Method and equipment for generating characteristic curve of centrifugal water pump and medium

Abstract

The invention provides a characteristic curve generation method and device of a centrifugal water pump and a medium. The method comprises the following steps that S1, a training set is constructed; s2, training a discriminator and a generator; s3, dual verification denoising is carried out; s4, circularly executing the training step and the de-noising step until the discriminator and the generator are completely converged; and S5, exporting a QH characteristic curve under the target frequency. According to the method, the characteristic curve of the centrifugal water pump can be generated in a complex actual production environment.

Description

Technical Field

[0001] The present invention relates to the technical field of water pumps, and in particular to a method for generating a characteristic curve of a centrifugal water pump based on data augmentation and double verification denoising. Background Art

[0002] Centrifugal water pump, also known as centrifugal water pump or centrifugal pump, is a pump that uses the centrifugal force generated by the high-speed rotation of the impeller to transport liquid. Its working principle is: when the motor drives the pump shaft and impeller to rotate, water or other liquids are thrown out under the action of centrifugal force, thereby forming a low-pressure area in the center of the impeller. The external liquid is pressed into this low-pressure area under the action of atmospheric pressure, and then thrown out by the impeller, and so on, forming a continuous liquid flow. In the actual production environment, the operating points of the centrifugal pump at a certain frequency are often concentrated or there is no data at this frequency. The distribution defects and scarcity of the data make it difficult to fit the curve. The data collected by data acquisition equipment such as flow meters will have certain errors. The worse the quality of the data acquisition equipment, the greater the error. That is, the fluctuation of the data collected by the hardware will bring noise data. Noise data under non-steady-state operation, such as when the water pump is switched on and off, the collected data often deviates from its characteristic curve. Summary of the invention

[0003] The present invention aims to at least solve the technical problems existing in the prior art, and in particular innovatively proposes a method for generating a characteristic curve of a centrifugal water pump based on data augmentation and double verification denoising.

[0004] In order to achieve the above-mentioned object of the present invention, the present invention provides a method for generating a characteristic curve of a centrifugal water pump based on data augmentation and double verification denoising, comprising the following steps:

[0005] Step 1: Construct a training set. Collect the real-time operation data of the water pump, calculate the instantaneous head of the water pump, obtain the actual operating point information of the water pump (frequency, flow, head), and generate the theoretical iso-efficiency operating point with a frequency change within ±10% and an interval of 0.1 based on the data augmentation strategy. Use the data collected in the recent period as training data.

[0006] Step 2: Train the discriminator and generator. Take the data of three dimensions, frequency, flow rate and lift, as input and whether it is real data as output, and train the discriminator of the generative adversarial network; take frequency and flow rate as input and lift as output, and train the generator of the generative adversarial network;

[0007] Step 3: Double verification denoising. Set thresholds η1 and η2, when the generator loss loss G <η1 and the loss of the discriminator DWhen <η2, double verification denoising is performed, and the discriminator and pump mathematical characteristics are used to detect and delete the noise operating points from the training set.

[0008] Step 4: Repeat the training step and denoising step until the discriminator and generator are fully converged.

[0009] Step 5: derive the QH characteristic curve at the target frequency. Set the specified frequency and flow range as needed, use the generator to generate the head, obtain a set of (flow, head) operating points at the specified frequency, and draw a smooth curve to obtain the QH characteristic curve.

[0010] The present invention also discloses a computer device, comprising a memory, a processor and a computer program stored in the memory and executable on the processor, wherein the processor implements the steps of the above method when executing the computer program.

[0011] The present invention also discloses a computer-readable storage medium on which a computer program is stored. When the computer program is executed by a processor, the steps of the above method are implemented.

[0012] In summary, due to the adoption of the above technical solution, the beneficial effects of the present invention are:

[0013] (1) The present invention can generate characteristic curves of centrifugal water pumps in complex actual production environments; (2) The generative adversarial framework ensures that the generator can accurately generate operating points that have not appeared in the history when deriving characteristic curves; (3) Data augmentation technology enriches the operating points of the pump operation, effectively improving the generalization ability and accuracy of the model. (4) Double verification denoising technology effectively eliminates noise data caused by various factors in the actual production environment.

[0014] Additional aspects and advantages of the present invention will be given in part in the following description and in part will be obvious from the following description, or will be learned through practice of the present invention. BRIEF DESCRIPTION OF THE DRAWINGS

[0015] The above and / or additional aspects and advantages of the present invention will become apparent and easily understood from the description of the embodiments in conjunction with the following drawings, in which:

[0016] Figure 1 It is a schematic block diagram of the process of the present invention. DETAILED DESCRIPTION

[0017] Embodiments of the present invention are described in detail below, examples of which are shown in the accompanying drawings, wherein the same or similar reference numerals throughout represent the same or similar elements or elements having the same or similar functions. The embodiments described below with reference to the accompanying drawings are exemplary and are only used to explain the present invention, and cannot be understood as limiting the present invention.

[0018] Combination Figure 1 , specifically including the following steps:

[0019] Step 1, build a training set. Collect the real-time operation data of the water pump, calculate the instantaneous head of the water pump, and obtain the operating point information of the actual operation of the water pump (F: frequency, Q: flow, H: head). Keep one decimal place for the pump frequency data, and generate the theoretical iso-efficiency operating point with a frequency change within ±10% and an interval of 0.1Hz based on the data augmentation strategy. Use the data collected in the recent period as training data;

[0020] For centrifugal water pumps, the pump efficiency does not change much within a certain speed adjustment range. After adjusting the speed, the performance of the centrifugal pump has the following relationship Where n is the original speed of the pump, is the changed speed, F, Q, H are the frequency, flow rate and head at the original speed, It is the frequency, flow rate and head after the speed changes. Based on the actual operation data collected, frequency data with an interval of 0.1Hz between [F*0.9, F*1.1] can be generated. And calculate the ratio Thus, the theoretical operating point (α*F, α*Q, α 2 *H). This allows a large number of operating points to be added to the training data.

[0021] The calculation method of the instantaneous head of the water pump is:

[0022]

[0023] Among them, H represents the instantaneous head of the pump;

[0024] p represents the pressure of the water pump;

[0025] α represents the pressure coefficient; its value is 102;

[0026] l indicates the installation height of the water pump pressure gauge;

[0027] M represents the liquid level elevation of the suction well;

[0028] Q represents the pump flow rate;

[0029] β represents the flow coefficient of the water pump; its value is 3600;

[0030] π represents pi, and its value is 3.14;

[0031] g represents the acceleration due to gravity, and its value is 9.8;

[0032] h represents water loss of the pump.

[0033] Step 2, train the discriminator and generator. Take the data of three dimensions of frequency, flow rate and head as input, and whether it is real data as output, and train the discriminator. The input data is divided into real production data and data generated by the generator, where the label of real data is true (1) and the label of data generated by the generator is false (0); take frequency and flow rate as input, and head as output, and train the generator;

[0034] The entire model framework is divided into two parts: the generator and the discriminator. Each time the entire data set is used to train the model, the discriminator and the generator are trained alternately. When training one module, the parameters of the other module are frozen, and the discriminator is trained first. Both the discriminator and the generator are three-layer neural networks with 32 nodes in each hidden layer. Specifically, for each set of frequency and flow data in the database (F i ,Q i ), multiply it by the random noise perturbation coefficient in the range [0.5,2], and get (F i ′,Q i ′). Input the disturbed data into the generator G to obtain Thus, a new operating point can be obtained through random noise perturbation and generator The discriminator inputs a working point, and the output layer outputs a number between (0,1) through the sigmoid activation function, indicating the probability that the working point is the actual working point of the pump, that is, When training the discriminator, the loss function is

[0035]

[0036] Where G(·) represents the generator mapping function, D(·) represents the discriminator mapping function,

[0037] N represents the number of working condition points;

[0038] F i ,Q i ,H i are the frequency, flow rate and head at the original speed of the i-th operating point respectively;

[0039] F i ′,Q i ′ are the frequency and flow rate after the speed changes at the i-th operating point;

[0040] When training the generator, the loss function is

[0041]

[0042] Step 3: Double verification denoising. Set thresholds η1 and η2, when the generator loss loss G<η1 and the loss of the discriminator is 0.5-η2 <loss D <0.5+η2, perform double verification denoising: use the discriminator to obtain the set of false operating points in the training data set S1={(F i ,Q i ,H i )|i=1,2,…,N;D(F i ,Q i ,H i )<0.4}; In addition, the operating point at the same frequency is fitted with the quadratic equation H=aQ for the relationship between flow and head using the least squares method 2 +bQ+c, marking the set of operating points S2 whose variance with the fitting curve is greater than the threshold σ. Thus, the intersection of operating points S=S1∩S2 that neither meet the judgment of the discriminator nor the mathematical law of the mechanism law is obtained. Delete the operating point set S from the training data set to reduce the noise data in the training set.

[0043] Step 4: Repeat the training step and denoising step until the discriminator and generator are fully converged.

[0044] Observe the changing trend of the loss function of the generator (G) and the discriminator (D) with the training iterations. Ideally, the loss of the generator should gradually decrease, while the loss of the discriminator should be close to 0.5. When the loss of the generator cannot be further reduced and the loss of the discriminator is close to 0.5, the training is terminated.

[0045] Step 5: derive the QH characteristic curve at the target frequency. After the training is completed, the generator G is extracted, and a frequency F is selected as needed, and the flow range [Q min ,Q max ] Generates evenly spaced pump flows M represents the selected flow Q i The number of min To Q max The denser the distance between them, the larger M will be. Input generator gets lift Therefore, a set of flow head operating points under frequency F can be obtained in the coordinate axis. Draw a smooth curve to obtain the QH characteristic curve of the pump at frequency F.

[0046] Although the embodiments of the present invention have been shown and described, it will be appreciated by those skilled in the art that various changes, modifications, substitutions and variations may be made to the embodiments without departing from the principles and spirit of the present invention, and that the scope of the present invention is defined by the claims and their equivalents.

Claims

1. A method for generating characteristic curves of a centrifugal water pump based on data augmentation and double verification denoising, characterized in that: The following steps are involved: S1, build training set; S2, training the discriminator and generator; S3, double verification denoising; S4, looping through the training and denoising steps until the discriminator and generator are fully converged; S5, deriving the QH characteristic curve at the target frequency.

2. The method for generating characteristic curve of a centrifugal water pump based on data augmentation and double verification denoising according to claim 1, characterized in that: In step S1, the method for constructing the training set is: Collect real-time operation data of the water pump, calculate the instantaneous head of the water pump, and obtain the actual operating point information of the water pump.

3. The method for generating characteristic curve of a centrifugal water pump based on data augmentation and double verification denoising according to claim 2, characterized in that: The real-time operation data of the water pump includes frequency and flow rate; The actual operating point information of the pump includes frequency, flow rate and instantaneous head.

4. The method for generating characteristic curve of a centrifugal water pump based on data augmentation and double verification denoising according to claim 1, characterized in that: In step S2, the method of training the discriminator and generator is: The frequency, flow rate and instantaneous head data are used as input, and whether it is real data is used as output to train the discriminator of the generative adversarial network; The generator of the generative adversarial network is trained by taking frequency and flow rate as input and instantaneous head as output.

5. The method for generating characteristic curve of a centrifugal water pump based on data augmentation and double verification denoising according to claim 1, characterized in that: In step S3, the double verification denoising method is: Set the thresholds η1 and η2, when the generator loss loss G <η1 and the loss of the discriminator D When <η2, double verification denoising is performed, and the discriminator and pump mathematical characteristics are used to detect and delete the noise operating points from the training set.

6. The method for generating characteristic curve of a centrifugal water pump based on data augmentation and double verification denoising according to claim 1, characterized in that: In step S5, the method for deriving the QH characteristic curve at the target frequency is: Set the specified frequency and flow range, use the generator to generate the head, get a set of operating points under the specified frequency, and draw a smooth curve to get the QH characteristic curve.

7. A computer device, characterized in that: The method comprises a memory, a processor and a computer program stored in the memory and executable on the processor, wherein the processor implements the steps of the method according to any one of claims 1 to 6 when executing the computer program.

8. A computer-readable storage medium, characterized in that: A computer program is stored thereon, and when the computer program is executed by a processor, the steps of the method described in any one of claims 1 to 6 are implemented.

Images