A parameter recommendation method for key factors affecting solid rocket engine performance

Abstract

The present application discloses a parameter recommendation method for key influencing factors of solid rocket engine performance. The method is based on a solid rocket engine (SRM) internal ballistic performance prediction neural network model, and includes three steps: solid rocket engine design parameter item influence calculation, solid rocket engine design main parameter item screening and solid rocket engine design parameter optimization. The influence of each input parameter of the model on the prediction result is obtained by parameter item influence calculation; the parameters with significant influence on the result are found by parameter item screening; and the parameter value combination that meets the performance index requirements is obtained by heuristic search through genetic algorithm.

Description

Technical Field

[0001] The invention relates to a parameter recommendation method for key influencing factors of solid rocket engine performance, and belongs to the technical field of practical application of machine learning. Background Art

[0002] In the field of research and development and design of solid rocket engines, performance indicators such as the engine's internal ballistic pressure-time curve (pt curve) are important evaluation indicators for engine design. In the past, the design process of solid rocket engines was usually that designers used empirical models to obtain the corresponding design parameters based on the performance index requirements of the target engine, and then used the obtained design parameters to make rockets for performance tests. If the performance meets the requirements, the test is successful; otherwise, the reasons are analyzed based on the differences between the actual performance indicators and the target performance indicators, the empirical model is corrected, and new process parameters are obtained for the next round of tests. This process often requires multiple rounds of iterations to design an engine that meets the requirements, which consumes a lot of manpower and material resources.

[0003] In recent years, with the rise and application of deep learning and neural networks in the field of data mining, the many successful applications of neural networks in industrial design, industrial data analysis and other fields have attracted more and more attention. In the field of solid rocket engine design, the internal ballistic performance prediction model based on the neural network model has achieved a high degree of accuracy. The formation of auxiliary design functions based on the neural network model is expected to significantly reduce the number of test rounds and shorten the design cycle.

[0004] However, there is still a deviation between the internal ballistic performance predicted by the model and the actual test performance. Therefore, in order to shorten the design cycle of auxiliary engine design and reduce the number of test rounds, further analysis is conducted to find out the influencing factors of the deviation, and on this basis, reasonable engine parameter values ​​are recommended, which becomes a new goal pursued by the current auxiliary design of solid rocket engines. However, due to the poor interpretability of the neural network model, it is difficult to directly determine the influencing factors that cause the deviation, and it is even more difficult to adjust these influencing factors accordingly, resulting in the inability to provide direct support to designers. Therefore, on the basis of the neural network model for solid rocket engine performance prediction based on neural networks, finding a method for analyzing influencing factors and recommending reasonable engine parameter values ​​is an urgent problem to be solved in order to achieve more efficient auxiliary engine design. Specifically, there are three key problems that need to be solved:

[0005] (1) How to evaluate the impact of relevant factors in engine design on engine performance;

[0006] (2) How to screen out the factors with the greatest impact and form a combination of key factors;

[0007] (3) How to adjust the parameter values ​​of key influencing factors to find parameter values ​​that meet performance requirements. Summary of the invention

[0008] In order to solve the deficiencies in the existing solid rocket engine design process, the present invention provides a design parameter recommendation method for a solid rocket engine internal ballistic performance prediction neural network model. For a specific solid rocket engine performance prediction model, the present invention can evaluate the influence and screen the input parameters, eliminate the interference items in the input parameters, retain the input parameters that have a greater impact on the results, and perform parameter optimization on this basis, providing powerful guidance for parameter adjustment in the design and improving the efficiency of the engine design process.

[0009] In order to achieve the above objectives, the present invention adopts the following technical solution: a parameter recommendation method for key influencing factors of solid rocket engine performance, the method comprising the following steps:

[0010] S1: Calculation of the influence of design parameters of solid rocket motors, using MIV to evaluate the influence of influencing factors;

[0011] S2: Screening of main parameters of solid rocket engine design;

[0012] S3: Optimization of main design parameters of solid rocket engines.

[0013] Step S1, for an existing solid rocket engine performance neural network prediction model, the present invention uses MIV (Mean Impact Value) to evaluate the influence of influencing factors. In view of the data characteristics in the field of solid rocket engine design, the present invention improves the calculation method of MIV and uses the improved MIV value to measure the importance of influencing factors to the model.

[0014] Step S2, sort the influencing factors from large to small according to the obtained MIV, and divide them from small to large to obtain multiple TOP-K influencing factor combinations according to the number of influencing factors contained in the prediction model. Then, train a solid rocket engine performance prediction model for each influencing factor combination obtained by division, and test it on the test set to obtain the prediction accuracy of the pressure curve segmentation. The number of influencing factors in the combination with the highest prediction accuracy of the influencing factor combination is recorded as parameter_max, and the influencing factors are divided into two parts according to parameter_max. The influencing factors with MIV ranking greater than parameter_max are irrelevant influencing factors; the influencing factors with MIV ranking less than parameter_max are relevant influencing factors.

[0015] Step S3: Using a genetic algorithm to search for prediction models corresponding to all relevant influencing factors, obtain a parameter value combination that meets the engine performance requirements.

[0016] Wherein, step S1 is specifically as follows:

[0017] Step S11, respectively transform the influencing factor matrix Each influencing factor is increased or decreased by 10% according to the difference between the maximum and minimum values ​​of the variable, and a 2×n influencing factor change matrix is ​​obtained.

[0018]

[0019] Where n is the number of influencing factors in the model, and m is the amount of data samples in the data set.

[0020] Step S12: input the 2×n matrices as input into the interior ballistic pressure performance curve prediction model to obtain 2×n×m predicted pressure curves.

[0021] Step S13, calculating the MIV value of each influencing factor according to the improved MIV calculation formula.

[0022] Among them, the MIV value is used in step S1 to measure the influence of the influencing factors on the model. The MIV calculation adopts a floating ratio, and the formula is as follows:

[0023] X i,up =X i +(X i,max -X i,min )×10% (1)

[0024] X i,down =X i -(X i,max -X i,min )×10% (2)

[0025] Where X i,up is the result after the i-th variable is floated, X i,down is the result after the i-th variable is reduced, X i,max is the maximum value of the i-th variable in the training set, X i,min is the minimum value of the i-th variable in the training set;

[0026] At the same time, the prediction result in this example is not a single value but the entire pressure curve, and the impact cannot be simply characterized by positive or negative, such as Figure 2 In this case, although the change of variable A has a much greater impact on the prediction curve than variable B, the traditional MIV value calculation result is that the MIV of variable A is much smaller than the MIV of variable B. Therefore, the present invention improves the calculation of the MIV value, and the formula is as follows:

[0027]

[0028]

[0029] Among them, MIV i is the average influence value of the i-th variable, IV i (j) is the influence value of the i-th variable on the j-th sample, where y ij,up,t is the prediction result of the pressure value of the jth training sample at time t after the i-th variable is increased by 10%, y ij,down,t It is the prediction result of the pressure value of the jth training sample at time t after the i-th variable decreases by 10%.

[0030] Among them, step S2 is to perform parameter screening, and the process is as follows Figure 3 As shown, including:

[0031] Step S21, sorting the influencing factors from large to small according to the obtained MIV, and dividing them from small to large to obtain multiple TOP-K influencing factor combinations according to the number of influencing factors included in the prediction model, wherein the TOP-K influencing factor combination refers to a set of influencing factors with the top k influencing factor MIV ranking in the model;

[0032] Step S22, respectively train the prediction model of the interior ballistic performance curve segment with each combination of the influencing factors obtained by the division. This step needs to ensure that the training rounds of each model are the same to avoid the difference in model accuracy due to different training rounds. Then, the prediction accuracy of the interior ballistic performance curve segment is obtained by testing on the test set. The prediction accuracy obtained at this time can be used as the confidence of the influencing factor combination for the interior ballistic performance curve segment.

[0033] Step S23, as the percentage of the influencing factor combination to the total number of influencing factors decreases, the prediction accuracy of the pressure-time curve segment, i.e., the confidence level, first increases and then decreases, and the model with the highest confidence level is found;

[0034] In step S24, the number of influencing factors in the combination with the highest confidence level of the influencing factor combination is recorded as parameter_max. According to parameter_max, the influencing factors are divided into two parts. One part is the influencing factors whose MIV ranking in the model is greater than parameter_max, which are irrelevant influencing factors; the other part is the influencing factors whose MIV ranking is less than parameter_max, which are the main relevant influencing factors, and the relevant influencing factors that have a greater impact on the model results are obtained after screening.

[0035] Among them, step S3, parameter optimization, the present invention uses genetic algorithm (GA) to find the optimal parameter combination. Each individual in the population in the algorithm is the result of encoding all the relevant influencing factors (i.e., the main influencing factors) previously screened. The process is as follows Figure 4 As shown, specifically including:

[0036] Step S31, the initial individuals in the population are coded as several sets of influencing factors, which include the design scheme to be optimized and the influencing factors obtained by continuing to randomly modify the design scheme;

[0037] Step S32, the fitness of the individual is predicted by an interior ballistic performance prediction model that takes all relevant influencing factors as inputs. The closer the predicted result is to the target, the higher the fitness;

[0038] Step S33, in the process of individual selection, in addition to screening out individuals with low fitness, individuals whose process parameters are out of the range or adjusted too much among the adjusted individuals are also screened out;

[0039] Step S34, the algorithm stops when all individuals in the population have reached the target or the number of iterations has been reached, and outputs all individuals in the population that have reached the optimization target at this time;

[0040] Step S35: At the end of the model, multiple solutions that meet the optimization goal will be output. At this time, these solutions are sorted. The methods with a smaller number of adjustment parameters and a smaller adjustment range will be recommended with a higher priority. The priority of a solution is as follows:

[0041]

[0042] Among them, Priority i is the priority of the ith solution, x k is the value of the kth parameter in the scheme, x k0 is the value of the kth parameter in the initial scheme to be optimized. After calculating the priorities of all schemes, the parameter value combination with the highest priority is the recommended design parameter.

[0043] Compared with the prior art, the advantages of the present invention are as follows: after testing on a typical solid rocket engine model, 4 to 9 key influencing factors can be discovered under 20 to 60 parameter conditions, and parameter values ​​that meet performance requirements can be recommended, thereby improving the efficiency of designers in optimizing parameters and finding out error influencing factors. BRIEF DESCRIPTION OF THE DRAWINGS

[0044] Figure 1 It is a schematic diagram of the overall process of the present invention;

[0045] Figure 2A schematic diagram of the process of calculating the influence of parameters of the present invention;

[0046] Figure 3 A schematic diagram of the process of parameter screening of the present invention;

[0047] Figure 4 A schematic diagram of the process of optimizing parameter values ​​of the present invention. DETAILED DESCRIPTION

[0048] The present application will be further described below by taking a specific neural network model (solid rocket motor interior ballistic performance curve prediction model) as an example in conjunction with the accompanying drawings.

[0049] Example 1: See Figure 1 —- Figure 4 , a parameter recommendation method for key influencing factors of solid rocket engine performance, Figure 1 The overall process of the present invention is shown; the present invention uses the MIV value to measure the influence of the influencing factors on the model. The MIV calculation adopts a floating ratio, and the formula is as follows:

[0050] X i,up =X i +(X i,max -X i,min )×10% (1)

[0051] X i,down =X i -(X i,max -X i,min )×10% (2)

[0052] Where X i,up is the result after the i-th variable is floated, X i,down is the result after the i-th variable is reduced, X i,max is the maximum value of the i-th variable in the training set, X i,min is the minimum value of the i-th variable in the training set.

[0053] At the same time, the prediction result in this example is not a single value but the entire pressure curve, and the impact cannot be simply characterized by positive or negative, such as Figure 2 In this case, although the change of variable A has a much greater impact on the prediction curve than variable B, the traditional MIV value calculation result is that the MIV of variable A is much smaller than the MIV of variable B. Therefore, the present invention improves the calculation of the MIV value, and the formula is as follows:

[0054]

[0055]

[0056] Among them, MIVi is the average influence value of the i-th variable, IV i (j) is the influence value of the i-th variable on the j-th sample, where y ij,up,t is the prediction result of the pressure value of the jth training sample at time t after the i-th variable is increased by 10%, y ij,down,t It is the prediction result of the pressure value of the jth training sample at time t after the i-th variable decreases by 10%.

[0057] Step S1: Use MIV to evaluate the influence of influencing factors. The process is as follows: Figure 2 Shown include:

[0058] Step S11, respectively transform the influencing factor matrix Each influencing factor is increased or decreased by 10% according to the difference between the maximum and minimum values ​​of the variable, and a 2×n influencing factor change matrix is ​​obtained.

[0059]

[0060] Where n is the number of influencing factors in the model, and m is the amount of data samples in the data set.

[0061] Step S12: input the 2×n matrices as input into the interior ballistic pressure performance curve prediction model to obtain 2×n×m predicted pressure curves.

[0062] Step S13, calculating the MIV value of each influencing factor according to the improved MIV calculation formula.

[0063] Step S2, perform parameter screening, the process is as follows Figure 3 As shown, including:

[0064] Step S21, sort the influencing factors from large to small according to the obtained MIV, and divide them from small to large according to the number of influencing factors included in the prediction model to obtain multiple TOP-K influencing factor combinations. The TOP-K influencing factor combination refers to the set of influencing factors with the top k influencing factor MIV ranking in the model.

[0065] Step S22, train the prediction model of the interior ballistic performance curve segment respectively with each combination of influencing factors obtained by division. This step needs to ensure that the training rounds of each model are the same, so as to avoid the difference in model accuracy due to different training rounds. Then, test on the test set to obtain the prediction accuracy of the interior ballistic performance curve segment, and the prediction accuracy obtained at this time can be used as the confidence of the influencing factor combination for the interior ballistic performance curve segment.

[0066] Step S23, as the percentage of the influencing factor combination to the total number of influencing factors decreases, the prediction accuracy of the pressure-time curve segment, that is, the confidence level, first increases and then decreases, and the model with the highest confidence level is found.

[0067] Step S24, the number of influencing factors in the combination with the highest confidence of the influencing factor combination is recorded as parameter_max, and the influencing factors are divided into two parts according to parameter_max, one part is the influencing factors whose MIV ranking in the model is greater than parameter_max, which are irrelevant influencing factors; the other part is the influencing factors whose MIV ranking is less than parameter_max, which are mainly relevant influencing factors. The relevant influencing factors that have a greater impact on the model results are obtained after screening.

[0068] Step S3, parameter optimization, the present invention uses a genetic algorithm (GA) to find the optimal parameter combination. Each individual in the population in the algorithm is the result of encoding all the relevant influencing factors (i.e., the main influencing factors) previously screened. The process is as follows Figure 4 As shown, including:

[0069] Step S31, the initial individuals in the population are coded as several sets of influencing factors, which include the design scheme to be optimized and the influencing factors obtained by continuing to randomly modify it.

[0070] In step S32, the fitness of the individual is predicted by an interior ballistic performance prediction model that takes all relevant influencing factors as input. The closer the predicted result is to the target, the higher the fitness.

[0071] Step S33, in the process of individual selection, in addition to screening out individuals with low fitness, individuals whose process parameters are out of the value range or adjusted too much are also screened out.

[0072] Step S34, the algorithm will stop after all individuals in the population have reached the target or the number of iterations has been reached, and output all individuals in the population that have reached the optimization target at this time.

[0073] Step S35: At the end of the model, multiple solutions that meet the optimization goal will be output, and these solutions will be ranked. The methods with a smaller number of adjustment parameters and a smaller adjustment range will be recommended with a higher priority. The priority of a solution is as follows:

[0074]

[0075] Among them, Priority i is the priority of the ith solution, x k is the value of the kth parameter in the scheme, x k0is the value of the kth parameter in the initial optimization scheme. After calculating the priorities of all schemes, the parameter value combination with the highest priority is the recommended design parameter.

[0076] In the description of this specification, the description with reference to the terms "one embodiment", "some embodiments", "example", "specific example", or "some examples" etc. means that the specific features, structures, materials or characteristics described in conjunction with the embodiment or example are included in at least one embodiment or example of the present invention. In this specification, the schematic representations of the above terms do not necessarily refer to the same embodiment or example. Moreover, the specific features, structures, materials or characteristics described may be combined in any one or more embodiments or examples in a suitable manner. In addition, those skilled in the art may combine and combine the different embodiments or examples described in this specification and the features of the different embodiments or examples, without contradiction.

[0077] Although the embodiments of the present invention have been shown and described above, it is to be understood that the above embodiments are exemplary and are not to be construed as limitations of the present invention. A person skilled in the art may change, modify, replace and vary the above embodiments within the scope of the present invention.

Claims

1. A parameter recommendation method for key factors affecting the performance of solid rocket engines. It is characterized in that The method comprises the following steps: S1: Calculation of the influence of design parameters of solid rocket motors, using MIV to evaluate the influence of influencing factors; S2: Screening of main parameters of solid rocket engine design; S3: Optimization of main design parameters of solid rocket motors; In step S1, the MIV value is used to measure the influence of the influencing factors on the model. The MIV calculation adopts a floating ratio, and the formula is as follows: X i,up =X i +(X i,max -X i,min )×10% (1) X i,down =X i -(X i,max -X i,min )×10% (2) Where X i,up is the result after the i-th variable is floated, X i,down is the result after the i-th variable is reduced, X i,max is the maximum value of the i-th variable in the training set, X i,min is the minimum value of the i-th variable in the training set; The calculation of the MIV value is improved, and the formula is as follows: Among them, MIV i is the average influence value of the i-th variable, IV i (j) is the influence value of the i-th variable on the j-th sample, where y ij,up,t is the prediction result of the pressure value of the jth training sample at time t after the i-th variable is increased by 10%, y ij,down,t is the prediction result of the pressure value of the jth training sample at time t after the i-th variable decreases by 10%; Step S2, performing parameter screening, including: Step S21, sorting the influencing factors from large to small according to the obtained MIV, and dividing them from small to large to obtain multiple TOP-K influencing factor combinations according to the number of influencing factors included in the prediction model, wherein the TOP-K influencing factor combination refers to a set of influencing factors with the top k influencing factor MIV ranking in the model; Step S22, respectively train the prediction model of the interior ballistic performance curve segment with each combination of the influencing factors obtained by the division. This step needs to ensure that the training rounds of each model are the same to avoid the difference in model accuracy due to different training rounds. Then, the prediction accuracy of the interior ballistic performance curve segment is obtained by testing on the test set. The prediction accuracy obtained at this time can be used as the confidence of the influencing factor combination for the interior ballistic performance curve segment. Step S23, as the percentage of the influencing factor combination to the total number of influencing factors decreases, the prediction accuracy of the pressure-time curve segment, i.e., the confidence level, first increases and then decreases, and the model with the highest confidence level is found; In step S24, the number of influencing factors in the combination with the highest confidence level of the influencing factor combination is recorded as parameter_max. According to parameter_max, the influencing factors are divided into two parts. One part is the influencing factors whose MIV ranking in the model is greater than parameter_max, which are irrelevant influencing factors; the other part is the influencing factors whose MIV ranking is less than parameter_max, which are the main relevant influencing factors, and the relevant influencing factors that have a greater impact on the model results are obtained after screening.

2. The method for recommending parameters of key influencing factors of solid rocket engine performance according to claim 1, It is characterized in that Step S1 is specifically as follows: Step S11, respectively transform the influencing factor matrix Each influencing factor is increased or decreased by 10% according to the difference between the maximum and minimum values ​​of the variable, and a 2×n influencing factor change matrix is ​​obtained. Where n is the number of influencing factors in the model, and m is the amount of data samples in the data set; Step S12, taking the 2×n matrices as inputs, and inputting them into the interior ballistic pressure performance curve prediction model to obtain 2×n×m predicted pressure curves; Step S13, calculating the MIV value of each influencing factor according to the improved MIV calculation formula.

3. The method for recommending parameters of key influencing factors of solid rocket engine performance according to claim 1, It is characterized in that Step S3, parameter optimization, specifically includes: Step S31, the initial individuals in the population are coded as several sets of influencing factors, including the design scheme to be optimized and the influencing factors obtained by continuing to randomly modify the design scheme; Step S32, the fitness of the individual is predicted by an interior ballistic performance prediction model that takes all relevant influencing factors as inputs. The closer the predicted result is to the target, the higher the fitness; Step S33, in the process of individual selection, in addition to screening out individuals with low fitness, individuals whose process parameters are out of the range or adjusted too much among the adjusted individuals are also screened out; Step S34, the algorithm stops when all individuals in the population have reached the target or the number of iterations has been reached, and outputs all individuals in the population that have reached the optimization target at this time; Step S35: At the end of the model, multiple solutions that meet the optimization goal will be output. At this time, these solutions are sorted. The methods with a smaller number of adjustment parameters and a smaller adjustment range will be recommended with a higher priority. The priority of a solution is as follows: Among them, Priority i is the priority of the ith solution, x k is the value of the kth parameter in the scheme, x k0 is the value of the kth parameter in the initial scheme to be optimized. After calculating the priorities of all schemes, the parameter value combination with the highest priority is the recommended design parameter.

Images