process, and recording them as second data;

[0018] First training: using the first data and the second data to train the BP neural network sub-model to obtain a trained BP neural network sub-model;

[0019] Fifth collection: Collect the weight coefficients of the trained BI-LSTM sub-model, recorded as the third data;

[0020] Second prediction: input the third data into the trained BP neural network sub-model to obtain predicted controller parameters;

[0021] Parameter update: Update the current controller parameters to the predicted controller parameters.

[0022] By adopting the above technical solution, the BP (back propagation) neural network excels at processing complex nonlinear mapping relationships. Introducing it into the control model can complement the time series data processed by the BI-LSTM sub-model, further improving the control model's prediction accuracy and generalization capabilities. This application collects the weight coefficients (first data) and corresponding controller parameters (second data) during the BI-LSTM sub-model training process, providing a rich set of training samples for the BP neural network sub-model. These training samples not only capture the learning process of the BI-LSTM sub-model but also directly relate to the controller parameters. This application uses the first and second data to train the BP neural network sub-model, enabling it to predict controller parameters given the BI-LSTM sub-model's weight coefficients. By collecting the weight coefficients (third data) of the trained BI-LSTM sub-model and inputting them into the trained BP neural network sub-model, predicted controller parameters are obtained. This step achieves a direct mapping from the BI-LSTM sub-model to the controller parameters, improving prediction accuracy and efficiency.

[0023] Optionally, after executing the parameter updating step and before executing the first prediction step, the method further includes:

[0024] Sixth acquisition: obtaining the negative adjustment amount of the gas generator under each second data, recorded as fourth data; obtaining the negative adjustment amount of the gas generator under the predicted controller parameters, recorded as fifth data;

[0025] Sorting: sorting the fourth data in ascending order to obtain a fourth data sequence;

[0026] Calculate distances: Calculate the distance between the fifth data and the first fourth data in the fourth data sequence, which is recorded as the first distance; calculate the distance between the fifth data and the second fourth data in the fourth data sequence, which is recorded as the second distance; calculate the distance between the fifth data and the third fourth data in the fourth data sequence, which is recorded as the third distance;

[0027] Obtaining the minimum value: recording the minimum value among the first distance, the second distance, and the third distance as the sixth data;

[0028] Distance judgment: judging whether the sixth data is greater than a preset distance threshold, if so, executing the step of adding data; if not, executing the step of first prediction;

[0029] Add data: Add the predicted controller parameters to the parameter set consisting of the second data and perform the first prediction steps.

[0030] By adopting the above technical solution, by obtaining the negative regulation (fourth data) of the gas generator under each historical controller parameter (second data) and the negative regulation (fifth data) under the predicted controller parameters, a practical basis is provided for evaluating the rationality of the predicted parameters. This application sorts the fourth data to obtain a fourth data sequence and calculates the distance between the fifth data and the first three fourth data in the fourth data sequence. This allows for an intuitive assessment of the closeness between the predicted parameters and the historical optimal parameters. This application selects the value with the smallest distance (sixth data) as the evaluation criterion. When the sixth data is greater than a preset distance, it indicates that the fifth data is far from the fourth data, indicating that the current controller parameters differ from the historical controller parameter distribution, possibly indicating a new operating condition or a gas generator anomaly requiring further processing and analysis. This application sets a preset distance threshold to determine whether to add the predicted controller parameters to the parameter set. This means that only predicted controller parameters that are sufficiently good (i.e., sufficiently close to the historical optimal parameters) will be adopted, reducing the risk of performance degradation due to poor parameters.

[0031] Optionally, after performing the step of determining the distance and before performing the step of adding data, the method further includes:

[0032] Pressure judgment: judge whether the current pressure of the gas generator meets the expectation. If so, execute the step of adding data; if not, execute the step of early warning;

[0033] Early warning: Send out an alarm signal.

[0034] By adopting the above technical solution, when the controller parameters do not conform to the historical controller parameter distribution, the current pressure of the gas generator is verified to determine whether it meets expectations. It is determined whether the gas generator has an abnormality (whether the gas generator has failed). If an abnormality occurs, an early warning is issued. Otherwise, it indicates that the current controller parameters are due to new operating conditions, and the first prediction step is executed.

[0035] Optionally, after performing the step of data collection and before performing the step of first modeling, the method further includes:

[0036] Data processing: including calculation of correlation, correlation judgment, retention and deletion;

[0037] Calculate correlation: Calculate the correlation between historical gas flow and historical valve opening;

[0038] Determination of correlation: Determine whether the correlation is greater than zero. If so, execute the retaining step; if not, execute the deleting step;

[0039] Retention: Retain the historical gas flow corresponding to the current correlation, the historical pressure data of the gas generator corresponding to the historical gas flow, and the historical valve opening, and use the retained historical gas flow, the historical pressure data of the gas generator corresponding to the historical gas flow, and the historical valve opening as a new training set;

[0040] Deletion: Delete the historical gas flow corresponding to the current correlation degree, the historical pressure data of the gas generator corresponding to the historical gas flow, and the historical valve opening.

[0041] By adopting the above technical solution, in the process of adjusting the valve opening so that the current gas flow reaches the target gas flow, the phenomenon of negative adjustment will occur. This application can quantify the degree of correlation between the historical gas flow and the historical valve opening by calculating the correlation between the two. When the valve is adjusted, the gas generator is in the stage of increasing negative adjustment, and the historical gas flow and the historical valve opening are negatively correlated (that is, the gas flow decreases when the valve opening increases, or the gas flow increases when the valve opening decreases). After the stage of increasing negative adjustment, the two are positively correlated. This application deletes the historical gas flow corresponding to the stage of increasing negative adjustment, the historical pressure data of the gas generator corresponding to the historical gas flow, and the historical valve opening, that is, deletes the data that is confusing to the control model, optimizes the training set, and reduces the risk of the model learning the data characteristics of the negative correlation stage during the training process.

[0042] Optionally, the control model further includes a Bayesian network sub-model, and after performing the data processing step and before performing the first modeling step, further includes:

[0043] Determine nodes: The deleted historical gas flow, the historical pressure data of the gas generator corresponding to the historical gas flow, and the historical valve opening are used as nodes of the Bayesian network sub-model, wherein the directed edges of the Bayesian network sub-model are the dependencies between the nodes;

[0044] Reasoning: Input the target gas flow rate and the current pressure of the gas generator into the Bayesian network sub-model, and output the inferred value of the valve opening, which is recorded as the seventh data;

[0045] Calculating the average: calculating the average of the seventh data and the target valve opening, and using the average as the new target valve opening.

[0046] By adopting the above technical solution, deleted historical data (including historical gas flow, historical gas generator pressure data corresponding to the historical gas flow, and historical valve opening) are used as nodes of the Bayesian network sub-model, realizing the reuse of these deleted data. These deleted data correspond to the stage of increasing negative regulation of the gas generator. Through the inference mechanism of the Bayesian network, the data characteristics of the stage of increasing negative regulation are specifically learned. This data not only provides additional information for the control model, but also helps the control model to more comprehensively understand the behavior patterns of the gas generator in different operating stages. By learning the data characteristics of the stage of increasing negative regulation, the Bayesian network sub-model can more accurately predict the gas flow and valve opening requirements under similar conditions.

[0047] Optionally, after executing the step of calculating the degree of association and before executing the step of determining the degree of association, the method further includes:

[0048] Second modeling: Obtain valve degradation data and establish a degradation model based on the degradation data. The calculation model of the degradation model is as follows:

[0049] ;

[0050] in, is the actual measured gas flow; is the theoretical maximum gas flow rate; is the degradation coefficient; t is the service time of the valve;

[0051] Updating the correlation degree: updating the correlation degree based on the degradation coefficient, and using the updated correlation degree as a new correlation degree.

[0052] By adopting the above technical solution, valves, as key components in gas generators, gradually degrade over time. This degradation directly affects the relationship between valve opening and gas flow. By introducing a degradation model, the performance changes of valves during actual use can be more accurately reflected, thereby improving the accuracy of the correlation. Traditional correlation calculations are often based on static historical data, ignoring the dynamic factor of valve degradation. Updating the correlation using the degradation coefficient calculated by the degradation model can make the correlation more accurate to current conditions and improve the timeliness of the control model. Incorporating valve degradation information into the gas flow prediction model allows the control model to more accurately predict future gas flow trends. Based on the updated correlation, control parameters (such as valve opening) can be more intelligently adjusted to accommodate performance changes caused by valve degradation, helping to reduce control errors and improve the stability and accuracy of gas flow.

[0053] Optionally, after performing the step of data collection and before performing the step of first modeling, the method further includes:

[0054] Quantity judgment: judge whether the number of training samples in the training set is greater than a preset threshold. If so, execute the first modeling step; if not, execute the step of expanding the training set;

[0055] Expand the training set: Based on the training set, use the generative adversarial network to generate new training samples and add the new training samples to the training set.

[0056] By employing this technical solution, sufficient data support is ensured before model training by determining whether the number of training samples in the training set exceeds a preset threshold. If the number of training samples is insufficient, a generative adversarial network (GAN) is used to generate new training samples, thereby increasing the diversity and richness of the training set. A diverse training set helps the control model better learn the essential characteristics of the data, rather than simply memorizing specific samples in the training set. By expanding the training set, the model can learn more data features, thereby improving the accuracy of gas flow predictions.

[0057] Optionally, the step of expanding the training set further includes:

[0058] Calculate similarity: Calculate the similarity between the new training sample and the training sample in the original sample set;

[0059] Similarity judgment: judge whether the similarity is greater than a preset threshold. If so, do not retain the new training sample; if not, retain the new training sample.

[0060] By using this technical solution, we can quantify the differences between new training samples and existing samples by calculating their similarity with those in the original sample set. If the new sample is too similar to a sample in the original sample set, it means that the information they provide may be duplicated. In this case, the new training sample is not retained. Otherwise, the new training sample is retained. This step can increase the diversity and information richness of the training set.

[0061] In a second aspect, the present application provides a gas generator gas flow automatic regulation system, which adopts the following technical solutions:

[0062] A gas generator gas flow automatic regulation system, comprising:

[0063] The data acquisition module includes a first acquisition unit and a second acquisition unit;

[0064] The first acquisition unit is used to collect historical gas flow, historical pressure data of the gas generator, and historical valve opening, and integrate the historical gas flow, historical pressure data of the gas generator, and historical valve opening into a training set;

[0065] The second acquisition unit is used to collect the target gas flow rate and the current pressure of the gas generator;

[0066] A first modeling module is used to establish a control model, wherein the control model includes a BI-LSTM sub-model;

[0067] A model training module, which is in communication with the first acquisition unit and the first modeling module, and is used to train the BI-LSTM sub-model using the training set to obtain a trained BI-LSTM sub-model;

[0068] The first prediction module is in communication with the second acquisition unit and the model training module, and is used to input the target gas flow rate and the current pressure of the gas generator into the trained BI-LSTM sub-model to obtain the target valve opening;

[0069] The regulating module is in communication with the first prediction module and is configured to regulate the current valve opening to a target valve opening.

[0070] By adopting the above technical solution, this application realizes comprehensive and targeted data collection by integrating the first acquisition unit and the second acquisition unit, constructs an advanced control model using the BI-LSTM sub-model, and realizes efficient and accurate gas flow automatic control through model training, real-time prediction and precise adjustment. This system has high automation, high-precision control, strong adaptability and good scalability, providing strong support for the stable operation and energy efficiency optimization of the gas generator.

[0071] In summary, this application includes at least one of the following beneficial technical effects:

[0072] 1. This application collects historical gas flow, historical gas generator pressure data, and historical valve openings, providing a rich dataset for model training. This data covers the performance of the gas generator under different operating conditions and helps the control model learn the complex relationship between gas flow, pressure, and valve opening. This application's control model includes a BI-LSTM sub-model, which can capture long-term dependencies in time series data while considering both past and future information. This makes it suitable for predicting problems with time series characteristics, such as gas flow. This application also uses a training set to train the BI-LSTM sub-model, enabling it to learn the mapping relationship between gas flow, pressure, and valve opening. Through training, the BI-LSTM sub-model can more accurately predict the target valve opening for a given target gas flow and current pressure, can adapt to different operating conditions and gas generator characteristics, and has strong generalization capabilities. By combining data-driven and BI-LSTM models, this application can achieve precise control of gas flow and improve the operating efficiency and stability of the gas generator.

[0073] 2. BP (back-propagation) neural networks excel at processing complex nonlinear mapping relationships. Introducing them into the control model can complement the time series data processed by the BI-LSTM sub-model, further improving the control model's prediction accuracy and generalization capabilities. This application collects the weight coefficients (first data) and corresponding controller parameters (second data) during the BI-LSTM sub-model training process, providing a rich set of training samples for the BP neural network sub-model. These training samples not only capture the BI-LSTM sub-model's learning process but also directly correlate to the controller parameters. This application uses the first and second data to train the BP neural network sub-model, enabling it to predict controller parameters given the BI-LSTM sub-model's weight coefficients. By collecting the weight coefficients (third data) of the trained BI-LSTM sub-model and inputting them into the trained BP neural network sub-model, predicted controller parameters are obtained. This step achieves a direct mapping from the BI-LSTM sub-model to the controller parameters, improving prediction accuracy and efficiency.

[0074] 3. By obtaining the negative regulation (fourth data) of the gas generator under each historical controller parameter (second data) and the negative regulation (fifth data) under the predicted controller parameters, a practical basis is provided for evaluating the rationality of the predicted parameters. This application sorts the fourth data to obtain a fourth data sequence and calculates the distance between the fifth data and the first three fourth data in the fourth data sequence. This allows for an intuitive assessment of the closeness between the predicted parameters and the historical optimal parameters. This application selects the value with the smallest distance (sixth data) as the evaluation criterion. When the sixth data is greater than a preset distance, it indicates that the fifth data is relatively far from the fourth data, indicating that the current controller parameters differ from the historical controller parameter distribution, possibly indicating a new operating condition or a gas generator anomaly requiring further processing and analysis. This application sets a preset distance threshold to determine whether to add the predicted controller parameters to the parameter set. This means that only predicted controller parameters that are sufficiently good (i.e., sufficiently close to the historical optimal parameters) will be adopted, reducing the risk of performance degradation due to poor parameters. BRIEF DESCRIPTION OF THE DRAWINGS

[0075] Figure 1 is a flow chart of Example 1 of the present application;

[0076] Figure 2 This is a flowchart from S31 third acquisition to S36 parameter update in Example 2 of the present application;

[0077] Figure 3 This is a flow chart from S61 sixth acquisition to S68 adding data in Example 2 of the present application;

[0078] Figure 4This is a flowchart of data processing from S7 to mean calculation in S83 in Example 3 of the present application. DETAILED DESCRIPTION

[0079] The following combination Figures 1 to 4 This application is described in further detail.

[0080] Example 1: This example discloses a method for automatically adjusting the gas flow of a gas generator, referring to Figure 1 The method includes: S1 data acquisition, S2 first modeling, S3 model training, S4 first prediction and S5 adjustment. First, historical data and real-time data are collected, then a control model is established, and then the control model is trained using historical data. Then, the real-time data is input into the trained control model, and the predicted target valve opening is output. Then, the current valve opening is adjusted according to the target valve opening. This embodiment includes the following steps:

[0081] S1 data acquisition includes S11 first acquisition and S12 second acquisition.

[0082] S11 is the first collection step, which extracts historical gas flow, historical pressure data of the gas generator, and historical valve opening, and integrates the historical gas flow, historical pressure data of the gas generator, and historical valve opening into a training set.

[0083] The historical gas flow, the historical pressure data of the gas generator and the historical valve opening are collected at the same time.

[0084] S12 is the second collection step, which collects target gas flow rate and current pressure data of the gas generator.

[0085] S2 first builds a control model, which includes a BI-LSTM sub-model. This sub-model learns the complex relationship between gas flow, gas generator pressure, and valve opening, providing a foundation for subsequent prediction and control.

[0086] In S3 model training, the BI-LSTM sub-model is trained using the training set collected in the first collection of S11 to obtain a trained BI-LSTM sub-model.

[0087] In the process of training the BI-LSTM sub-model, the historical gas flow rate and historical pressure data of the gas generator in the training set are used as training data, and the historical valve opening is used as the true label of the training data.

[0088] In the first prediction step S4, the target gas flow rate and the current pressure of the gas generator collected in the second collection step S12 are input into the trained BI-LSTM sub-model to obtain the target valve opening.

[0089] S5 adjusts the current valve opening to the target valve opening.

[0090] This embodiment achieves efficient and precise automated control of gas flow through the steps of data collection, first modeling, model training, first prediction and adjustment, thereby improving the operating efficiency and energy efficiency performance of the gas generator.

[0091] Example 2: Reference Figure 2 The difference between this embodiment and embodiment 1 is that the control model further includes a BP neural network sub-model, the BI-LSTM sub-model is communicatively connected to the BP neural network sub-model, and in the S3 model training, it also includes:

[0092] S31 is the third collection step. Each time the BI-LSTM sub-model is trained, its weight coefficient is recorded, and the weight coefficient is recorded as the first data. The first data reflects the degree of fit of the BI-LSTM sub-model to the data and the learned features.

[0093] S32 is the fourth acquisition, which inputs the target gas flow rate and the current pressure of the gas generator into the BI-LSTM sub-model after each training, obtains the target valve opening predicted by the BI-LSTM sub-model after each training, uses a controller to control the valve to transition from the current opening to the target opening, obtains the controller parameters in this process, and records the controller parameters as the second data.

[0094] S33 is a first training step, in which the first data and the second data are used to train the BP neural network sub-model to obtain a trained BP neural network sub-model.

[0095] In the process of training the BP neural network sub-model, the first data is used as training data, and the second data is used as the true label of the first data.

[0096] S34 is the fifth collection, collecting the weight coefficients of the BI-LSTM sub-model trained in S3 model training, and recording the weight coefficients as the third data.

[0097] S35 second prediction: input the third data into the trained BP neural network sub-model to obtain predicted controller parameters.

[0098] S36 parameter update, updates the current controller parameters to the controller parameters predicted in S35 second prediction.

[0099] This embodiment enhances the model training process by introducing a BP neural network sub-model, thereby achieving intelligent prediction and optimization of controller parameters, which not only improves the accuracy and stability of gas flow control, but also enhances the adaptability and robustness.

[0100] Reference Figure 3In other embodiments, after executing S36 parameter update and before executing S4 first prediction, the method further includes:

[0101] S61 is a sixth collection step, in which the negative adjustment amount of the gas generator under each second data is obtained and recorded as the fourth data.

[0102] Then, the negative adjustment amount of the gas generator under the predicted controller parameters is obtained and recorded as the fifth data.

[0103] S62 sorts the fourth data in ascending order to form an ordered data sequence, which is recorded as a fourth data sequence.

[0104] S63 calculates the distance, calculates the distance between the fifth data and the fourth data ranked first in the fourth data sequence, recorded as the first distance; calculates the distance between the fifth data and the fourth data ranked second in the fourth data sequence, recorded as the second distance; calculates the distance between the fifth data and the fourth data ranked third in the fourth data sequence, recorded as the third distance.

[0105] For example, in the fourth data sequence, the first three fourth data are 3%, 5%, and 8%, respectively. The fifth data is 4%, so the distance between the fifth data and the first three fourth data is:

[0106] First distance: |4% - 3%|=1%;

[0107] Second distance: |4% - 5%|=1%;

[0108] Third distance: |4% - 8%|=4%.

[0109] S64 obtains a minimum value by comparing the first distance, the second distance, and the third distance to obtain the minimum value among the three, and records the minimum value as the sixth data.

[0110] Based on the first distance, the second distance, and the third distance calculated in S63 calculating the distance, the sixth data is 1%.

[0111] S65 distance judgment, judge whether the sixth data is greater than the preset distance threshold. If so, it means that the difference between the predicted controller parameters and the existing optimal parameters is large, which may correspond to different working conditions or there may be a fault in the gas generator. It is necessary to perform S66 pressure judgment for further analysis; if not, it means that the predicted controller parameters are close enough to the optimal parameters, and the parameters can be directly updated, and then the first prediction in S4 is executed.

[0112] Based on the comparison between the sixth data (1%) obtained in obtaining the minimum value in S64 and the preset distance threshold (eg, 0.5%), it is found that 1%>0.5%, and at this time, it is necessary to perform S66 pressure determination.

[0113] S66 pressure judgment, judges whether the current pressure of the gas generator meets expectations. If so, it means that the sixth data is greater than the preset distance threshold due to new working conditions, and then executes S68 to add data; if not, it means that it is caused by a gas generator failure, and it is necessary to execute S67 warning.

[0114] The current pressure of the gas generator meets the expected standard: the current pressure of the gas generator is within a safe range.

[0115] S67 early warning, issuing an alarm signal.

[0116] S68 adds data, adds the predicted controller parameters to the parameter set composed of the second data, and executes S4 first prediction.

[0117] This embodiment can intelligently identify new operating conditions or fault conditions by comparing the distance between the negative adjustment amount under the predicted controller parameters and the optimal parameters, and combining pressure judgment, so as to decide whether to update the parameter set or issue an early warning, thereby effectively improving the accuracy and safety of valve control.

[0118] Example 3: Reference Figure 4 The difference between this embodiment and embodiment 1 is that the control model further includes a Bayesian network sub-model, and after executing S1 data acquisition and before executing S2 first modeling, it also includes:

[0119] S7 data processing includes S71 calculating correlation, S72 second modeling, S73 updating correlation, S74 correlation judgment, S75 retaining and S76 deleting.

[0120] S71 calculates the correlation between historical gas flow and historical valve opening. By using statistical methods (such as correlation coefficient and mutual information) to calculate the correlation between the two, we can understand the strength of the linear or nonlinear relationship between them.

[0121] For example, the correlation between the historical gas flow rate and the historical valve opening is calculated by the correlation coefficient. The historical data records involved in this step are shown in Table 1.

[0122] Table 1 Historical gas flow and historical valve opening data

[0123]

[0124] The calculation model of the correlation coefficient is as follows:

[0125] ;

[0126] in, is the historical gas flow at time point i; is the historical valve opening at time point i; is the mean of historical gas flow; is the average value of historical valve opening.

[0127] After calculation, we can know the average value of historical gas flow , the average value of historical valve opening .

[0128] ;

[0129] ;

[0130] .

[0131] S72: Second modeling, obtaining valve degradation data, and establishing a degradation model based on the degradation data. The calculation model of the degradation model is as follows:

[0132] ;

[0133] in, is the actual measured gas flow rate, in m³ / h; The theoretical maximum gas flow rate, in m³ / h; is the degradation coefficient; t is the service life of the valve, unit is h. In the calculation model of the degradation model, t is only entered as a numerical value without using a unit.

[0134] The degradation model describes how valve performance changes over time, particularly its impact on gas flow. The degradation coefficient is a key parameter in the degradation model, reflecting the degree of valve performance degradation due to factors such as wear and aging. The degradation coefficient can be calculated using the actual measured gas flow, the theoretical maximum gas flow, and the valve's operating time.

[0135] S73 updates the correlation degree, updates the correlation degree based on the degradation coefficient, and uses the updated correlation degree as a new correlation degree.

[0136] Updating the correlation degree based on the degradation coefficient may be based on the original correlation degree, adding the degradation coefficient (or degradation coefficient multiples of), or directly multiply the original correlation by .

[0137] S74 determines the correlation degree, determines whether the updated correlation degree described in S73 is greater than zero. If so, it indicates that there is a positive correlation between the gas flow and the valve opening. At this time, the data corresponding to the correlation degree is not the data in the stage of increasing negative adjustment. Then, S75 is executed to retain. If not, it indicates that there is a negative correlation between the gas flow and the valve opening. At this time, the data corresponding to the correlation degree is the data in the stage of increasing negative adjustment. These data are no longer representative and need to be deleted by executing S76.

[0138] S75 retains the historical gas flow corresponding to the current correlation degree, the historical pressure data of the gas generator corresponding to the historical gas flow, and the historical valve opening, and integrates the retained historical gas flow, the historical pressure data of the gas generator corresponding to the historical gas flow, and the historical valve opening into a new training set.

[0139] S76 deletion deletes the historical gas flow corresponding to the current correlation, the historical pressure data of the gas generator corresponding to the historical gas flow, and the historical valve opening, reducing the negative impact of these data on model training, so that the control model can learn and predict based on high-quality data.

[0140] S81 determines the nodes, and uses the deleted historical gas flow, the historical pressure data of the gas generator corresponding to the historical gas flow, and the historical valve opening as nodes of the Bayesian network sub-model, wherein the directed edges of the Bayesian network sub-model are the dependencies between the nodes.

[0141] S82 reasoning uses the conditional probability table (CPT) in the Bayesian network and known evidence variables (such as the target gas flow rate and the current pressure of the gas generator) to calculate the posterior probability distribution of unknown variables (such as valve opening).

[0142] This step uses the target gas flow rate and the current gas generator pressure as inputs, performs inference through the Bayesian network sub-model, and outputs an inferred value for the valve opening, recorded as the seventh data point. This inferred value is calculated based on the dependency relationship between the current input and historical data, and reflects the most likely value of the valve opening under given conditions.

[0143] S83 calculates the average value, that is, calculates the average value of the seventh data and the target valve opening, and uses the average value as the new target valve opening.

[0144] In this embodiment, the relevant data of the negative regulation increasing stage is deleted and the deleted data is reused. The Bayesian network model is used to learn the data characteristics of the negative regulation increasing stage, and the Bayesian network is used to perform inference based on the data of the negative regulation increasing stage to obtain the inferred value of the valve opening. Then, the mean of the target valve opening and the seventh data is solved to obtain the new target valve opening.

[0145] Example 4: This example differs from Example 1 in that, after executing S1 data collection and before executing S2 first modeling, it further includes:

[0146] S91 Quantity judgment, determines whether the number of training samples in the training set is greater than the preset threshold. If so, it is considered that the data volume of the training set is sufficient, and S2 first modeling is executed; if not, it is considered that the data volume of the training set is insufficient, and it is necessary to execute S92 to expand the training set and increase the training data in the training set.

[0147] S92 expands the training set and uses a generative adversarial network to generate new training samples based on the training set.

[0148] In other embodiments, the S92 expanded training set further includes:

[0149] S921 calculates similarity, using algorithms such as Euclidean distance and cosine similarity to calculate the similarity between the new training sample and the training samples in the original sample set.

[0150] S922 similarity determination, determining whether the similarity is greater than a preset threshold, if so, not retaining the new training sample; if not, retaining the new training sample and adding the retained training sample to the training set.

[0151] This embodiment enables the training set to meet the modeling requirements in terms of data volume and diversity, thereby improving the accuracy and generalization ability of the control model.

[0152] Example 5: This example discloses a gas generator gas flow automatic regulation system, the system comprising:

[0153] The data acquisition module includes a first acquisition unit and a second acquisition unit.

[0154] The first acquisition unit is used to collect historical data, where the historical data includes historical gas flow, historical pressure data of the gas generator, and historical valve opening information.

[0155] Then the historical gas flow, historical pressure data of the gas generator and historical valve opening are integrated into the training set.

[0156] The second acquisition unit is used to acquire the target gas flow rate and the current pressure of the gas generator.

[0157] The first modeling module is used to establish a control model, where the control model includes a BI-LSTM sub-model.

[0158] The model training module is in communication with the first acquisition unit and the first modeling module and is configured to train the BI-LSTM sub-model using the training set to obtain a trained BI-LSTM sub-model. Through iterative learning, the BI-LSTM sub-model is able to accurately predict the relationship between gas flow, pressure, and valve opening.

[0159] The first prediction module is in communication with the second acquisition unit and the model training module, and is used to input the target gas flow rate and the current pressure of the gas generator into the trained BI-LSTM sub-model to obtain the target valve opening.

[0160] The regulating module is in communication with the first prediction module and is configured to regulate the current valve opening to a target valve opening.

[0161] This embodiment achieves precise control of the gas generator valve opening through the collaborative work of multiple modules such as data acquisition, first modeling, model training, first prediction and adjustment. This control method based on the BI-LSTM sub-model can make full use of historical data and real-time data to improve the accurate control of the valve.

[0162] The above are all preferred embodiments of the present application, and are not intended to limit the scope of protection of the present application. Therefore, any equivalent changes made based on the structure, shape, and principle of the present application should be included in the scope of protection of the present application.

Claims

1. A method for automatically adjusting the gas flow of a gas generator, characterized in that: include: Data collection: including the first collection and the second collection; First, collect historical gas flow, historical pressure data of the gas generator, and historical valve opening, and integrate the historical gas flow, historical pressure data of the gas generator, and historical valve opening into a training set; Second collection: collect target gas flow and current pressure data of the gas generator; First modeling: establishing a control model, wherein the control model includes a BI-LSTM sub-model; Model training: Use the training set to train the BI-LSTM sub-model to obtain the trained BI-LSTM sub-model; First prediction: Input the target gas flow rate and the current gas generator pressure into the trained BI-LSTM sub-model to obtain the target valve opening; Adjustment: adjust the current valve opening to the target valve opening; The control model also includes a BP neural network sub-model, and in the model training step, also includes: The third collection: collect the weight coefficients of the BI-LSTM sub-model during each training process, which is recorded as the first data; Fourth collection: collecting controller parameters corresponding to the historical gas flow during each training process, and recording them as second data; First training: using the first data and the second data to train the BP neural network sub-model to obtain a trained BP neural network sub-model; Fifth collection: Collect the weight coefficients of the trained BI-LSTM sub-model, recorded as the third data; Second prediction: input the third data into the trained BP neural network sub-model to obtain predicted controller parameters; Parameter update: Update the current controller parameters to the predicted controller parameters; After the step of performing parameter updating and before the step of performing the first prediction, the method further includes: Sixth acquisition: obtaining the negative adjustment amount of the gas generator under each second data, recorded as fourth data; obtaining the negative adjustment amount of the gas generator under the predicted controller parameters, recorded as fifth data; Sorting: sorting the fourth data in ascending order to obtain a fourth data sequence; Calculate distances: Calculate the distance between the fifth data and the first fourth data in the fourth data sequence, which is recorded as the first distance; calculate the distance between the fifth data and the second fourth data in the fourth data sequence, which is recorded as the second distance; calculate the distance between the fifth data and the third fourth data in the fourth data sequence, which is recorded as the third distance; Obtaining the minimum value: recording the minimum value among the first distance, the second distance, and the third distance as the sixth data; Distance judgment: judging whether the sixth data is greater than a preset distance threshold, if so, executing the step of adding data; if not, executing the step of first prediction; Adding data: adding the predicted controller parameters to the parameter set consisting of the second data and executing the first prediction step; After executing the step of distance determination and before executing the step of adding data, the method further includes: Pressure judgment: judge whether the current pressure of the gas generator meets the expectation. If so, execute the step of adding data; if not, execute the step of early warning; Early warning: Send out an alarm signal.

2. The method for automatically adjusting the gas flow of a gas generator according to claim 1, characterized in that: After executing the step of data collection and before executing the step of first modeling, the method further includes: Data processing: including calculation of correlation, correlation judgment, retention and deletion; Calculate correlation: Calculate the correlation between historical gas flow and historical valve opening; Determination of correlation: Determine whether the correlation is greater than zero. If so, execute the retaining step; if not, execute the deleting step; Retention: Retain the historical gas flow corresponding to the current correlation, the historical pressure data of the gas generator corresponding to the historical gas flow, and the historical valve opening, and use the retained historical gas flow, the historical pressure data of the gas generator corresponding to the historical gas flow, and the historical valve opening as a new training set; Deletion: Delete the historical gas flow corresponding to the current correlation degree, the historical pressure data of the gas generator corresponding to the historical gas flow, and the historical valve opening.

3. The method for automatically adjusting the gas flow of a gas generator according to claim 2, characterized in that: The control model further includes a Bayesian network sub-model, and after performing the data processing step and before performing the first modeling step, further includes: Determine nodes: The deleted historical gas flow, the historical pressure data of the gas generator corresponding to the historical gas flow, and the historical valve opening are used as nodes of the Bayesian network sub-model, wherein the directed edges of the Bayesian network sub-model are the dependencies between the nodes; Reasoning: Input the target gas flow rate and the current pressure of the gas generator into the Bayesian network sub-model, and output the inferred value of the valve opening, which is recorded as the seventh data; Calculating the average: calculating the average of the seventh data and the target valve opening, and using the average as the new target valve opening.

4. The method for automatically adjusting the gas flow of a gas generator according to claim 3, characterized in that: After executing the step of calculating the degree of association and before executing the step of determining the degree of association, the method further includes: Second modeling: Obtain valve degradation data and establish a degradation model based on the degradation data. The calculation model of the degradation model is as follows: ； in, is the actual measured gas flow; is the theoretical maximum gas flow rate; is the degradation coefficient; t is the service time of the valve; Updating the correlation degree: updating the correlation degree based on the degradation coefficient, and using the updated correlation degree as a new correlation degree.

5. The method for automatically adjusting the gas flow of a gas generator according to claim 1, characterized in that: After executing the step of data collection and before executing the step of first modeling, the method further includes: Quantity judgment: judge whether the number of training samples in the training set is greater than a preset threshold. If so, execute the first modeling step; if not, execute the step of expanding the training set; Expand the training set: Based on the training set, use the generative adversarial network to generate new training samples and add the new training samples to the training set.

6. The method for automatically adjusting the gas flow of a gas generator according to claim 5, characterized in that: The step of expanding the training set further includes: Calculate similarity: Calculate the similarity between the new training sample and the training sample in the original sample set; Similarity judgment: judge whether the similarity is greater than a preset threshold. If so, do not retain the new training sample; if not, retain the new training sample.

7. A gas generator gas flow automatic regulation system, the system being used to execute the method according to any one of claims 1 to 6, characterized in that: include: The data acquisition module includes a first acquisition unit and a second acquisition unit; The first acquisition unit is used to collect historical gas flow, historical pressure data of the gas generator, and historical valve opening, and integrate the historical gas flow, historical pressure data of the gas generator, and historical valve opening into a training set; The second acquisition unit is used to collect the target gas flow rate and the current pressure of the gas generator; A first modeling module is used to establish a control model, wherein the control model includes a BI-LSTM sub-model; A model training module, which is in communication with the first acquisition unit and the first modeling module, and is used to train the BI-LSTM sub-model using the training set to obtain a trained BI-LSTM sub-model; The first prediction module is in communication with the second acquisition unit and the model training module, and is used to input the target gas flow rate and the current pressure of the gas generator into the trained BI-LSTM sub-model to obtain the target valve opening; The regulating module is in communication with the first prediction module and is configured to regulate the current valve opening to a target valve opening.

Images