Axial flow pump, self-learning method and regulating method

Abstract

This invention relates to an axial flow pump, a self-learning method, and an adjustment method, relating to the technical field of axial flow pumps, and includes the following steps: Step S1: Determine the target performance point; Step S2: Based on the target performance point, search for adjacent working intervals in the database, and solve for the theoretical installation angle θ based on the adjacent working intervals. * Step S3: Install angle θ according to theory * The blade angle is adjusted. This invention has the advantage of enabling more precise angle adjustment.

Description

Technical Field

[0001] This invention relates to the technical field of axial flow pumps, and in particular to an axial flow pump, a self-learning method, and an adjustment method. Background Technology

[0002] Traditional axial flow pumps typically have blade angles that are either fixed (non-adjustable) or adjustable (hydraulic or electric). However, hydraulic adjustment is costly and difficult to maintain, making it unsuitable for small and medium-sized axial flow pumps. Electric adjustment only provides simple angle settings. Summary of the Invention

[0003] To address the shortcomings of existing technologies, one of the objectives of this invention is to provide an axial flow pump, a self-learning method, and an adjustment method, which have the advantage of enabling more precise angle adjustment.

[0004] The above-mentioned objective of the present invention is achieved through the following technical solution:

[0005] A method for adjusting an axial flow pump includes the following steps:

[0006] Step S1: Determine the target performance point;

[0007] Step S2: Based on the target performance point, search for adjacent working intervals in the database, and solve for the theoretical installation angle θ based on the adjacent working intervals. * ;

[0008] Step S3: Install angle θ according to theory * Adjust the blade angle.

[0009] In a preferred embodiment, the present invention can be further configured such that, in step S3, the formula is used: = ,in, For the final installation angle; X * θ represents the target performance point; X represents the measured performance of the sensor; K represents the corrected gain; and θ represents the target performance point. * The theoretical installation angle.

[0010] In a preferred embodiment, the present invention can be further configured as follows: In step S2, m installation angle values ​​are selected from the database, such that the difference between the value corresponding to the installation angle and the target performance point meets the standard, and the theoretical installation angle θ is solved using the m installation angle values. * .

[0011] In a preferred embodiment, the present invention can be further configured such that: during the solution process, multiple iterations are performed, and each iteration provides a new angle estimate, θ. * N This represents the angle estimate obtained in the Nth iteration. Then the target installation angle θ is output. * N , where ε is the set convergence threshold.

[0012] This invention also discloses a self-learning method for axial flow pumps, comprising the following steps:

[0013] Step L1: Collect hydraulic performance data of the pump under different operating conditions and classify them;

[0014] Step L2: Remove outlier data;

[0015] Step L3: Use the data processed in step L2 to solve for the theoretical installation angle θ. * Model updates;

[0016] Step L4: Use an error verification mechanism to verify the updated characteristic relationships, and only write the data into the formal database when the error meets the threshold.

[0017] In a preferred embodiment, the present invention may be further configured to include step L5: periodically organizing, cleaning, and reconstructing the database.

[0018] In a preferred embodiment, the present invention can be further configured as follows: In step L3, when an interpolation node model is used, let the existing nodes in the database be (θ). m ,η m Q m H m The newly collected valid data points are (θ) i ,η i Q i H i If the distance d between the newly collected data point and the existing sample is:

[0019] d=((θ i -θ m ) 2 +(η i -η m ) 2 +(Q i -Q m ) 2 +(H i -H m ) 2 ) 1 / 2

[0020] like:

[0021] d min >β

[0022] If the newly collected data point is in an area not covered by the database, it needs to be added to the database; otherwise, it does not need to be added. Here, β is the distance threshold. If the distance of the newly added data point from an existing node is greater than β, then the new data point is added.

[0023] In a preferred embodiment, the present invention can be further configured such that, in step L4, the error acquisition method is to re-predict the target parameter M using the updated model. pred The actual traffic flow at the current moment is M. c The error is:

[0024] e M =|M pred -M c |

[0025] like:

[0026] e M / M c < 5%

[0027] If the error judgment condition is met, the sample is written into the formal database; otherwise, the sample is discarded.

[0028] In a preferred embodiment, the present invention may be further configured as follows: step L5 includes the following steps: cleaning up redundant nodes in the database, merging adjacent nodes in the database, and periodically updating the model based on the latest cleaned data.

[0029] The present invention also discloses an axial flow pump, which employs the above-described adjustment method for an axial flow pump and / or the above-described self-learning method for an axial flow pump.

[0030] The present invention has the following advantages:

[0031] The pump's hydraulic characteristics are digitized (database), and the optimal blade angle is automatically calculated based on real-time operating data. Precise adjustment is achieved through motor drive, and it has self-learning capabilities, making it more and more accurate the longer it runs. Attached Figure Description

[0032] Figure 1 This is a schematic diagram of the structure of the present invention.

[0033] Figure 2 This is a schematic diagram of the blade mounting angle of the present invention.

[0034] Figure 3 This is the flow chart of the angle adjustment control of the present invention.

[0035] Reference numerals in the attached diagram: 1. Blade; 2. Adjustment motor; 3. Impeller hub; 4. Mounting base. Detailed Implementation

[0036] The present invention will be further described in detail below with reference to the accompanying drawings.

[0037] Reference Figures 1-3 The present invention discloses an axial flow pump adjustment method, comprising the following steps:

[0038] Step S1: Determine the target performance point;

[0039] Step S2: Based on the target performance point, search for adjacent working intervals in the database. Based on the adjacent working intervals, select m installation angle values ​​from the database, ensuring that the difference between the installation angle value and the target performance point meets the standard. Solve for the theoretical installation angle θ using the m installation angle values. * During the solution process, multiple iterations will be performed, and each iteration will provide a new angle estimate, θ. * N Let |θ| represent the angle estimate obtained in the Nth iteration. * N − θ * N−1 If | < ε, then the target installation angle θ will be output. * N , where ε is the set convergence threshold;

[0040] Step S3: Install angle θ according to theory * The blade angle is adjusted using the following formula: = ,in, For the final installation angle; X * θ represents the target performance point; X represents the measured performance of the sensor; K represents the corrected gain; and θ represents the target performance point. * The theoretical installation angle.

[0041] More specifically:

[0042] Step S1: Determine the target performance point

[0043] According to the control mode (target flow Q) * Target lift H * Target efficiency η * (etc.), determine the target performance point:

[0044] If it is a target traffic mode: the target point is Q. * With flow error e Q = Q*-Q is used as the target for subsequent optimization and adjustment.

[0045] If it is target head mode: the target point is H * With head error e H = H*-H serves as the target for subsequent optimization and adjustment.

[0046] If it is the optimal efficiency mode: monitor the maximum efficiency η in real time. max This will serve as the target for subsequent optimization and adjustment.

[0047] Step S2: Reverse lookup of blade angle from database

[0048] Based on the target performance point (Q) * H * or η * The following calculation process is performed in the feature database:

[0049] ① Search adjacent work areas

[0050] The installation angle θ is selected from the database using the minimum error criterion, such that |Q i -Q * | or |H i -H * |or|η i -η * | The minimum number of m points (the number m can be set), where the subscript i represents the i-th feature data point (θ) in the database. i ,η i Q i H i That is, if the target performance point is Q... * The installation angle θ is selected from the database using the minimum error criterion so that |Q i -Q * The minimum number of m points (the number m can be set).

[0051] ② Interpolation calculation or fitting solution

[0052] The target installation angle θ is calculated using the following three-dimensional interpolation model. * :

[0053]

[0054] in f Methods that can be used include: three-dimensional linear interpolation, spline interpolation, polynomial fitting of surfaces, and small neural network regression model NN(Q, H, n); η is the current efficiency.

[0055] ③ Convergence Criterion

[0056] Iterative calculations are required during interpolation or fitting, and each iteration provides a new angle estimate, denoted as θ. * N This represents the angle estimate obtained in the Nth iteration. If the solution is considered convergent, the target installation angle θ is output. * N, where ε is the set convergence threshold (e.g., 0.1°).

[0057] Step S3: Correction

[0058] The theoretical target blade angle θ was calculated. * Then, based on the deviation between the actual operating data and the target performance point, the angle is fine-tuned in real time and dynamically to make the final angle closer to the actual optimal value. Correction is performed when there is a deviation between the actual operating point and the database prediction.

[0059]

[0060] in The corrected target angle (i.e., the final installation angle); X * For the target performance point (e.g., Q) * H * , η max X represents the actual measured performance of the sensor (such as Q, H, η); K represents the correction gain, which can be fixed or adaptively adjusted according to the operating conditions.

[0061] This invention also discloses a self-learning method for axial flow pumps, comprising the following steps:

[0062] Step L1: Collect hydraulic performance data of the pump under different operating conditions and classify them;

[0063] Step L2: Remove outlier data;

[0064] Step L3: Use the data processed in step L2 to solve for the theoretical installation angle θ. * For model updates, when using an interpolation node model, let the existing nodes in the database be (θ). m ,η m Q m H m The newly collected valid data points are (θ) i ,η i Q i H i If the distance d between the newly collected data point and the existing sample is:

[0065] d=((θ i -θ m ) 2 +(η i -η m ) 2 +(Q i -Q m ) 2 +(H i -H m ) 2 )1 / 2

[0066] like:

[0067] d min >β

[0068] If the newly collected data points are in areas not covered by the database, they need to be added to the database; otherwise, they do not need to be added. Here, β is the distance threshold. If the distance of the new data point from an existing node is greater than β, then the new data point is added.

[0069] Step L4: The updated characteristic relationships are verified using an error verification mechanism. Data is only written to the formal database when the error meets the threshold. The error is obtained by re-predicting the target parameter M using the updated model. pred The actual traffic flow at the current moment is M. c The error is:

[0070] e M =|M pred -M c |

[0071] like:

[0072] e M / M c < 5%

[0073] If the error judgment condition is met, the sample is written into the formal database; otherwise, the sample is discarded.

[0074] Step L5: Regularly clean, refactor, and reorganize the database, including the following steps: clean up redundant nodes in the database, merge adjacent nodes in the database, and periodically update the model based on the latest cleaned data.

[0075] More specifically:

[0076] Step L1: Real-time sampling of runtime data

[0077] As needed, one or more of the actual hydraulic performance data (flow rate Q, head H, power P, speed n, installation angle θ) of the pump under different operating conditions are collected in real time, and these data are used as "samples" to update the database (self-learning).

[0078] Collect data sets (Q,H,P,N,θ) and store them in a temporary buffer according to time windows (e.g., 10 s).

[0079] Step L2: Outlier Removal

[0080] The following methods were used to identify unstable conditions and invalid data:

[0081] ① Removal based on fluctuation amplitude (removal when operating conditions are unstable):

[0082] If the data fluctuates too much within a sampling window (e.g., 10 seconds), the data in that window is considered invalid.

[0083] (X max -X min ) / μ X > δ

[0084] Where X is a performance parameter (such as Q, H, η), μ X δ is the average value of this parameter within the sampling window; δ is the set fluctuation threshold.

[0085] ② Elimination of blade angle dynamic change stage (velocity > threshold):

[0086] When the blades are rotating and the angle changes rapidly, it can cause instability in data such as flow rate and head, so it is necessary to reject such data.

[0087] The judgment is made by the blade angular velocity.

[0088] (θ t -θ t-1 ) / Δt>ω

[0089] Then it is considered to be in an angle-adjusting transient state, and all sampled data at that moment are discarded (Q). t H t ,P t ,n t ,θ t ).

[0090] Where θ t Install the angle θ at the current sampling time. t-1 Δt is the installation angle at the previous sampling time; Δt is the time difference between adjacent sampling times; ω is a given threshold.

[0091] Step L3: Incremental update of the database

[0092] The filtered, valid data is added to the database incrementally, and online updates are performed on the interpolation nodes or fitted models. One of the following three self-learning mechanisms can be used:

[0093] ① Increase the increment of interpolation nodes

[0094] Suppose the database already has nodes (θ) m ,η m Q m H m The newly collected valid data points are (θ) i , η i Q i H iIf the distance d between the newly collected data point and the existing sample is:

[0095] d=((θ i -θ m ) 2 +(η i -η m ) 2 +(Q i -Q m ) 2 +(H i -H m ) 2 ) 1 / 2

[0096] like:

[0097] d min >β

[0098] If the newly collected data points are from areas not covered by the database, they need to be added to the database; otherwise, they do not need to be added. Here, β is the distance threshold.

[0099] If the distance of the new data point from an existing node is greater than β, then the new data point is added.

[0100] ②Online updating of the fitted model

[0101] Recursive least squares (RLS) is used for updating.

[0102] ③ Neural network fine-tuning

[0103] Mini-batch training is performed based on the running data to update the neural network regression model.

[0104] Step L4: Verification and Writeback

[0105] An error verification mechanism is used to validate the updated characteristic relationships (model), and the data is only written to the formal database when the error meets a threshold. The updated model is validated as follows, taking flow rate as an example (head and efficiency are similar), using the updated model to re-predict the flow rate Q at the current time. pred The actual flow rate at the current moment is Q. c The error is:

[0106] e Q =|Q pred -Q c |

[0107] like:

[0108] e Q / Q c < 5%

[0109] If the error judgment condition is met, the data points are written to the formal database; otherwise, the samples are discarded and the model before the update is restored.

[0110] Step L5: Long-term learning strategies

[0111] Regularly clean, refactor, and reorganize the database to maintain high accuracy, low redundancy, and high stability during long-term operation, preventing model degradation over time. Specifically, after writing new feature data points in step L3, the following operations are performed, mainly including:

[0112] ① Clean up redundant nodes in the database

[0113] If multiple very close data points appear in the database, i.e. when d < a given threshold, these points are considered "duplicates," and only the one with the highest quality (e.g., the one with the smallest error e) is retained.

[0114] ② Merge adjacent nodes in the database

[0115] When multiple nodes are excessively densely packed in the database (e.g., many points are collected at a certain angle), calculate their center point parameters (θ', η', Q', H'):

[0116] θ'=(θ1+θ2+θ3…+θ M ) / M

[0117] η'=(η1+η2+η3…+η M ) / M

[0118] Q'=(Q1+Q2+Q3…+θ M ) / M

[0119] H'=(H1+H2+H3…+θ M ) / M

[0120] Replace the surrounding adjacent nodes with this center point. Where M is the number of adjacent points.

[0121] ③ Model Reconstruction

[0122] Every fixed interval (e.g., 24 hours), the model is refitted and updated using the method described in L3 based on the latest cleaned data.

[0123] The present invention also discloses an axial flow pump, which adopts the above-mentioned adjustment method and / or self-learning method of an axial flow pump, including an impeller hub 3, a mounting base 4, an adjustment motor 2 and blades 1. The adjustment motor 2 has a self-locking function. The blades 1 are mounted on the mounting base 4. The adjustment motor 2 is mounted in the impeller hub 3. The adjustment motor 2 is used to drive the mounting base 4 to rotate, thereby driving the blades 1 to rotate, so as to adjust the mounting angle θ.

[0124] The implementation principle of this embodiment is as follows: the hydraulic characteristics of the pump are digitized (database), the optimal blade angle is automatically calculated based on real-time operating data, and precise adjustment is achieved through motor drive. It has self-learning ability, making it more and more accurate the longer it runs.

[0125] The embodiments described herein are preferred embodiments of the present invention and are not intended to limit the scope of protection of the present invention. Therefore, all equivalent changes made in accordance with the structure, shape, and principle of the present invention should be covered within the scope of protection of the present invention.

Claims

1. A method for regulating an axial flow pump, characterized in that: Includes the following steps: Step S1: Determine the target performance point; Step S2: Based on the target performance point, search for adjacent working intervals in the database, and solve for the theoretical installation angle θ based on the adjacent working intervals. * ; Step S3: Install angle θ according to theory * Adjust the blade angle; In step S2, m installation angle values ​​are selected from the database, ensuring that the difference between the performance point corresponding to the installation angle and the target performance point meets the standard. The theoretical installation angle θ is then calculated using the m installation angle values. * ; In step S3, the formula is used: ,in, For the final installation angle; X * X represents the target performance point; K represents the measured performance point of the sensor; θ represents the corrected gain. * The theoretical installation angle.

2. The axial flow pump adjustment method according to claim 1, characterized in that: The solution process involves multiple iterations, each iteration providing a new angle estimate, θ. * N This represents the angle estimate obtained in the Nth iteration. Then the output θ * N The theoretical installation angle is ε, where ε is the set convergence threshold.

3. The axial flow pump adjustment method according to claim 1, characterized in that: It also includes a self-learning step, which includes the following steps: Step L1: Collect hydraulic performance data of the pump under different operating conditions and classify them; Step L2: Remove outlier data; Step L3: Use the data processed in step L2 to solve for the theoretical installation angle θ. * Model updates; Step L4: Use an error verification mechanism to verify the updated characteristic relationships, and only write the data into the formal database when the error meets the threshold.

4. The axial flow pump adjustment method according to claim 3, characterized in that: It also includes step L5: regularly organizing, cleaning and reconstructing the database.

5. A method for adjusting an axial flow pump according to claim 3, characterized in that: In step L3, when an interpolation node model is used, let the existing nodes in the database be (θ). m ,η m Q m H m The newly collected valid data points are (θ) i ,η i Q i H i If the distance d between the newly collected data point and the existing sample is: d=((θ i -θ m ) 2 +(the i -or m ) 2 +(Q i -Q m ) 2 +(H i -H m ) 2 ) 1 / 2 like: d min >β If the newly collected data points are in areas not covered by the database, they need to be added to the database; otherwise, they do not need to be added. Here, β is the distance threshold, and d... min Let d be the minimum value. If the distance of the new data point from the existing node is greater than β, then the new data point is added.

6. A method for adjusting an axial flow pump according to claim 3, characterized in that: In step L4, the error is obtained by re-predicting the target parameter M using the updated model. pred The actual traffic flow at the current moment is M. c The error is: e M =|M pred -M c | like: have been M / M c < 5% If the error judgment condition is met, the sample is written into the formal database; otherwise, the sample is discarded.

7. The axial flow pump adjustment method according to claim 4, characterized in that: Step L5 includes the following steps: cleaning up redundant nodes in the database, merging adjacent nodes in the database, and periodically updating the model based on the latest cleaned data.

8. An axial flow pump, characterized in that: An axial flow pump adjustment method as described in any one of claims 1-7 is adopted.

Images