A simulation analysis-based pressure reducing valve optimization method

Abstract

This invention relates to the field of simulation analysis, and more particularly to a pressure relief valve optimization method based on simulation analysis. The method includes: sorting candidate participants according to data coverage to obtain a candidate sequence; determining whether to perform secondary screening of participants based on the initial screening quantity; determining a compensation setting method based on the data deviation of anchored participants; when the compensation setting method is a mixture of participant compensation based on deviation compensation value and selected evaluation value, determining the deviation compensation ratio based on the difference in data deviation; when the compensation setting method is a rotation of participant compensation based on deviation compensation value and selected evaluation value, determining the optimization state based on the amount of anchored data, and determining the rotation method as either baseline gradient rotation or adaptive gradient rotation based on the optimization state; and optimizing the participant data selection method according to actual conditions to improve simulation analysis efficiency.

Description

Technical Field

[0001] This invention relates to the field of simulation analysis, and in particular to a method for optimizing pressure reducing valves based on simulation analysis. Background Technology

[0002] As hydraulic systems develop towards higher pressure, larger flow, and greater intelligence, the steady-state accuracy and dynamic response indicators of pressure reducing valves continue to improve. The traditional "experiment-correction" iteration cycle can no longer meet the stringent requirements of OEMs for development cycle and cost. In recent years, optimization methods based on CFD and system-level simulation have gradually matured, enabling structure-parameter optimization to be completed in a virtual environment, significantly shortening the number of prototype iterations. However, the simulation accuracy heavily depends on the realism of the boundary conditions. But the actual operating data of pressure reducing valves is scattered among multiple parties, and due to data confidentiality, the parties cannot directly share the original data. Therefore, technicians use federated learning technology for simulation optimization.

[0003] Chinese Patent Publication No. CN113962401B discloses a federated learning system, a feature selection method and apparatus within the federated learning system. The system includes at least a first participant and a second participant. The first participant holds a first feature and a sample label, while the second participant holds a second feature. The first participant is configured to: calculate a first distance between each sample in the joint sample set and the central sample based on the first feature, and obtain a second distance between each sample in the joint sample set and the central sample calculated by the second participant based on the second feature; update the first distance based on the second distance to obtain a comprehensive distance between each sample in the joint sample set and the central sample; filter out first similar samples and first dissimilar samples based on the sample label and the comprehensive distance; update the weight of the first feature and filter based on the first feature. However, this technical solution suffers from the following problems: in multi-round federated learning simulations, it cannot optimize the participant data selection method according to actual conditions, and a single selection method easily leads to simulation analysis efficiency failing to meet user needs. Summary of the Invention

[0004] To address this, the present invention provides a pressure reducing valve optimization method based on simulation analysis, which overcomes the problem that existing simulations for multi-round federated learning cannot optimize the selection of participant data according to actual conditions, and that a single selection method easily leads to simulation analysis efficiency that fails to meet user needs.

[0005] To achieve the above objectives, this invention provides a simulation-based method for optimizing a pressure-reducing valve, comprising:

[0006] The candidate participants are sorted according to data coverage to obtain a candidate sequence, and a secondary screening of participants is determined based on data coverage.

[0007] The compensation setting method is determined based on the data deviation of the anchoring participants;

[0008] When the compensation setting method is a combination of compensation based on deviation compensation value and selection of evaluation value for participating parties, the deviation compensation ratio is determined based on the difference in data deviation.

[0009] When the compensation setting method is to perform participant compensation based on the deviation compensation value and the selected evaluation value, the optimization state is determined based on the anchored data volume, and the rotation method is determined as either baseline gradient rotation or adaptive gradient rotation according to the optimization state.

[0010] Furthermore, the candidate participants are sorted in descending order of data coverage to obtain the candidate sequence;

[0011] If the number of participants in the first screening is greater than or equal to the preset required number, a second screening of participants will be conducted.

[0012] If the number of initial screenings is less than the preset required number, then the preset required number of candidate participants within the candidate sequence are selected in descending order of data coverage and recorded as anchor participants.

[0013] Furthermore, a second screening of participants is conducted, including:

[0014] Select the candidate participants whose data coverage is greater than the preset data coverage and select the preset number of candidate participants in descending order of coverage balance. These are called anchored participants.

[0015] Furthermore, the compensation setting method is determined based on the data deviation of the anchoring participants;

[0016] If the data deviation is less than the preset data deviation, the compensation setting method is to rotate the compensation for participants based on the deviation compensation value and the selected evaluation value.

[0017] If the data deviation is greater than or equal to the preset data deviation, the compensation participation method is to compensate the participants by combining the deviation compensation value and the selected evaluation value.

[0018] Furthermore, when compensating participants based on both the deviation compensation value and the selected evaluation value, the deviation compensation ratio is determined based on the difference in data deviation.

[0019] The deviation compensation ratio is positively correlated with the difference in data deviation.

[0020] Furthermore, for any candidate participant, the deviation compensation value is positively correlated with the number of discriminative data points.

[0021] For any candidate participant, its selection evaluation value is positively correlated with the number of times the candidate participant has participated.

[0022] Furthermore, when rotating participants to compensate based on deviation compensation values ​​and selection evaluation values, participant compensation is performed once every few learning rounds, and a single participant compensation uses deviation compensation values ​​or selection evaluation values ​​to select participants.

[0023] Furthermore, the optimization state is determined based on the anchored data volume, and the rotation method is determined based on the optimization state;

[0024] If the optimized state is that the anchored data volume is less than or equal to the preset anchored data volume, then the rotation method is the baseline gradient rotation;

[0025] If the optimized state is that the anchored data volume is greater than the preset anchored data volume, then the rotation method is adaptive gradient rotation.

[0026] Furthermore, in adaptive gradient rotation, gradient transformation parameters are determined based on the amount of anchored data;

[0027] The gradient transformation parameters are positively correlated with the amount of anchored data.

[0028] Compared with the prior art, the beneficial effect of the present invention is that the technical solution of the present invention has a secondary screening of participants, which further improves the screening accuracy of participants while ensuring data coverage, thereby making the screening participants more suitable for subsequent simulation learning, avoiding the problem of low screening accuracy caused by screening participants based solely on data coverage, and further improving the efficiency of simulation learning.

[0029] Furthermore, in this invention, the compensation setting method is determined based on the data deviation of the anchored participants. Under the premise of ensuring that the participants are selected based on data coverage, different compensation setting methods are selected based on the adaptability of the data deviation of the anchored participants. This makes the selection of compensation setting methods more in line with the actual application scenario and further improves the efficiency of subsequent simulation learning.

[0030] Furthermore, in this invention, the deviation compensation value for any participant is determined based on the number of distinguishing data. The distinguishing data enables the calculation of the deviation compensation value while protecting the privacy and security of the specific data values, thereby improving data security during the simulation analysis process.

[0031] Furthermore, in this invention, the optimization state is determined based on the anchored data volume, and the rotation method is determined according to the optimization state. The adaptive selection of the rotation method improves the compensation accuracy of the participants and avoids the problem of data oscillation caused by premature compensation by the participants, thereby further improving the optimization efficiency of this invention. Attached Figure Description

[0032] Figure 1 This is a schematic diagram of the pressure reducing valve optimization method based on simulation analysis according to the present invention;

[0033] Figure 2This is a flowchart illustrating the method for determining compensation settings based on the data deviation of anchoring participants in this invention.

[0034] Figure 3 This is a flowchart illustrating how the rotation method is determined based on the optimized state according to the present invention. Detailed Implementation

[0035] To make the objectives and advantages of the present invention clearer, the present invention will be further described below with reference to embodiments; it should be understood that the specific embodiments described herein are merely for explaining the present invention and are not intended to limit the present invention.

[0036] Preferred embodiments of the present invention will now be described with reference to the accompanying drawings. Those skilled in the art should understand that these embodiments are merely illustrative of the technical principles of the present invention and are not intended to limit the scope of protection of the present invention.

[0037] It should be noted that in the description of this invention, the terms "upper", "lower", "left", "right", "inner", "outer", etc., which indicate directions or positional relationships, are based on the directions or positional relationships shown in the accompanying drawings. This is only for the convenience of description and is not intended to indicate or imply that the device or element must have a specific orientation, or be constructed and operated in a specific orientation. Therefore, it should not be construed as a limitation of this invention.

[0038] Please see Figures 1 to 3 As shown, this invention provides a simulation-based method for optimizing a pressure-reducing valve, comprising:

[0039] The candidate participants are sorted according to data coverage to obtain a candidate sequence, and a secondary screening of participants is determined based on data coverage.

[0040] The compensation setting method is determined based on the data deviation of the anchoring participants;

[0041] When the compensation setting method is a combination of compensation based on deviation compensation value and selection of evaluation value for participating parties, the deviation compensation ratio is determined based on the difference in data deviation.

[0042] When the compensation setting method is to perform participant compensation based on the deviation compensation value and the selected evaluation value, the optimization state is determined based on the anchored data volume, and the rotation method is determined as either baseline gradient rotation or adaptive gradient rotation according to the optimization state.

[0043] This invention is applied to the optimization of pressure-reducing valves. Users solicit participants through centralized federated learning, simulating the pressure-reducing valve and performing several rounds of learning. Each round of learning includes: each participant independently training its model using its local data; each participant uploading its locally trained parameters to a central server; the central server integrating the uploaded parameters using a specific aggregation algorithm to generate a global model; each participant receiving the global model and training it again based on its local data. Multiple rounds of learning are performed until the global model reaches convergence or completes a predetermined number of iterations. Participants include, but are not limited to, experimental centers, component suppliers, and production centers. Each participant possesses several local data entries, each containing several data items, including but not limited to inlet pressure, outlet pressure, flow rate, and temperature. Users can perform virtual experiments to optimize the final model. The specific number of learning rounds, whether the global model has reached convergence, and the predetermined number of iterations are all set by the user according to actual needs. This is content already known to those skilled in the art and will not be elaborated upon here. The specific problem solved by this invention is that in simulations of multi-round federated learning, it is impossible to optimize the selection of participant data based on actual conditions. A single selection method easily leads to simulation analysis efficiency failing to meet user needs, thereby improving subsequent learning efficiency and model accuracy, and ultimately improving optimization efficiency.

[0044] The present invention utilizes several historical records. Each historical record records at least once the data coverage, initial screening quantity, data deviation, and anchor data quantity during the historical process. Each historical record is also set with a qualified mark, which indicates whether the historical record meets the user's requirements. The user can determine whether the historical record meets the user's requirements based on the simulation accuracy. It is understood that using self-defined indicators to determine whether the historical record meets the user's requirements is a skill that is already known to those skilled in the art, and will not be elaborated here.

[0045] Specifically, the candidate participants are sorted in descending order of data coverage to obtain the candidate sequence;

[0046] If the number of participants in the first screening is greater than or equal to the preset required number, a second screening of participants will be conducted.

[0047] If the number of initial screenings is less than the preset required number, then the preset required number of candidate participants within the candidate sequence are selected in descending order of data coverage and recorded as anchor participants.

[0048] The number of candidate participants whose data coverage exceeds the preset data coverage is recorded as the initial screening number.

[0049] Among them, the candidate participants are those not currently counted as anchor participants. For a single candidate participant, the corresponding data coverage is confirmed as follows: the user sets the required data range. For a single piece of local data of the candidate participant, if all data in the local data is within the required data range, then the local data is counted as covered data. The number of local data is the total number of local data pieces of the candidate participant. The required data range is several ranges corresponding to the local data required by the user. For example, for inlet pressure, the user sets the required maximum and minimum values ​​for inlet pressure (the maximum and minimum values ​​are set by the user, who can record the maximum and minimum values ​​of inlet pressure in the offline experiment as the required maximum and minimum values ​​for inlet pressure). The maximum and minimum values ​​are evenly divided into several intervals. The number of intervals is set by the user. The greater the user's requirements for the amount of local data and the accuracy of analysis, the greater the number of intervals. One possible value for the number of intervals is 4.

[0050] Users can set the preset data coverage and preset required quantity values ​​according to their actual scenarios. It is understandable that the greater the user's need for local data coverage, the larger the preset data coverage value will be; the greater the user's acceptance of the number of anchor participants, the larger the preset required quantity value will be. A method for setting these values ​​is provided: extract the data coverage and initial screening quantity corresponding to the historical records that meet the user's needs, remove outliers from the data coverage and initial screening quantity respectively, and record the average values ​​of the data coverage and initial screening quantity after removing outliers as the preset data coverage and preset required quantity, respectively. The methods for removing outliers include, but are not limited to, the 3σ criterion method or the IQR method.

[0051] Specifically, the secondary screening of participants includes:

[0052] Select the candidate participants whose data coverage is greater than the preset data coverage and select the preset number of candidate participants in descending order of coverage balance. These are called anchored participants.

[0053] For a single candidate participant, the coverage balance is determined by detecting the coverage count of each demand data range corresponding to the candidate participant. The absolute value of the difference between the maximum and minimum coverage counts is recorded as the coverage balance. For a single demand data range, the corresponding coverage count is the total number of data items in the coverage data of the candidate participant that are within that demand data range.

[0054] Specifically, the compensation setting method is determined based on the data deviation of the anchoring participants;

[0055] If the data deviation is less than the preset data deviation, the compensation setting method is to rotate the compensation for participants based on the deviation compensation value and the selected evaluation value.

[0056] If the data deviation is greater than or equal to the preset data deviation, the compensation participation method is to compensate the participants by combining the deviation compensation value and the selected evaluation value.

[0057] The data deviation P of the anchoring participants is calculated as follows:

[0058]

[0059] Where Ki is the average number of required data ranges for each historical request corresponding to the i-th anchor participant, K0 is the number of required data ranges for the current user, M is the total number of anchor participants, and for a single anchor participant, its corresponding single historical request is federated learning in the historical process of that anchor participant.

[0060] The preset data deviation value can be set by the user according to the actual scenario. It can be understood that the data deviation value reflects the similarity between the anchor participant's historical learning process and the current user's needs. The greater the user's need for similarity, the smaller the preset data deviation value. When the data deviation value is greater than the preset data deviation value, it reflects that the similarity between the anchor participant's historical learning process and the current user's needs is lower than the user's need for similarity. Therefore, a hybrid approach of using deviation compensation value and selected evaluation value is adopted for participant compensation, and the deviation compensation ratio is determined based on the difference in data deviation values, making the deviation compensation ratio more in line with the actual scenario. Based on the extracted data deviation values ​​corresponding to the historical records that meet the user's needs, outliers in the data deviation values ​​are removed, and the average value of the data deviation values ​​after removing outliers is recorded as the preset data deviation value.

[0061] Specifically, when compensating participants based on a combination of deviation compensation values ​​and selected evaluation values, the deviation compensation ratio is determined based on the difference in data deviation.

[0062] The deviation compensation ratio is positively correlated with the difference in data deviation.

[0063] Data deviation difference = preset data deviation - data deviation, deviation compensation ratio = benchmark ratio + L × data deviation difference, where L is the adjustment coefficient. The values ​​of the benchmark ratio and the adjustment coefficient can be set by the user according to their needs. The greater the user's need for data deviation compensation, the larger the values ​​of the benchmark ratio and L. One benchmark ratio and L value is provided: benchmark ratio = 50%, L = 2. The unit of deviation compensation ratio is %, and the deviation compensation ratio is an integer rounded up with a maximum value of 100.

[0064] When participant compensation is performed using a hybrid approach based on deviation compensation value and selection evaluation value, participant compensation is conducted once every X rounds of learning, where X is the preset number of learning rounds. In each participant compensation round, participants are selected in descending order based on deviation compensation value and are designated as anchor participants. The number of anchor participants selected based on deviation compensation value = deviation compensation percentage × compensation quantity. The number of anchor participants selected based on deviation compensation value is a rounded-up integer. Participants are then selected in ascending order based on selection evaluation value and are designated as anchor participants. The number of anchor participants selected based on selection evaluation value = compensation quantity - number of anchor participants selected based on deviation compensation value. The preset number of learning rounds and compensation quantity are set by the user. The greater the user's demand for participant compensation, the smaller the preset number of learning rounds and the larger the compensation quantity. One preset number of learning rounds and compensation quantity is provided: preset number of learning rounds = 3, compensation quantity = 10.

[0065] Specifically, for any candidate participant, the deviation compensation value is positively correlated with the number of discriminative data points.

[0066] For any candidate participant, its selection evaluation value is positively correlated with the number of times the candidate participant has participated.

[0067] The difference data is the number of compensation data for the candidate participant. If the number of data items in a candidate participant's local data that are within the range of required data is 50% to 80% (inclusive) of the total number of data items in the local data, then this local data is recorded as the difference data. The deviation compensation value = the number of difference data / the preset number of difference data.

[0068] The evaluation value is calculated as the number of times a candidate participant has participated, which is the total number of times the candidate participant has been identified as an anchor participant in the past.

[0069] Users can set the preset values ​​for the number of distinguishing data and the preset number of participations according to their actual application scenarios. It can be understood that the greater the user's demand for compensation from the participants, the greater the value of the preset number of distinguishing data, and the preset number of participations is the average number of participations of the candidate participants.

[0070] Specifically, when rotating participant compensation based on bias compensation value and selection evaluation value, participant compensation is performed once every few learning rounds. Each participant compensation selects participants in descending order of bias compensation value or ascending order of selection evaluation value. The participants selected for compensation are denoted as anchor participants. The number of learning rounds between each participant compensation is gradient ascent. The number of learning rounds between the Cth participant compensation and the (C+1)th participant compensation is X + g × C, where g × C is an integer rounded up, and c is the gradient transformation parameter.

[0071] Specifically, the optimization state is determined based on the amount of anchored data, and the rotation method is determined based on the optimization state;

[0072] If the optimized state is that the anchored data volume is less than or equal to the preset anchored data volume, then the rotation method is the baseline gradient rotation;

[0073] If the optimized state is that the anchored data volume is greater than the preset anchored data volume, then the rotation method is adaptive gradient rotation.

[0074] The anchor data volume is the total number of local data entries of the anchor participants. The preset anchor data volume is set by the user. The shorter the user's requirement for data processing time, the smaller the preset anchor data volume. A method for determining the preset anchor data volume is provided, which extracts the anchor data volume corresponding to the historical records that meet the user's needs, and records the average value of the anchor data volume after removing outliers as the preset anchor data volume.

[0075] In the baseline gradient rotation, g is set to 1; in the adaptive gradient rotation, the gradient transformation parameters are determined based on the amount of anchored data.

[0076] The gradient transformation parameters are positively correlated with the amount of anchored data.

[0077] Gradient transformation parameter = anchor data size / preset anchor data size. The gradient transformation parameter is an integer rounded up.

[0078] The technical solution of the present invention has been described above with reference to the preferred embodiments shown in the accompanying drawings. However, it will be readily understood by those skilled in the art that the scope of protection of the present invention is obviously not limited to these specific embodiments. Without departing from the principles of the present invention, those skilled in the art can make equivalent changes or substitutions to the relevant technical features, and the technical solutions after these changes or substitutions will all fall within the scope of protection of the present invention.

[0079] The above description is merely a preferred embodiment of the present invention and is not intended to limit the invention. Various modifications and variations can be made to the present invention by those skilled in the art. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the scope of protection of the present invention.

Claims

1. A method for optimizing a pressure reducing valve based on simulation analysis, characterized in that, include: The candidate participants are sorted according to data coverage to obtain a candidate sequence, and a second screening of participants is determined based on the number of participants in the first screening. The candidate participants are those not currently counted as anchor participants. For a single candidate participant, the corresponding data coverage is confirmed as follows: the user sets the required data range, and for a single piece of local data of the candidate participant, if all data items in the local data are within the required data range, the local data is counted as covered data. The number of local data items is the total number of local data items of the candidate participants, and the required data range is the range of local data items corresponding to the user's needs. The participants include experimental centers, component suppliers, and production centers. Each participant has local data, and a single piece of local data includes several data items, including inlet pressure, outlet pressure, flow rate, and temperature. The compensation setting method is determined based on the data deviation of the anchoring participants; the data deviation reflects the similarity between the historical learning process of the anchoring participants and the current user needs; When the compensation setting method is to compensate participants using a combination of deviation compensation value and selected evaluation value, the deviation compensation ratio is determined based on the difference in data deviation. Deviation compensation value = number of distinguishing data / preset number of distinguishing data, where distinguishing data refers to the number of compensation data for the candidate participants. Selected evaluation value = number of participations / preset number of participations. Deviation compensation ratio = benchmark ratio + adjustment coefficient × data deviation difference, data deviation difference = preset data deviation - data deviation; When the compensation setting method is to perform participant compensation based on the deviation compensation value and the selected evaluation value, the optimization state is determined based on the anchored data volume, and the rotation method is determined according to the optimization state as either baseline gradient rotation or adaptive gradient rotation. For each participant compensation, participants are selected in descending order of deviation compensation value or ascending order of selected evaluation value. The participants selected for compensation are denoted as anchored participants. The number of learning rounds between each participant compensation is the gradient ascent. The number of learning rounds between the Cth participant compensation and the (C+1th)th participant compensation is = X + g × C, where X is the preset number of learning rounds, C is the number of participant compensation rounds, and g × C is an integer rounded up. In baseline gradient rotation, g is set to 1. In adaptive gradient rotation, the gradient transformation parameter is determined based on the anchored data volume. Gradient transformation parameter = anchored data volume / preset anchored data volume.

2. The pressure reducing valve optimization method based on simulation analysis according to claim 1, characterized in that, The candidate participants are sorted in descending order of data coverage to obtain the candidate sequence; If the number of participants in the first screening is greater than or equal to the preset required number, a second screening of participants will be conducted. If the number of initial screenings is less than the preset required number, then the preset required number of candidates within the candidate sequence are selected in descending order of data coverage and recorded as anchored candidates.

3. The pressure reducing valve optimization method based on simulation analysis according to claim 2, characterized in that, Secondary screening of participants, including: The selected participants, whose data coverage is greater than the preset data coverage, are selected in descending order of coverage balance. The number of participants with the preset demand is recorded as the anchor participants. For a single candidate participant, the coverage balance is determined by detecting the number of times each demand data range is covered. The absolute value of the difference between the maximum and minimum number of coverage is recorded as the coverage balance. For a single demand data range, the corresponding number of coverage is the total number of data items in the covered data of the candidate participant that are within the demand data range.

4. The pressure reducing valve optimization method based on simulation analysis according to claim 3, characterized in that, The compensation setting method is determined based on the data deviation of the anchoring participants; If the data deviation is less than the preset data deviation, the compensation setting method is to rotate the compensation for participants based on the deviation compensation value and the selected evaluation value. If the data deviation is greater than or equal to the preset data deviation, the compensation participation method is to compensate the participants by combining the deviation compensation value and the selected evaluation value.

5. The pressure reducing valve optimization method based on simulation analysis according to claim 4, characterized in that, When compensating participants based on a combination of deviation compensation values ​​and selected evaluation values, the deviation compensation ratio is determined based on the difference in data deviation. The deviation compensation ratio is positively correlated with the difference in data deviation.

6. The pressure reducing valve optimization method based on simulation analysis according to claim 4, characterized in that, For any candidate participant, the deviation compensation value is positively correlated with the number of discrepancies. For any candidate participant, its selection evaluation value is positively correlated with the number of times the candidate participant has participated.

7. The pressure reducing valve optimization method based on simulation analysis according to claim 1, characterized in that, The optimization state is determined based on the amount of anchored data, and the rotation method is determined based on the optimization state. If the optimized state is that the anchored data volume is less than or equal to the preset anchored data volume, then the rotation method is the baseline gradient rotation; If the optimized state is that the anchored data volume is greater than the preset anchored data volume, then the rotation method is adaptive gradient rotation.

8. The pressure reducing valve optimization method based on simulation analysis according to claim 1, characterized in that, In adaptive gradient rotation, gradient transformation parameters are determined based on the amount of anchored data. The gradient transformation parameters are positively correlated with the amount of anchored data.

Images