A parallel fan balance control method based on machine learning

Abstract

The present invention is a parallel fan balance control method based on machine learning, comprising the following steps: step 1, obtaining historical data of the parallel fan; step 2, constructing a machine learning model, training the machine learning model through historical data, and obtaining the parameters of the machine learning model; step 3, correcting the observed data of the air duct pressure through the parameters of the machine learning model; step 4, predicting the target frequency of the parallel fan using the parameters of the machine learning model, the corrected air duct pressure and the current operating data; step 5, adjusting the parallel fan according to the target frequency of the parallel fan until the target frequency with equivalent output is reached. The present invention can continuously and automatically perform balance control on the parallel fan; saves labor and improves the efficiency during operation; at the same time, it can quantitatively and accurately perform balance control on the parallel fan, and improve the efficiency of the fan operation.

Description

Technical Field

[0001] The present invention relates to the technical field of fan control, and in particular to a parallel fan balancing control method based on machine learning. Background Art

[0002] In engineering, the hardware of parallel fans will use products of the same model from the same manufacturer, and the parameters of the fan pipelines and equipment are exactly the same or close. The existing control method of parallel fans is to manually perform startup control on the two fans at startup to avoid "wind grabbing" or even fan reversal. During the operation of the parallel fans, if "wind grabbing" occurs, the normal fan's rotor blades should be manually opened wide, and the other fan should be manually closed to the minimum. Wait until the air main pipe pressure returns to normal and all parameters are stable before restarting the parallel operation. After starting the parallel fans, the rotor blade baffle should be opened to a certain degree to prevent reverse wind. To maintain the stable wind pressure of the parallel fans in operation, manual small operations are performed to ensure that the wind pressure changes slowly, relying on the level and experience of the operating personnel.

[0003] In the prior art, the patent application number CN201811600062.0, "A method and system for automatically paralleling primary fans for thermal power generators", provides an automatic paralleling method for the start-up of a primary fan. The disadvantage is that only the current is used to determine whether the paralleling is successful, and there is no balance control during operation. The patent application number CN201410126530.0, "A control method for eliminating surge of axial flow fans", has the disadvantage of narrow scenario applicability, and requires the operator to cooperate with the program to perform multiple manual operations during operation. The existing technology can provide a bottom-line guarantee for the safety of parallel fans, prevent major accidents and hidden dangers, and can smoothly control the start-up and shutdown of parallel fans, and keep the wind pressure basically stable during operation. However, the control scenarios of these technologies are limited, covering only the paralleling and surge scenarios at startup, and lacking the scenarios of control during operation; at the same time, manual operation leads to low control efficiency, coarse granularity of controlling fans, and cannot be continuously controlled automatically. At the same time, the automatic parallel method only has blowers, lacks a parallel control method for induced draft fans, and has no quantitative method for balancing control of parallel fans, making it difficult to maintain the optimal parallel efficiency of parallel fans. Summary of the invention

[0004] The present invention proposes a universal and efficient parallel fan balancing control method based on machine learning, so that the output of the parallel fans can be continuously and effectively automatically balanced during the operation of the fans after they are started.

[0005] The present invention provides a parallel fan balance control method based on machine learning, the method comprising the following steps:

[0006] Step 1, obtaining historical data of parallel fans;

[0007] Step 2: Build a machine learning model, train the machine learning model with historical data, and obtain the parameters of the machine learning model;

[0008] Step 3, correcting the observed data of the air duct pressure through the parameters of the machine learning model;

[0009] Step 4, predicting the target frequency of the parallel fans using the parameters of the machine learning model, the corrected air duct pressure, and the current operating data;

[0010] Step 5, adjusting the parallel fans according to the target frequencies of the parallel fans until the target frequencies with equivalent outputs are reached.

[0011] Furthermore, in step 1, the parallel blower includes two identical pipes, and the volume flow rate of the pipes is:

[0012]

[0013] Among them, Q A represents the volume flow rate of pipe A, Q B represents the volume flow rate of pipe B, Q represents the total air volume in the air duct; α=(u A S) 2 , β=(u B S) 2 ; F A Indicates the actual duct pressure of pipe A, F B Indicates the actual duct pressure of pipe B; u A represents the flow coefficient of pipe A, u B represents the flow coefficient of pipe B, u A ≈u B ; S represents the cross-sectional area of ​​the pipe.

[0014] Furthermore, in step 2, the machine learning model includes first to third machine learning models when the actual current of the pipeline can be obtained, wherein the first machine learning model is:

[0015]

[0016] Among them, I A Indicates the actual fan current of pipeline A, I B Indicates the actual fan current of pipeline B;

[0017] The second machine learning model and the third machine learning model are:

[0018]

[0019] Among them, λ A1 represents the first parameter of pipeline A, λ A2represents the second parameter of pipeline A, λ A3 represents the third parameter of pipeline A, λ B1 represents the first parameter of pipeline B, λ B2 represents the second parameter of pipeline B, λ B3 Indicates the third parameter of pipeline B; f A Indicates the actual frequency of the fan in pipeline A; f B Indicates the actual frequency of the fan in duct B.

[0020] Furthermore, in step 2, when the machine learning model cannot obtain the actual pipeline current, a fourth machine learning model is constructed using the least squares method:

[0021]

[0022] Among them, λ A4 The fourth parameter of pipeline A, λ A5 The fifth parameter of pipeline A, λ B4 The fourth parameter of pipeline B, λ B5 Indicates the fifth parameter of pipeline B; f A Indicates the actual frequency of the fan in duct A, f B represents the actual frequency of the fan in duct B, and min T means minimizing the expression.

[0023] Furthermore, in step 3, when the machine learning model is able to obtain the actual current of the pipeline, the observed data of the air duct pressure is corrected by the actual current:

[0024]

[0025] in, represents the observed data of the air duct pressure in pipe A, Represents the observed data of the air duct pressure in pipe B.

[0026] Furthermore, in step 3, when the machine learning model cannot obtain the actual current of the pipeline, the observation number of the air duct pressure is corrected by parameters α and β:

[0027]

[0028] in, represents the observed data of the air duct pressure in pipe A, Represents the observed data of the air duct pressure in pipe B.

[0029] Furthermore, in step 4, the machine learning model predicts the target frequency of the fan when the actual pipeline current is available:

[0030]

[0031] Among them, f' A represents the fan target frequency of duct A, f' B Indicates the target frequency of the fan in duct B.

[0032] Furthermore, in step 4, the target frequency of the wind turbine is predicted:

[0033]

[0034] Among them, Q A =Q B =Q / 2, f' A represents the fan target frequency of duct A, f' B Indicates the target frequency of the fan in duct B.

[0035] Furthermore, in step 5, the target frequency of the parallel fans is predicted and the difference is made between the current actual frequency of the parallel fans. The fan with the larger absolute value of the difference starts to adjust and is adjusted close to the target frequency. Then the other fan follows the adjustment. The two fans are gradually adjusted in multiple steps according to the configuration parameters, gradually reaching the target frequency with equivalent output.

[0036] The beneficial effects achieved by the present invention are:

[0037] The present invention is applicable to a wide range of scenarios, and is applicable to all scenarios during the operation of parallel fans, including during startup and stable operation.

[0038] The present invention can continuously and automatically perform balancing control on parallel fans, which has the positive effect of saving manpower and improving the efficiency during operation; at the same time, it can quantitatively and accurately perform balancing control on parallel fans, which has the positive effect of improving the efficiency of fan operation. BRIEF DESCRIPTION OF THE DRAWINGS

[0039] Figure 1 A flow chart of a parallel fan balancing control method based on machine learning provided in an embodiment of the present invention;

[0040] Figure 2 A logical schematic diagram of parallel fans in a parallel fan balancing control method based on machine learning provided in an embodiment of the present invention. DETAILED DESCRIPTION

[0041] The technical solution of the present invention is described in more detail below in conjunction with the accompanying drawings. The present invention includes but is not limited to the following embodiments.

[0042] As attached Figure 1 As shown, the present invention proposes a parallel fan balance control method based on machine learning, comprising the following steps:

[0043] Step 1, obtaining historical data of parallel fans;

[0044] Step 2: Build a machine learning model, train the machine learning model with historical data, and obtain the parameters of the machine learning model;

[0045] Step 3, correcting the observed data of the air duct pressure through the parameters of the machine learning model;

[0046] Step 4, predicting the target frequency of the parallel fans using the parameters of the machine learning model, the corrected air duct pressure, and the current operating data;

[0047] Step 5: Adjust the parallel fans according to the predicted parallel fan frequency until the target frequency with equivalent output is reached.

[0048] In the following examples, as shown in the attached Figure 2 As shown in the figure, the logic diagram of parallel fans is abstracted into two pipes A and B, with a connecting pipe C in the middle. The same liquid flows in the pipes. One direction is from the pipe to the main pipe D, and the other is from the main pipe D to the two pipes in the opposite direction. In engineering, the equipment on pipes A and B are the same type of products manufactured by the same manufacturer, with the same or very close equipment parameters, and the special case of A or B failure is not considered. The pressure F of the two pipes is known. A and F B , the total flow rate of the main pipe Q, the frequency of the fan equipment on pipes A and B f A 、f B Or opening K A , K B , so that the output of the two pipelines is equal, the frequency or opening is actually used to control the flow in the pipeline. According to the definition of engineering, the output is equal to the flow rate Q in the pipeline A and Q B The values ​​are the same and this goal is achieved by adjusting the frequency of the fan equipment.

[0049] The present invention is further explained below in conjunction with the accompanying drawings and specific embodiments.

[0050] The embodiments in which the parallel fans are blowers (such as primary and secondary fans) or induced draft fans are described in detail as follows.

[0051] Embodiment 1:

[0052] In this embodiment, historical data from the past week is used, updated once a week, remodeled, and predicted using the new model.

[0053] Step 1: Obtain the historical data of the following parameters for the past week: the actual frequency f of the two wind turbines A and f B, the actual duct pressure F of the two fan equipment A and F B , total air volume in the duct Q, actual current of the two fans I A and I B .

[0054] The known fan air volume model is:

[0055]

[0056] For pipes A and B, according to the “pipeline flow-pressure relationship”, there is the following relationship:

[0057]

[0058]

[0059] Among them, Q A represents the volume flow rate of pipe A, Q B represents the volume flow rate of pipe B; u represents the flow coefficient, which is related to the valve, fluid material and pipe shape. Once these are determined, the coefficient will be determined accordingly; S represents the cross-sectional area of ​​the pipe; F AB Indicates the pressure before the main pipe D.

[0060] According to the law of conservation of mass: Q = Q A +Q B , so we can get the system of equations:

[0061]

[0062] In general, since the parameters of pipelines A and B and the equipment are exactly the same or very close, the flow coefficients u of pipelines A and B are considered to be equal if the failure of A or B is not considered. A 、u B Very close, that is, approximately equal: u A ≈u B ,so:

[0063]

[0064]

[0065]

[0066] Get Q A , Q B , and then let α=(u A S) 2 andβ=(u B S) 2 ,have to:

[0067]

[0068] Usually there is no Q A , Q B The measuring point, α, β cannot be directly calculated. There are two different ways to solve them. In engineering, which method to use is determined based on whether there is actual current of parallel fans.

[0069] Step 2: After the actual current I can be obtained A and I B When the linear regression method is used, three machine learning models are constructed.

[0070] According to the fan air volume model, since the fan efficiency, mechanical transmission efficiency, rated power and rated current of parallel fans are the same, the uncertainties can be offset by using the air volume ratio for calculation.

[0071]

[0072] Combined with Q = Q A +Q B We can calculate:

[0073]

[0074] According to formulas (5) and (7),

[0075]

[0076] The first machine learning model is obtained by joint subtraction:

[0077]

[0078] The linear regression method without intercept was used to fit α and β.

[0079] According to (6) and (8), substitute α and β, use two linear regression models to fit the parameters of the second machine learning model and the third machine learning model, and obtain the first parameter λ of pipeline A A1 , the second parameter λ of pipeline A A2 , the third parameter λ of pipeline A A3 , the first parameter λ of pipeline B B1 , the second parameter λ of pipeline B B2 , the third parameter λ of pipeline B B3 .

[0080]

[0081] The independent variable in the second machine learning model is and F A , the dependent variable is f A; The third machine learning model independent variable is and F B , the dependent variable is f B .

[0082] If there is no actual current, a fourth machine learning model is constructed using a least squares method.

[0083] According to the fan air volume model, since the fan efficiency, mechanical transmission efficiency, rated power, and rated current of the parallel fans are the same, the actual current is proportional to the actual air volume * actual pressure:

[0084]

[0085] Because the actual fan current is equivalent to the actual frequency,

[0086]

[0087] Converted into the equation:

[0088]

[0089] Minimize the following:

[0090] min T=(λ A4 (Q A ×F A )+λ A5 -f A ) 2 +(λ B4 (Q B ×F B )+λ B5 -f B ) 2 (14)

[0091] Substituting formula (5) into formula (14), we get:

[0092]

[0093] Using the least squares method, minimizing (15), we can find the fourth parameter λ of pipeline A A4 , the fifth parameter λ of pipeline A A5 、The fourth parameter λ of pipeline B B4 、The fifth parameter λ of pipeline B B5 and α, β.

[0094] Step 3: To make the prediction result more accurate, the observed data of the air duct pressure is corrected;

[0095] If there is actual current, the observed data of the air duct pressure is corrected by the actual current so that the observed data is adjusted to the actual air duct pressure data:

[0096]

[0097] If there is no actual current, the observed data of the air duct pressure is corrected by using parameters α and β to adjust the observed data to the actual air duct pressure data:

[0098]

[0099] Step 4: The parameters of the machine learning model, the corrected duct pressure, and the current runtime data are used to predict the target frequency of the two parallel fans to obtain the fan target frequency.

[0100] For actual current, the following formula is used for prediction:

[0101]

[0102] For the case where there is no actual current, here Q A =Q B =Q / 2, use the following formula to make predictions,

[0103]

[0104] Among them, f' A represents the fan target frequency of duct A, f' B Indicates the target frequency of the fan in duct B.

[0105] Step 5, predict the target frequency of the parallel fans, and make a difference with the current actual frequency of the parallel fans. The fan with the larger absolute value of the difference starts to adjust, and is adjusted close to the target frequency. Then the other fan follows the adjustment. The two fans are gradually adjusted in multiple steps according to the configuration parameters, gradually reaching the target frequency with equivalent output.

[0106] At this point, the complete control method for parallel fan balancing has been completed. The five steps include acquiring data, establishing different models according to different situations, and predicting and adjusting the frequency of parallel fans in three major stages, which efficiently and quantitatively balances the air supply of parallel fans.

[0107] The invention is not limited to the above-mentioned specific implementation modes. A person skilled in the art can implement the invention in various other specific implementation modes according to the embodiments and the disclosure of the drawings. Therefore, any design that adopts the design structure and concept of the invention and makes some simple transformations or changes falls within the scope of protection of the invention.

Claims

1. A parallel fan balance control method based on machine learning, characterized in that: The method comprises the following steps: Step 1, obtaining historical data of parallel fans; Step 2: Build a machine learning model, train the machine learning model with historical data, and obtain the parameters of the machine learning model; Step 3: Correct the observed data of air duct pressure through the parameters of the machine learning model; Step 4, predicting the target frequency of the parallel fans using the parameters of the machine learning model, the corrected air duct pressure, and the current operating data; Step 5, adjusting the parallel fans according to the target frequency of the parallel fans until the target frequency with equivalent output is reached; In step 1, the parallel blower includes two identical pipes, and the volume flow rate of the pipes is: Among them, Q A represents the volume flow rate of pipe A, Q B represents the volume flow rate of pipe B, Q represents the total air volume in the air duct; α=(u A S) 2 , β=(u B S) 2 ; F A Indicates the actual duct pressure of pipe A, F B Indicates the actual duct pressure of pipe B; u A represents the flow coefficient of pipe A, u B represents the flow coefficient of pipe B, u A ≈u B ; S represents the cross-sectional area of ​​the pipe; In step 2, the machine learning model includes first to third machine learning models when the actual pipeline current can be obtained, wherein the first machine learning model is: Among them, I A Indicates the actual fan current of pipeline A, I B Indicates the actual fan current of pipeline B; The second machine learning model and the third machine learning model are: Among them, λ A1 represents the first parameter of pipeline A, λ A2 represents the second parameter of pipeline A, λ A3 represents the third parameter of pipeline A, λ B1 represents the first parameter of pipeline B, λ B2 represents the second parameter of pipeline B, λ B3 Indicates the third parameter of pipeline B; f A Indicates the actual frequency of the fan in pipeline A; f B Indicates the actual frequency of the fan in pipeline B; In step 2, when the machine learning model cannot obtain the actual pipeline current, the fourth machine learning model is constructed using the least squares method: Among them, λ A4 The fourth parameter of pipeline A, λ A5 The fifth parameter of pipeline A, λ B4 The fourth parameter of pipeline B, λ B5 Indicates the fifth parameter of pipeline B; f A Indicates the actual frequency of the fan in duct A, f B represents the actual frequency of the fan in pipeline B, and minT represents minimizing the expression; In step 3, when the machine learning model is able to obtain the actual current of the pipeline, it corrects the observed data of the duct pressure by the actual current: in, represents the observed data of the air duct pressure in pipe A, Represents the observed data of the air duct pressure in pipe B.

2. The parallel fan balance control method according to claim 1, characterized in that: In step 3, when the machine learning model cannot obtain the actual current of the pipeline, the observed number of the duct pressure is corrected by parameters α and β: in, represents the observed data of the air duct pressure in pipe A, Represents the observed data of the air duct pressure in pipe B.

3. The parallel fan balance control method according to claim 1, characterized in that: In step 4, the machine learning model predicts the target frequency of the fan when the actual pipeline current is available: in, represents the fan target frequency of duct A, Indicates the target frequency of the fan in duct B.

4. The parallel fan balance control method according to claim 2, characterized in that: In step 4, the target frequency of the wind turbine is predicted: Among them, Q A =Q B =Q / 2, represents the fan target frequency of duct A, Indicates the target frequency of the fan in duct B.

5. The parallel fan balance control method according to claim 1, characterized in that: In step 5, the target frequency of the parallel fans is predicted and the difference is made between the current actual frequency of the parallel fans. The fan with the larger absolute value of the difference starts to adjust and is adjusted close to the target frequency. Then the other fan follows the adjustment. The two fans are adjusted step by step according to the configuration parameters, gradually reaching the target frequency with equivalent output.

Images