A water pump characteristic curve self-adaptive calibration and prediction method

Abstract

The application discloses a kind of water pump characteristic curve self-adapting calibration and prediction method, comprising the following steps: real-time acquisition of the data of each feature of water pump by external device, the data collected is uploaded to specified platform, and current and historical data storage;Establish ARMA model based on classification improvement, and each feature of water pump is predicted;The predicted value of each feature of water pump in step S2 and the acquisition value of each feature of water pump at current time are substituted into the operation law formula of water pump at rated frequency, to correct and predict the characteristic curve of water pump at current time;Least square fitting is carried out by formula based on the pipeline parameters collected;Draw the pipeline characteristic curve and the water pump characteristic curve, and the intersection of the two curves is the actual working point of the predicted water pump. The application has the characteristics of being able to effectively improve accuracy and expand the scope of application.

Description

Technical Field

[0001] This invention relates to a method for processing water pump characteristics, and more particularly to a method for adaptive calibration and prediction of water pump characteristic curves. Background Technology

[0002] Current traditional pump modeling methods typically focus on mechanistic modeling, failing to consider the changes in characteristic curves of the same type of pump under different operating conditions and over time. Therefore, traditional models tend to have significant errors. Furthermore, due to the specific characteristics of different industrial scenarios, the collected data is limited (equipment cannot be arbitrarily started or stopped in certain industrial environments), only collecting partial data points at fixed or specific frequencies. For example, in the existing technology titled "Modeling Method and System for Mathematical Model of Variable Frequency Speed ​​Control Pump," it uses massive amounts of historical pump operating data, grouping them according to the same frequency band and fitting the flow-head and flow-efficiency curves for the corresponding frequency bands. However, it still has the following shortcomings:

[0003] 1. This model family is only suitable for fitting the flow-head and flow-efficiency curves of a water pump to data within the same frequency band. It will be limited when the collected data cannot be in the same frequency band or when there are few data points in the same frequency band.

[0004] 2. The model's fitting of the pump characteristic curve based on historical data can, to some extent, address the problem of characteristic curve deviation caused by long-term use of the pump. However, it cannot detect pump failures or large deviations in the collected data in a timely manner.

[0005] Therefore, existing technologies suffer from poor accuracy and a narrow range of applications. Summary of the Invention

[0006] The purpose of this invention is to address the shortcomings of existing technologies by providing an adaptive calibration and prediction method for water pump characteristic curves. This invention effectively improves accuracy and expands its applicability.

[0007] The technical solution of this invention: an adaptive calibration and prediction method for water pump characteristic curves, comprising the following steps:

[0008] S1. Data on various characteristics of the water pump and pipeline are collected in real time by external equipment, the collected data is uploaded to the designated platform, and current and historical data are stored.

[0009] S2. Establish an ARMA model based on classification improvement and predict various features of the water pump;

[0010] S3. Substitute the predicted values ​​of each characteristic of the water pump in step S2 and the collected values ​​of each characteristic of the water pump at the current moment into the operating law formula of the water pump at the rated frequency, including the flow-head curve equation, the flow-power curve equation and the efficiency curve equation, to correct the characteristic curve of the water pump at the current moment, and obtain the predicted flow-head and flow-efficiency curves of the water pump in the next time unit.

[0011] S4. Based on the collected pipeline parameters, the pipeline characteristic curve is obtained by least squares fitting. The fitting formula is as follows: ,in For the head of the pipeline, For the flow rate of the pipeline, The parameters to be fitted are: plot the pipeline characteristic curve and the pump characteristic curve, and the intersection of the two curves is the predicted actual operating point of the pump.

[0012] Furthermore, in step S1, data on various characteristics of the water pump and pipeline are collected once a day, with each data point corresponding to the flow rate and head of the water pump and pipeline.

[0013] For each collected data point, the average value is calculated by taking multiple data points and then using that average as the measurement value for that data point.

[0014] Furthermore, in step S2, the classification-based improved ARMA model is implemented through the following steps:

[0015] 1) Classify the data of various characteristics of the water pump collected in step S1 into data at the same frequency, i.e., fixed frequency data, and data at different frequencies, i.e., variable frequency data;

[0016] 2) Classify the fixed-frequency and variable-frequency data by quarter, and extract the characteristic data information of each category, including the maximum, minimum, mean, truncated mean, and median of each characteristic. Select one data point of the characteristic to describe that characteristic of the category, denoted as . , represents the median of the data collected from the j-th feature point in the i-th class;

[0017] 3) Construct a vector from the data information of each feature in each class. Based on the Legendre best approximation criterion, find the collected data vector that is closest to the feature data information vector, and record the corresponding date in this class. ;

[0018] 4) Take the m features from the first N classes in the classified dataset, and predict the m feature points of the current class (N+1th class) based on the ARMA model. The specific formula is as follows:

[0019] ;

[0020] Where p is the order of the autoregressive term AR. Autoregressive coefficients; q is the order of the moving average term MA. The moving average coefficient is... It is white noise. This defines the boundaries of the noise; E represents the expected value, and Var represents the variance. This indicates any data collected from the same type of water pump.

[0021] 5) Substitute the values ​​of each feature variable obtained from the collection into the data information of each feature of the next predicted class obtained in the above steps, with the corresponding date being... ;

[0022] ;

[0023] 6) Based on the feature values ​​of the last kt days in the measured Nth class and the predicted future feature data information vector, fit the change curves of each feature, and use a function... This is expressed as follows: Predictions are made on a 15-day cycle, allowing for the updating and iteration of the pump characteristic curve every 15 days for the next quarter; the specific formula is as follows:

[0024]

[0025] Solving the formula using the least squares method yields the function. , where the function Let represent the prediction function fitted to the i-th feature, and let be the prediction value for any time step in the next class based on the prediction function.

[0026] Furthermore, in step S3, the equations for the flow-head curve, the flow-power curve, and the efficiency curve are as follows:

[0027]

[0028]

[0029]

[0030] in, It refers to the pump head. It is the water pump flow rate. It's the pump efficiency. It's the power of the water pump. These are the parameters for fitting a quadratic curve.

[0031] Furthermore, data collected at the same frequency can be used to derive flow-head curve equations and flow-power curve equations at a specific frequency through the flow-head curve equation and the flow-power curve equation.

[0032] Data collected at different frequencies were fitted using a least-squares method combined with a mechanistic model, as shown below:

[0033]

[0034]

[0035]

[0036] in, To ensure that the water pump operates at a frequency of The head, flow rate, and power at that time For the rated frequency, This is the ratio of the operating frequency to the rated frequency.

[0037] The beneficial effects of this invention are:

[0038] Compared to existing technologies, this invention improves the accuracy of the pump's mathematical model by employing two separate modeling methods for data fusion to model the pump's characteristic curves at specific frequencies: First, fitting the pump's characteristic curve under variable frequency speed control, collecting parameters such as flow rate, head, and power at different frequencies, and converting these data into data for the same frequency through a mechanistic model to further provide the characteristic curve for that frequency; second, fitting the pump's characteristic curve under fixed frequency, by collecting pump characteristic parameters at the same frequency and fitting the characteristic curve for that frequency; and finally, fitting the pipeline characteristic curve, by measuring the pipeline head at different flow rates and fitting the pipeline characteristic curve for that scenario. This invention integrates and automates data acquisition, data processing, data verification, and real-time updates of the pump's characteristic curves, eliminating the need for additional work by R&D personnel. By collecting data at the same and different frequencies, it can predict the pump's flow-head and flow-efficiency curves for one or more future time units. This method reduces workload while maintaining the accuracy of the characteristic curves. Moreover, its application scenarios are broad and not limited to a specific industrial scenario.

[0039] This invention acquires data in real time and calibrates the characteristic curves of water pumps and pipelines based on this data. The model is adaptively updated, and historical data is analyzed. Using an improved time-series ARMA model, it determines whether the data acquired at the current moment is abnormal, further predicting future flow-head and flow-efficiency curves. Anomaly detection in water pumps can be used as a judgment criterion. It is applicable to different industrial environments and can simultaneously fit water pump characteristic curves at specified frequencies to data acquired at both fixed and variable frequencies. In summary, this invention effectively improves accuracy and expands its applicability. Attached Figure Description

[0040] Figure 1 The present invention provides a flowchart of an adaptive calibration and prediction method for water pump characteristic curves. Detailed Implementation

[0041] The specific embodiments of the present invention will be further described in detail below with reference to the accompanying drawings.

[0042] This invention provides an adaptive calibration and prediction method for water pump characteristic curves.

[0043] This invention implements a method for adaptive verification, calibration, and prediction of pump characteristic curves, including four aspects: data acquisition, data processing and rationality analysis, pump model calibration, and pump characteristic curve prediction. This is achieved through the following steps: Figure 1 The following is a detailed flowchart:

[0044] 1. External devices collect data in real time, once a day. Each data point corresponds to the flow rate, head, etc. of the water pump / pipeline. The data collected in real time by the external devices is uploaded to the designated platform, and current and historical data are stored.

[0045] For each data point collected, the average value is obtained by taking multiple data points and using it as the measurement value for that data point.

[0046] 2. Establishment of an Improved ARMA Model Based on Classification. Traditional ARMA models can predict data for several future days based on daily collected data. However, for data points such as those collected from pump characteristic curves, which have small short-term fluctuations and require prediction over a longer timescale, the predicted data trend is unclear, and significant computational resources are wasted. Therefore, the traditional model is improved to fully extract information from the data and improve the overall accuracy of data prediction. The improved model is implemented through the following steps:

[0047] The data collected in step 1 is classified into two parts: data at the same frequency (fixed frequency) and data at different frequencies (variable frequency).

[0048] The data from both parts are categorized by quarter, and information on each feature (flow rate, head, etc.) within the same category is extracted, including the maximum, minimum, mean, truncated mean (mean after removing the maximum and minimum values ​​within the same category), and median. The median of the feature is selected to describe that feature within that category, denoted as . , where represents the median of the data collected from the j-th feature point in the i-th class.

[0049] The medians of each feature in each class can be used to construct a vector. The data collected each day can be represented by a vector as follows: ,in Let q represent the q-th feature on day t under class i. According to the Legendre best approximation criterion, with equation (1-1) as the objective, find the collected data vector that is closest to the median vector of the data, and record the corresponding date in this class. .

[0050] (1-1)

[0051] In the classified dataset, take m features from the first N classes, and use the ARMA model to predict these m feature points for the current class (the (N+1)th class). The specific formula is as follows:

[0052] (1-2)

[0053] Where p is the order of the autoregressive term AR, which can be determined by the PACF plot. The autoregressive coefficient; q is the order of the moving average term MA, which can be determined from the ACF plot. The moving average coefficient is... It is white noise.

[0054] By substituting the measured values ​​of each feature variable into the steps described above, we can obtain the median of each feature for the predicted next class, corresponding to the date. .

[0055] (1-3)

[0056] Based on the feature values ​​of the last kt days (divided into k segments) in the Nth class of measurements and the predicted future median vector, fit the change curves of each feature (using a function). This is expressed as a 15-day cycle for forecasting, enabling the pump characteristic curve to be updated and iterated every 15 days for the next quarter. The specific implementation process is described by formula (1-4), which is solved using least squares to obtain the function.

[0057] (1-4)

[0058] Where the function Let represent the prediction function fitted to the i-th feature, and let be the prediction value for any time step in the next class based on the prediction function.

[0059] 3. Based on the predicted values ​​of each feature in step 2 and the measured values ​​of each data point at the current moment, substitute the operating law formulas of the water pump at the rated frequency, including the flow-head curve equation, the flow-power curve equation, and the efficiency curve equation, as shown below:

[0060] (1-5)

[0061] (1-6)

[0062] (1-7)

[0063] in It refers to the pump head. It is the water pump flow rate. It's the pump efficiency. It's the power of the water pump. These are the parameters for fitting a quadratic curve. Data collected at the same frequency and at different frequencies are processed separately for different industrial scenarios and data collection methods.

[0064] Data collected at the same frequency can be used to obtain the flow-head curve equation and flow-power curve equation at a specific frequency through formulas (1-5) and (1-6).

[0065] Data collected at different frequencies were fitted using a least-squares method combined with a mechanistic model, as shown below:

[0066] (1-8)

[0067] (1-9)

[0068] (1-10)

[0069] in To ensure that the water pump operates at a frequency of The head, flow rate, and power at that time For the rated frequency, This is the ratio of the operating frequency to the rated frequency.

[0070] Assuming data were collected at n different frequencies, the following calculations were performed using the least squares method from (1-5), (1-9), (1-6), and (1-10):

[0071] (1-11)

[0072] (1-12)

[0073] in:

[0074] (1-13)

[0075] This step not only corrects the pump's characteristic curve at the current moment, but also provides a prediction of the pump's flow-head and flow-efficiency curves for the next time unit.

[0076] 4. The collected pipeline parameters are fitted using the least squares method according to formula (1-14).

[0077] (1-14)

[0078] in For the head of the pipeline, For the flow rate of the pipeline, These are the parameters that need to be fitted.

[0079] Plot the pipeline characteristic curve and the pump characteristic curve. Their intersection point is the predicted actual operating point of the pump.

[0080] The embodiments of the present invention can achieve the following functions: each component on the platform should be able to interact independently with systems outside the platform, collect and transmit data in real time, and upload it to the data processing platform;

[0081] Based on real-time data acquisition, a script is created to automatically fit and correct the characteristic curves of the water pump and pipeline, calculate the real-time pipeline impedance coefficient, and update and correct the original mechanism model of the water pump in real time.

[0082] Based on historical data, predict the changes in the pump characteristic curve;

[0083] It is applicable to different industrial environments, and the collected data is processed in blocks to obtain more accurate pump characteristic curves.

[0084] To fulfill the above functions, an external system is required to collect and input data in real time and update the characteristic curve of the water pump in real time, without requiring additional manual processing by the platform's developers, thus avoiding secondary development of the platform and improving efficiency.

[0085] The above embodiments are used to explain and illustrate the present invention, but not to limit the present invention. Any modifications and changes made to the present invention within the spirit and scope of the claims shall fall within the protection scope of the present invention.

Claims

1. A method for adaptive calibration and prediction of water pump characteristic curves, characterized in that, Includes the following steps: S1. Data on various characteristics of the water pump and pipeline are collected in real time by external equipment, the collected data is uploaded to the designated platform, and current and historical data are stored. S2. Establish an ARMA model based on classification improvement and predict various features of the water pump; In step S2, the ARMA model based on classification improvement is implemented through the following steps: 1) Classify the data of various characteristics of the water pump collected in step S1 into data at the same frequency, i.e., fixed frequency data, and data at different frequencies, i.e., variable frequency data; 2) Classify the fixed-frequency and variable-frequency data by quarter, and extract the characteristic data information of each category, including the maximum, minimum, mean, truncated mean, and median of each characteristic. Select one data point of the characteristic to describe that characteristic of the category, denoted as . , represents the median of the data collected from the j-th feature point in the i-th class; 3) Construct a vector from the data information of each feature in each class. Based on the Legendre best approximation criterion, find the collected data vector that is closest to the feature data information vector, and record the corresponding date in this class. ; 4) Take the m features from the first N classes in the classified dataset, and predict the m feature points of the current class (N+1th class) based on the ARMA model. The specific formula is as follows: ； Where p is the order of the autoregressive term AR. is the autoregressive coefficient, where q is the order of the moving average term MA. is the moving average coefficient, where , It is white noise. This defines the boundaries of the noise; E represents the expected value, and Var represents the variance. This indicates any data collected from the same type of water pump. 5) Substitute the values ​​of each feature variable obtained from the collection into the data information of each feature of the next predicted class obtained in the above steps, with the corresponding date being... ; ； 6) Based on the feature values ​​of the last kt days in the measured Nth class and the predicted future feature data information vector, fit the change curves of each feature, and use a function... This is described in terms of selecting a forecast period and updating and iterating the pump characteristic curve for each forecast period in the next quarter; the specific formula is as follows: Solving the formula using the least squares method yields the function. , where the function Let represent the prediction function fitted to the i-th feature, and obtain the predicted value at any time in the next class based on the prediction function; S3. Substitute the predicted values ​​of each characteristic of the water pump in step S2 and the collected values ​​of each characteristic of the water pump at the current moment into the operating law formula of the water pump at the rated frequency, including the flow-head curve equation, the flow-power curve equation and the efficiency curve equation, to correct the characteristic curve of the water pump at the current moment, and obtain the predicted flow-head and flow-efficiency curves of the water pump in the next time unit. S4. Based on the collected pipeline parameters, the pipeline characteristic curve is obtained by least squares fitting. The fitting formula is as follows: ,in For the head of the pipeline, For the flow rate of the pipeline, The parameters to be fitted are: plot the pipeline characteristic curve and the pump characteristic curve, and the intersection of the two curves is the predicted actual operating point of the pump.

2. The method for adaptive calibration and prediction of pump characteristic curves according to claim 1, characterized in that, In step S1, data on various characteristics of the water pump and pipeline are collected once a day, with each data point corresponding to the flow rate and head of the water pump and pipeline. For each data point collected, the average value is obtained by taking multiple data points and using it as the measurement value for that data point.

3. The method for adaptive calibration and prediction of pump characteristic curves according to claim 1, characterized in that, In step S3, the equations for the flow-head curve, the flow-power curve, and the efficiency curve are shown below: in, It refers to the pump head. It is the water pump flow rate. It's the pump efficiency. It's the power of the water pump. These are the parameters for fitting a quadratic curve.

4. The method for adaptive calibration and prediction of pump characteristic curves according to claim 3, characterized in that, Data collected at the same frequency can be used to derive flow-head curve equations and flow-power curve equations at a specific frequency through flow-head curve equations and flow-power curve equations. Data collected at different frequencies were fitted using a least-squares method combined with a mechanistic model, as shown below: in, To ensure that the water pump operates at a frequency of The head, flow rate, and power at that time For the rated frequency, This is the ratio of the operating frequency to the rated frequency.

Images