Method and system for evaluating energy efficiency of pump

Abstract

The invention discloses a method and system for evaluating the energy efficiency of a machine pump, and the method comprises the steps: collecting the historical data of a plurality of machine pumps, the historical data comprising the historical operation data related to the energy efficiency of each machine pump and the corresponding historical energy efficiency data; obtaining operation data, related to energy efficiency, of the to-be-evaluated pump in a current time period in real time, and determining an actual data type included in the operation data in the current time period; extracting a first type of operation data and a first type of energy efficiency data related to the actual data type from the historical data, constructing a training data set, and training a first preset model by using the training data set so as to obtain an energy efficiency soft measurement model used for evaluating the energy efficiency of the pump in the next time period; and obtaining a corresponding energy efficiency evaluation result by using the energy efficiency soft measurement model according to the operation data of the next time period. According to the method, the pump energy efficiency is accurately evaluated under the condition that part of pump operation parameters are lacked.

Description

Technical Field

[0001] The present invention belongs to the field of machine pump energy efficiency monitoring, and in particular relates to a method and system for evaluating machine pump energy efficiency. Background Art

[0002] Pumps are the most common fluid conveying equipment in industries such as petrochemicals. According to statistics, the electricity consumption of pump equipment accounts for more than 21% of the total electricity consumption in the country. Therefore, how to accurately evaluate the real-time energy efficiency of pumps and understand the current status of pump energy consumption is the key to further reducing pump energy consumption.

[0003] At present, the energy efficiency of pumps is mainly evaluated by calculating the ratio of the electrical energy input to the mechanical energy output of the pump. This energy efficiency calculation method requires complete pressure and flow meters on site, and relies on expert experience and knowledge to query the experience curve or consult the motor manufacturer to obtain the power factor and efficiency of the motor under actual working conditions. Most petrochemical companies conduct an energy efficiency survey of pumps every year, but due to insufficient on-site instrument configuration and the huge workload of traditional energy efficiency calculation methods, the coverage rate of energy efficiency surveys is low.

[0004] In the soft-sensing technology of pump energy efficiency, there are only a few studies on mechanism-driven or data-driven prediction of pump energy efficiency in recent years. For example:

[0005] (1) The centrifugal pump efficiency prediction method based on data-driven modeling, firstly, by analyzing the relevant process knowledge of the centrifugal pump, the efficiency of the centrifugal pump is divided into two different intervals; then, the prediction variance output by the Gaussian process regression (GPR) model is used to identify the process of the efficiency curve of the centrifugal pump at different speeds, and the two intervals are automatically divided; finally, the prediction variance of GPR is combined with the posterior probability index of the Bayesian theorem to measure the approximate working conditions, the data at different speeds are measured, and the groups of data with the highest approximate variance are selected to form a training set to train the centrifugal pump efficiency prediction model. However, this method mainly studies the energy efficiency of the variable frequency drive centrifugal pump under different working conditions. By dividing the efficiency curve into two stages, namely, rising and falling, two GPR models are used for prediction respectively. The model is relatively cumbersome, and different working conditions of the same centrifugal pump need to be predicted separately, and the centrifugal pump efficiency prediction cannot be realized in the absence of operating parameters such as the inlet and outlet pressure of the centrifugal pump. It can be seen that this method obtains a centrifugal pump efficiency prediction model based on Gaussian process regression. To obtain this model, it is necessary to construct prediction models for a certain variable frequency centrifugal pump under the conditions of efficiency improvement and efficiency reduction. The operability is poor and it is impossible to achieve energy efficiency prediction in the absence of pump inlet and outlet pressure.

[0006] (2) A method for predicting the efficiency of a centrifugal pump. First, a standard is established based on the process knowledge of the external characteristics of the centrifugal pump to divide the efficiency curve of the centrifugal pump at different speeds into two stages. The similarity law of the centrifugal pump and the least squares support vector regression (LSSVR) are used to build models based on the characteristics of the two stages. Finally, a hybrid model is used to dynamically predict the efficiency of the two stages at different speeds. The technical focus of this method is to predict the efficiency of the centrifugal pump at different speeds. It is necessary to effectively integrate the mechanism knowledge of the centrifugal pump and the LSSVR model into the hybrid model. It is impossible to predict the efficiency of the centrifugal pump in the absence of operating parameters such as the inlet and outlet pressures of the centrifugal pump.

[0007] (3) A method for predicting the efficiency of a centrifugal pump based on correction of simulation results. In view of the deviation between the efficiency values ​​of the centrifugal pump obtained by computer simulation and experimental testing, multiple corrections are made through the conversion of various fitting curves to establish a correction method for the simulation results of the centrifugal pump, so as to accurately predict the change of the efficiency value of the centrifugal pump with the flow rate under various new design schemes. Although this method is objective, scientific, and versatile, and helps to improve the reliability of the simulation results of the efficiency of the centrifugal pump, it needs to rely on the experience and knowledge of experts to formulate a complex mechanism simulation model for a certain centrifugal pump, that is, there is a problem of relying on the experience and knowledge of experts to establish a complex fluid mechanics model, which is not suitable for the energy efficiency prediction of large-scale pump groups. At the same time, this method cannot realize the efficiency prediction of the centrifugal pump in the absence of operating parameters such as the inlet and outlet pressures of the centrifugal pump.

[0008] With the improvement of computer performance, data-driven energy efficiency soft measurement models have been fully developed. There is no need to study complex mechanism models or rely on expert experience and knowledge to establish a soft measurement model that can accurately reflect the energy efficiency of the pump. Summary of the invention

[0009] In order to solve the above problems, an embodiment of the present invention provides a method for evaluating the energy efficiency of a machine pump, comprising: collecting historical data of multiple machine pumps, wherein the historical data includes historical operating data related to energy efficiency and corresponding historical energy efficiency data of each machine pump; acquiring the operating data of the current time period related to the energy efficiency of the machine pump to be evaluated in real time, and determining the actual data type contained in the operating data of the current time period; extracting a first type of operating data and a first type of energy efficiency data related to the actual data type from the historical data, constructing a training data set, and using the training data set to train a first preset model, so as to obtain an energy efficiency soft measurement model for evaluating the energy efficiency of the machine pump in the next time period; and obtaining a corresponding energy efficiency evaluation result based on the operating data of the next time period using the energy efficiency soft measurement model.

[0010] Preferably, the method also includes: first performing data cleaning on the collected historical data to remove erroneous data, then uniformly normalizing the non-numerical data in the remaining data, and normalizing the numerical data in the remaining data, so as to use the uniformly normalized non-numerical data and normalized numerical data to train the first preset model.

[0011] Preferably, the process of uniformly normalizing non-numerical data includes: classifying the non-numerical data according to data sources to form multiple groups of data sequences corresponding to different data sources; encoding each group of data sequences according to different encoding methods to obtain encoded non-numerical data.

[0012] Preferably, the encoding method is determined based on the statistical results of the number of source categories corresponding to different elements in each group of data sequences, including but not limited to: onehot encoding, mean encoding, label encoding, Bayesian Target encoding and Leave-one-out encoding, wherein when the number of source categories in the corresponding group is less than or equal to 5, the onehot encoding is used, or, when the number of source categories in the corresponding group is greater than 5, the mean encoding or the label encoding is used.

[0013] Preferably, the normalization processing method adopted is selected from one of maximum-minimum value normalization and Z-Score normalization.

[0014] Preferably, before the remaining data are uniformly normalized and standardized, the method further includes: performing null value analysis on the remaining data, and assigning the corresponding null value data to zero according to the analysis result, so as to eliminate the influence of the null value phenomenon on the model training result.

[0015] Preferably, the first preset model is trained using a machine learning regression algorithm, and the machine learning regression algorithm is selected from one of a back propagation neural network BPNN, a lightweight and efficient gradient boosting tree LGBM, an extreme gradient boosting XGBoost, an extreme learning machine ELM and a KNN algorithm.

[0016] Preferably, the method also includes: when using the back propagation neural network BPNN, the framework of the first preset model connects each layer in a node connection manner, and the layers include a fully connected input layer, an output layer, three hidden layers, three BN layers and a dropout layer, wherein the input data of the input layer is the historical operation data in the training data set, and its dimension is the number of actual data types contained in the historical operation data, wherein, during the training process, the training of the first preset model is completed by configuring the initial learning rate, number of iterations and activation function for the back propagation neural network BPNN, and the activation function adopts the ReLu function.

[0017] Preferably, the training data set is composed of 80% of the extracted data. After obtaining the energy efficiency soft measurement model, the method further includes: constructing the data in the extracted data that are not selected into the training data set as a test data set, and using the test data set to perform error analysis on the energy efficiency soft measurement model to determine the adaptability of the current energy efficiency soft measurement model to the actual application based on the analysis results.

[0018] Preferably, the process of performing error analysis on the energy efficiency soft measurement model includes: calculating the mean square error and mean absolute error percentage between the predicted energy efficiency data predicted using the actual operating data in the test data set and the actual energy efficiency data directly extracted from the test data set, so as to obtain the error analysis result of the energy efficiency soft measurement model.

[0019] Preferably, the data types contained in the historical operation data include but are not limited to: installation entity, manufacturer, rated flow, rated head, motor rated power, motor rated voltage, motor rated current, medium specific gravity, operating years, motor power factor, motor efficiency, inlet pressure, outlet pressure, mass flow, volume flow, operating voltage, operating current and operating efficiency.

[0020] On the other hand, the present invention also provides a system for evaluating the energy efficiency of machine pumps, the system comprising the following modules: a historical data acquisition module, which is used to collect historical data of multiple machine pumps, wherein the historical data includes historical operating data related to energy efficiency and corresponding historical energy efficiency data of each machine pump; a real-time data acquisition module, which is used to acquire in real time the operating data of the machine pump to be evaluated in the current time period related to the energy efficiency, and determine the actual data type contained in the operating data of the current time period; a model generation module, which is used to extract a first type of operating data and a first type of energy efficiency data related to the actual data type from the historical data, construct a training data set, and use the training data set to train a first preset model, so as to obtain an energy efficiency soft measurement model for evaluating the energy efficiency of the machine pump in the next time period; an energy efficiency evaluation module, which is used to obtain the corresponding energy efficiency evaluation result based on the operating data of the next time period using the energy efficiency soft measurement model.

[0021] Compared with the prior art, one or more embodiments of the above scheme may have the following advantages or beneficial effects:

[0022] The present invention proposes a method and system for evaluating the energy efficiency of a pump. The method first collects historical data of multiple pumps, then uses the collected historical data, combined with the real-time data of the pump to be evaluated, to establish an energy efficiency soft measurement model based on machine learning technology for evaluating the energy efficiency of the pump in the next period; finally, based on the established energy efficiency soft measurement model, the operating data of the pump to be evaluated in the next period is used to obtain the energy efficiency evaluation result corresponding to the period. The present invention overcomes the defects of the traditional method that relies on expert experience and knowledge to set key parameters and construct complex mechanism models, and realizes accurate evaluation of the energy efficiency of the pump under the condition of missing some pump operating parameters.

[0023] Other features and advantages of the present invention will be described in the following description, and partly become apparent from the description, or understood by practicing the present invention. The purpose and other advantages of the present invention can be realized and obtained by the structures particularly pointed out in the description, claims and drawings. BRIEF DESCRIPTION OF THE DRAWINGS

[0024] The accompanying drawings are used to provide a further understanding of the present invention and constitute a part of the specification. Together with the embodiments of the present invention, they are used to explain the present invention and do not constitute a limitation of the present invention. In the accompanying drawings:

[0025] Figure 1 It is a step diagram of a method for evaluating the energy efficiency of a machine pump according to an embodiment of the present application.

[0026] Figure 2 It is a schematic diagram of the evaluation result of the missing machine pump inlet pressure of the method for evaluating the machine pump energy efficiency according to an embodiment of the present application.

[0027] Figure 3 It is a schematic diagram of the evaluation result of the missing motor efficiency of the method for evaluating the energy efficiency of a machine pump according to an embodiment of the present application.

[0028] Figure 4 It is a schematic diagram of the overall structure of a system for evaluating the energy efficiency of a pump according to an embodiment of the present application. DETAILED DESCRIPTION

[0029] The following will describe the implementation methods of the present invention in detail with reference to the accompanying drawings and embodiments, so that the implementation process of how the present invention applies technical means to solve technical problems and achieve technical effects can be fully understood and implemented accordingly. It should be noted that as long as there is no conflict, the various embodiments of the present invention and the various features in the embodiments can be combined with each other, and the technical solutions formed are all within the protection scope of the present invention.

[0030] In addition, the steps shown in the flowcharts of the accompanying drawings can be executed in a computer system such as a set of computer-executable instructions, and although a logical order is shown in the flowcharts, in some cases, the steps shown or described can be performed in an order different from that shown here.

[0031] Pumps are the most common fluid conveying equipment in industries such as petrochemicals. According to statistics, the electricity consumption of pump equipment accounts for more than 21% of the total electricity consumption in the country. Therefore, how to accurately evaluate the real-time energy efficiency of pumps and understand the current status of pump energy consumption is the key to further reducing pump energy consumption.

[0032] At present, the energy efficiency of pumps is mainly evaluated by calculating the ratio of the electrical energy input to the mechanical energy output of the pump. This energy efficiency calculation method requires complete pressure and flow meters on site, and relies on expert experience and knowledge to query the experience curve or consult the motor manufacturer to obtain the power factor and efficiency of the motor under actual working conditions. Most petrochemical companies conduct an energy efficiency survey of pumps every year, but due to insufficient on-site instrument configuration and the huge workload of traditional energy efficiency calculation methods, the coverage rate of energy efficiency surveys is low.

[0033] In the soft-sensing technology of pump energy efficiency, there are only a few studies on mechanism-driven or data-driven prediction of pump energy efficiency in recent years. For example:

[0034] (1) The centrifugal pump efficiency prediction method based on data-driven modeling, firstly, by analyzing the relevant process knowledge of the centrifugal pump, the efficiency of the centrifugal pump is divided into two different intervals; then, the prediction variance output by the Gaussian process regression (GPR) model is used to identify the process of the efficiency curve of the centrifugal pump at different speeds, and the two intervals are automatically divided; finally, the prediction variance of GPR is combined with the posterior probability index of the Bayesian theorem to measure the approximate working conditions, the data at different speeds are measured, and the groups of data with the highest approximate variance are selected to form a training set to train the centrifugal pump efficiency prediction model. However, this method mainly studies the energy efficiency of the variable frequency drive centrifugal pump under different working conditions. By dividing the efficiency curve into two stages, namely, rising and falling, two GPR models are used for prediction respectively. The model is relatively cumbersome, and different working conditions of the same centrifugal pump need to be predicted separately, and the centrifugal pump efficiency prediction cannot be realized in the absence of operating parameters such as the inlet and outlet pressure of the centrifugal pump. It can be seen that this method obtains a centrifugal pump efficiency prediction model based on Gaussian process regression. To obtain this model, it is necessary to construct prediction models for a certain variable frequency centrifugal pump under the conditions of efficiency improvement and efficiency reduction. The operability is poor and it is impossible to achieve energy efficiency prediction in the absence of pump inlet and outlet pressure.

[0035] (2) A method for predicting the efficiency of a centrifugal pump. First, a standard is established based on the process knowledge of the external characteristics of the centrifugal pump to divide the efficiency curve of the centrifugal pump at different speeds into two stages. The similarity law of the centrifugal pump and the least squares support vector regression (LSSVR) are used to build models based on the characteristics of the two stages. Finally, a hybrid model is used to dynamically predict the efficiency of the two stages at different speeds. The technical focus of this method is to predict the efficiency of the centrifugal pump at different speeds. It is necessary to effectively integrate the mechanism knowledge of the centrifugal pump and the LSSVR model into the hybrid model. It is impossible to predict the efficiency of the centrifugal pump in the absence of operating parameters such as the inlet and outlet pressures of the centrifugal pump.

[0036] (3) A method for predicting the efficiency of a centrifugal pump based on correction of simulation results. In view of the deviation between the efficiency values ​​of the centrifugal pump obtained by computer simulation and experimental testing, multiple corrections are made through the conversion of various fitting curves to establish a correction method for the simulation results of the centrifugal pump, so as to accurately predict the change of the efficiency value of the centrifugal pump with the flow rate under various new design schemes. Although this method is objective, scientific, and versatile, and helps to improve the reliability of the simulation results of the efficiency of the centrifugal pump, it needs to rely on the experience and knowledge of experts to formulate a complex mechanism simulation model for a certain centrifugal pump, that is, there is a problem of relying on the experience and knowledge of experts to establish a complex fluid mechanics model, which is not suitable for the energy efficiency prediction of large-scale pump groups. At the same time, this method cannot realize the efficiency prediction of the centrifugal pump in the absence of operating parameters such as the inlet and outlet pressures of the centrifugal pump.

[0037] With the improvement of computer performance, data-driven energy efficiency soft measurement models have been fully developed. There is no need to study complex mechanism models or rely on expert experience and knowledge to establish a soft measurement model that can accurately reflect the energy efficiency of the pump.

[0038] Therefore, in order to solve the above problems, the embodiment of the present invention proposes a method and system for evaluating the energy efficiency of a machine pump. The method first collects historical data of multiple machine pumps, and then uses the collected historical data, combined with the real-time data of the machine pump to be evaluated, to establish an energy efficiency soft measurement model based on machine learning technology for evaluating the energy efficiency of the machine pump in the next period; finally, based on the established energy efficiency soft measurement model, the operating data of the machine pump to be evaluated in the next period is used to obtain the energy efficiency evaluation result corresponding to the period. The present invention overcomes the defects of traditional methods that rely on expert experience and knowledge to set key parameters and construct complex mechanism models, and realizes accurate evaluation of the energy efficiency of the machine pump under the condition of missing some machine pump operating parameters.

[0039] Embodiment 1

[0040] In view of the fact that traditional formula methods and conventional data-driven and mechanism-driven energy efficiency models cannot carry out energy efficiency evaluation in the absence of relevant pump operating parameters, the present invention adds auxiliary parameters other than operating parameters to participate in the soft measurement of pump energy efficiency, and fully explores the manufacturer, the installation entity where the pump is located, the pump design parameters, the pump operation time and other easy-to-collect and stable auxiliary parameters to participate in model training, and fully learns the hidden information between the auxiliary parameters, the real-time operating parameters of the pump and the pump energy efficiency through machine learning technology, so as to realize accurate evaluation of the pump energy efficiency in the absence of relevant pump operating parameters.

[0041] Figure 1 This is a step diagram of a method for evaluating the energy efficiency of a pump according to an embodiment of the present application. Figure 1 To illustrate the various steps of this method.

[0042] like Figure 1 As shown, in step S110, historical data of multiple pumps are collected. The auxiliary parameters of this embodiment are historical data from multiple pumps. In order to build an energy efficiency soft measurement model with good accuracy in the absence of a small number of pump operating parameters, it is necessary to obtain complete and comprehensive auxiliary parameters other than operating parameters related to energy efficiency. Therefore, the multiple pumps from the production site in this embodiment cover multiple types, and each type includes several pumps. In this embodiment of the application, historical data includes historical operating data related to energy efficiency and corresponding historical energy efficiency data of each pump.

[0043] Furthermore, the data types included in the historical operation data include but are not limited to: installation entity, manufacturer, rated flow, rated head, motor rated power, motor rated voltage, motor rated current, medium specific gravity, operating years, motor power factor, motor efficiency, inlet pressure, outlet pressure, mass flow, volume flow, operating voltage, operating current and operating efficiency.

[0044] After collecting the historical data, the present invention preprocesses the collected historical data to use the preprocessed historical data as valid data to establish an energy efficiency soft measurement model based on machine learning technology. In the process of preprocessing the collected historical data, the collected historical data is first cleaned to remove erroneous data, and then the non-numerical data in the remaining data is uniformly normalized, and the numerical data in the remaining data is normalized, so as to train the first preset model using the uniformly normalized non-numerical data and the normalized numerical data.

[0045] First, the collected historical data is cleaned to remove erroneous data, so as to eliminate the influence of erroneous data on the measurement accuracy of the trained energy efficiency soft measurement model during the model training process. The removed erroneous data include but are not limited to: energy efficiency data with energy efficiency exceeding 100% or less than 1% and its corresponding operating data, operating data with serious deviation from the actual mass flow and volume flow and its corresponding energy efficiency data, and all operating data of a certain type lacking more than two or more data and its corresponding energy efficiency data.

[0046] Next, this embodiment uses the historical data that has completed data cleaning as the remaining data, and performs unified normalization processing on the non-numerical data in the remaining data, and normalizes the numerical data in the remaining data, so as to train the first preset model using the unified normalized non-numerical data and the normalized numerical data. Specifically, since the collected historical operation data contains a large number of data types, directly using the unprocessed historical data for model training, although it can achieve the expected goal, the amount of calculation is large and the operation is inconvenient. Therefore, this embodiment first divides the collected historical data into non-numerical data without specific numerical values ​​(for example: installation entity, manufacturer) and numerical data with specific numerical values ​​(for example: rated flow, rated head, motor rated power, etc.). Then, by processing the non-numerical data and numerical data respectively, the input data and output data used for model training are simplified to improve the speed of model training and reduce the complexity of the model training process. For the non-numerical data in the remaining data, this embodiment adopts a unified normalization processing method, so as to train the first preset model using the unified normalized non-numerical data. For the numerical data in the remaining data, this embodiment adopts a normalization processing method, so as to use the normalized numerical data to train the first preset model.

[0047] In the process of uniformly normalizing non-numerical data, the non-numerical data is first classified according to the data source to form multiple groups of data sequences corresponding to different data sources; then, each group of data sequences is encoded according to different encoding methods, thereby obtaining encoded non-numerical data. In an embodiment of the present application, the non-numerical data is first classified according to the source, for example: the source categories of the non-numerical data belonging to the manufacturer are Dalian Danai Pump Industry Co., Ltd., Danai Pump Industry, and Danai Pump Industry Co., Ltd., which are subsidiaries of the same parent company. At this time, the subsidiaries belonging to the aforementioned parent company in the current manufacturer are determined to belong to the same total source, and the same total source is used as Danai Pump Industry to form a data sequence corresponding to Danai Pump Industry containing the aforementioned subsidiaries (or source categories). Then, for different groups of data sequences, different encoding methods are matched for each group of data sequences, so that the data sequences of different groups are encoded according to the corresponding encoding methods to obtain encoded non-numerical data. Therefore, the present invention uses encoded non-numerical data to train the first preset model, which effectively simplifies the complexity of the operation.

[0048] Further, the encoding method is determined according to the quantitative statistical results of the source categories corresponding to different elements in each group of data sequences. Based on the fact that the source categories of the elements contained in each group of data sequences have different characteristics in number, the present invention adopts different encoding methods to adapt to data sequences with different numbers of source categories, so that the energy efficiency soft measurement model obtained by the encoding training of each group of data sequences obtained has a wider scope of application. In an embodiment of the present application, each group of data sequences contains multiple elements, each element corresponds to a source category, and the source categories of different elements may be the same or different. Therefore, if different elements in the data sequence correspond to different source categories, the corresponding quantitative statistical results are the same as the number of elements, and if some elements in the data sequence correspond to the same source category, the corresponding quantitative statistical results are less than the number of elements. Among them, the encoding method includes but is not limited to: onehot encoding, mean encoding, label encoding, Bayesian Target encoding and Leave-one-out encoding. In a specific embodiment of the present application, when the number of source categories of the corresponding group is less than or equal to 5, onehot encoding is used, or, when the number of source categories of the corresponding group is greater than 5, mean encoding or label encoding is used.

[0049] In the process of normalizing the numerical data, the normalization method used is selected from one of maximum-minimum value normalization and Z-Score normalization.

[0050] Before the remaining data is uniformly standardized and normalized, the present invention performs a null value analysis on the remaining data, and assigns the corresponding null value data to zero according to the analysis results to eliminate the influence of the null value phenomenon on the model training results. In practical applications, the collected historical data may be affected by factors such as external environmental interference or the characteristics of the parameter acquisition device itself, so that some data bits have null values. Therefore, this embodiment determines the data bits of the null value by performing a null value analysis on the remaining data. After determining the data bits of the null value, the null value data of the corresponding bit is filled with 0 to obtain coded null value data, and the filled data is marked as 1 for later data verification. Accordingly, the present invention obtains the remaining data with complete numerical data and non-numerical data by filling the null value data, thereby effectively eliminating the influence of the null value phenomenon on the model training results, and greatly improving the accuracy of the measurement results of the energy efficiency soft measurement model.

[0051] Further, in step S120, the operation data of the current period of the pump to be evaluated related to the energy efficiency is obtained in real time, and the actual data type contained in the operation data of the current period is determined. In the embodiment of the present application, the operation data of the current period of the pump to be evaluated related to the energy efficiency is obtained in real time during the operation of the pump to be evaluated, and the data type related to the energy efficiency contained in the operation data of the current period is determined and used as the actual data type.

[0052] After the historical data is collected and the actual data type of the current period is determined, in step S130, the first type of operating data and the first type of energy efficiency data related to the actual data type are extracted from the historical data, a training data set is constructed, and the first preset model is trained using the training data set, so as to obtain an energy efficiency soft measurement model for evaluating the energy efficiency of the pump in the next period. In practical applications, there are situations where all data types contained in the historical data cannot be obtained on site, that is, the actual data type has the characteristics of missing some data types compared with the data type corresponding to the historical data. In order to obtain an energy efficiency soft measurement model with the ability to evaluate the energy efficiency of the pump to be evaluated, this embodiment removes the historical data corresponding to the missing data type in the historical data, extracts the first type of operating data and the first type of energy efficiency data related to the actual data type from the historical data, and uses the combination of the first type of operating data and the first type of energy efficiency data as the training data set. Afterwards, using the constructed training data set, the historical operating data in the training data set is used as the input data of the first preset model, and the historical energy efficiency data corresponding to the historical operating data is used as the output data of the first preset model to realize the training of the first preset model. After the training of the first preset model is completed, an energy efficiency soft-sensing model for evaluating the energy efficiency of the machine pump in the next period is obtained.

[0053] In an embodiment of the present application, a machine learning regression algorithm is used to train the first preset model, and the machine learning regression algorithm is selected from one of the back propagation neural network BPNN, lightweight and efficient gradient boosting tree LGBM, extreme gradient boosting XGBoost, extreme learning machine ELM and KNN algorithm.

[0054] In a specific embodiment of the present application, when using the back propagation neural network BPNN, the framework of the first preset model connects each layer in a node connection manner, and the layers include a fully connected input layer, an output layer, three hidden layers, three BN layers and a dropout layer. In the process of training the first preset model, the input data of the input layer is the historical operation data in the training data set, and its dimension is the number of actual data types contained in the historical operation data. Based on the initial learning rate, number of iterations and activation function configured for the current back propagation neural network BPNN, the historical operation data in the training data set is used as the input data of the input layer, and the number of actual data types contained in the historical operation data is used as the training dimension, combined with the adaptive adjustment of the learning rate of the back propagation neural network BPNN, so as to realize the training of the first preset model.

[0055] In a specific embodiment of the present application, the activation function adopts the ReLu function, which is expressed by the following expression:

[0056]

[0057] Among them, f(x) represents the output of the neuron, and x represents the input of the neuron.

[0058] In step S140, the energy efficiency soft measurement model is used to obtain the corresponding energy efficiency evaluation result according to the operation data of the next period. After obtaining the operation data of the next period, the operation data of the next period is used as the input data of the trained energy efficiency soft measurement model, and the obtained output data is the energy efficiency data corresponding to the next period.

[0059] Further, the training data set is composed of 80% of the data in the extracted data. After obtaining the energy efficiency soft measurement model, the present invention constructs the data in the extracted data that are not selected into the training data set as a test data set, and uses the test data set to perform error analysis on the energy efficiency soft measurement model, so as to determine the adaptability of the current energy efficiency soft measurement model to the actual application according to the analysis results. In the embodiment of the present application, the extracted data is divided into a training data set and a test data set, and 80% of the data in the extracted data constitute the training data set, so as to ensure that the first preset model has sufficient input data, thereby ensuring the reliability of the energy efficiency measurement soft measurement model finally obtained. At the same time, the other 20% of the data outside the training data set in the extracted data (the data in the extracted data that are not selected into the training data set) is constructed as a test data set. After obtaining the test data set, the historical operation data in the test data set is used to predict the corresponding predicted energy efficiency data, so as to determine the adaptability of the current energy efficiency soft measurement model to the actual application according to the predicted energy efficiency data, and then determine whether the current model can be applied to the energy efficiency measurement of the pump to be evaluated.

[0060] In the process of error analysis of the energy efficiency soft measurement model, the mean square error and mean absolute error percentage between the predicted energy efficiency data predicted by the actual operation data in the test data set and the actual energy efficiency data directly extracted from the test data set are calculated to obtain the error analysis result of the energy efficiency soft measurement model. Specifically, in this embodiment, the historical operation data in the test data set is used as the input data of the obtained energy efficiency soft measurement model, and the corresponding predicted energy efficiency data is predicted by the current energy efficiency soft measurement model, that is, the output data of the current model is the predicted energy efficiency data. Afterwards, the predicted energy efficiency data and the collected actual energy efficiency data about the same historical operation data are analyzed and compared, and the mean square error and mean absolute error percentage between the predicted energy efficiency data predicted by the actual operation data in the test data set and the actual energy efficiency data directly extracted from the test data set are calculated, thereby generating the error analysis result of the current energy efficiency soft measurement model according to the mean square error and mean absolute error percentage.

[0061] In a specific embodiment of the present application, the error of the energy efficiency soft sensor model obtained may be caused by the neural network loss. If a validation data set is also used, there may also be an error caused by the validation data set. However, no matter what kind of error it is, it can be obtained according to the mean square error (MSE) and the mean absolute percentage error (MAPE).

[0062] In the embodiment of the present application, the mean square error is calculated using the following expression:

[0063]

[0064] Among them, MSE represents mean square error, N represents the number of training times, n represents the number of input data, i represents the sequence number of input data, Y(i) represents the actual energy efficiency data corresponding to the i-th input data, and Y′(i) represents the predicted energy efficiency data corresponding to the i-th input data.

[0065] In the embodiment of the present application, the mean absolute error percentage is calculated using the following expression:

[0066]

[0067] Here, MAPE stands for mean absolute percentage error.

[0068] In a specific embodiment of the present application, the constructed energy efficiency soft measurement model is verified using engineering application data, and the verification results show that the constructed model has the advantages of high accuracy and good reliability. Therefore, the present invention effectively solves the problems that the traditional formula method relies on expert experience knowledge to set key parameters, resulting in a large workload for pump energy efficiency evaluation, which is not conducive to improving the coverage of the pump energy efficiency survey, the conventional method needs to rely on expert experience knowledge to build a complex mechanism model, and the conventional method only uses a small sample energy efficiency data of a certain centrifugal pump to train the model, resulting in poor model generalization. It realizes accurate evaluation of pump energy efficiency under the condition of missing some pump operating parameters, and provides technical support for enterprises to improve pump equipment management and further carry out energy conservation and emission reduction work.

[0069] Embodiment 2

[0070] In the embodiment of the present application, a total of 1043 pieces of historical data of pumps were collected from several petrochemical production enterprises. The data types contained in the current historical data include: installation subject, manufacturer, rated flow, rated head, motor rated power, motor rated voltage, motor rated current, medium specific gravity, motor power factor, motor efficiency, outlet pressure, mass flow, volume flow, operating voltage, operating current and operating efficiency. Compared with the method of measuring the energy efficiency of pumps using the traditional formula of pump energy efficiency, the key parameter of inlet pressure is missing, and auxiliary parameters such as installation subject, manufacturer, rated flow, rated head, motor rated power, motor rated voltage, etc. that are easy to collect are added.

[0071] In this embodiment, when the pump inlet pressure cannot be obtained on site, the energy efficiency soft measurement model without the pump inlet pressure is constructed according to the method for evaluating the pump energy efficiency described in the first embodiment.

[0072] First, the collected historical data is cleaned to remove erroneous data, and the remaining data is obtained as shown in Table 1:

[0073] Table 1 Historical data after data cleaning

[0074]

[0075] Next, the historical data that has completed data cleaning is used as the remaining data, and the non-numerical data in the remaining data is standardized and normalized. The installation entity and the manufacturer of the pump are standardized, and the same type of installation entity and the same pump manufacturer are standardized with the same name, and the source categories with less than 5% of the total number of samples are uniformly classified as others. Among them, in the process of standardizing the non-numerical data, label coding is used for the pump installation entity and manufacturer, and the auxiliary parameters of different source categories are represented by numbers from 0 to n (category types). In the process of normalizing the numerical data, the maximum-minimum normalization method is adopted, and the normalization of the numerical data is achieved using the following expression:

[0076]

[0077] Among them, y′ i represents the normalized ith auxiliary parameter, y i represents the i-th auxiliary parameter before normalization, max(y) and Respectively represent the maximum and minimum values ​​of a certain type of auxiliary parameters.

[0078] Before the remaining data are uniformly normalized and normalized, a null value analysis is performed on the remaining data in the manner described in Example 1, and the corresponding null value data is assigned to zero according to the analysis result to eliminate the influence of the null value phenomenon on the model training result.

[0079] This embodiment uses a back propagation neural network BPNN. The framework of the first preset model connects each layer in a node connection manner. The layers include a fully connected input layer, an output layer, three hidden layers, three BN layers, and a dropout layer. Among them, the input layer nodes are 15, the output layer node is 1, the hidden layer nodes are 64, 64, and 32 respectively, and the initial learning rate is 10 -3 , the number of iterations is 200. Based on the initial learning rate, number of iterations and activation function configured for the current back propagation neural network BPNN, the historical operation data in the training data set is used as the input data of the input layer, and the number of actual data types contained in the historical operation data is used as the training dimension, combined with the adaptive adjustment of the learning rate of the back propagation neural network BPNN, so as to realize the training of the first preset model.

[0080] The activation function of this embodiment adopts the ReLu function.

[0081] Furthermore, the training data set of this embodiment is composed of 80% of the data in the extracted data, and the other 20% of the data in the extracted data other than the training data set (the data in the extracted data not selected into the training data set) is constructed as a test data set. By calculating the mean square error and mean absolute error percentage between the predicted energy efficiency data predicted by using the actual operation data in the test data set and the actual energy efficiency data directly extracted from the test data set, it is determined that the energy efficiency value predicted by the currently obtained energy efficiency soft measurement model is not much different from the actual value in the absence of the pump inlet pressure. Figure 2 Schematic diagram of the evaluation result of the missing pump inlet pressure in the method for evaluating the pump energy efficiency in an embodiment of the present application. Figure 2 , this embodiment takes the average value of 20 model training and test results, and obtains a mean square error of 0.0017 and a mean absolute error percentage of 3.4%. It can be seen that the measurement results of the energy efficiency soft measurement model obtained by the present invention are highly accurate and can meet the measurement accuracy requirements in actual engineering applications.

[0082] Embodiment 3

[0083] In the embodiment of the present application, a total of 1043 pieces of historical data of pumps were collected from several petrochemical production enterprises. The data types contained in the current historical data include: installation subject, manufacturer, rated flow, rated head, motor rated power, motor rated voltage, motor rated current, medium specific gravity, motor power factor, inlet pressure, outlet pressure, mass flow, volume flow, operating voltage, operating current and operating efficiency. Compared with the method of measuring the energy efficiency of pumps using the traditional formula of pump energy efficiency, the key parameters of motor efficiency are missing, and auxiliary parameters such as installation subject, manufacturer, rated flow, rated head, motor rated power, motor rated voltage, etc. that are easy to collect are added.

[0084] In this embodiment, when the motor efficiency cannot be obtained on site, the energy efficiency soft measurement model without the inlet pressure of the pump is constructed according to the method for evaluating the energy efficiency of the pump described in the first embodiment.

[0085] First, the collected historical data is cleaned to remove erroneous data.

[0086] Next, the historical data that has completed data cleaning is used as the remaining data, and the non-numerical data in the remaining data is standardized and normalized. The installation entity and the manufacturer of the pump are standardized, and the same type of installation entity and the same pump manufacturer are standardized with the same name, and the source categories with less than 5% of the total number of samples are uniformly classified as others. Among them, in the process of standardizing the non-numerical data, label coding is used for the pump installation entity and manufacturer, and the auxiliary parameters of different source categories are represented by numbers from 0 to n (category types). In the process of normalizing the numerical data, the maximum-minimum value normalization method is used to achieve the normalization of the numerical data.

[0087] Before the remaining data are uniformly normalized and normalized, a null value analysis is performed on the remaining data in the manner described in Example 1, and the corresponding null value data is assigned to zero according to the analysis result to eliminate the influence of the null value phenomenon on the model training result.

[0088] This embodiment uses a back propagation neural network BPNN. The framework of the first preset model connects each layer in a node connection manner. The layers include a fully connected input layer, an output layer, three hidden layers, three BN layers, and a dropout layer. Among them, the input layer nodes are 15, the output layer node is 1, the hidden layer nodes are 64, 64, and 32 respectively, and the initial learning rate is 10 -3 , the number of iterations is 200. Based on the initial learning rate, number of iterations and activation function configured for the current back propagation neural network BPNN, the historical operation data in the training data set is used as the input data of the input layer, and the number of actual data types contained in the historical operation data is used as the training dimension, combined with the adaptive adjustment of the learning rate of the back propagation neural network BPNN, so as to realize the training of the first preset model.

[0089] The activation function of this embodiment adopts the ReLu function.

[0090] Furthermore, the training data set of this embodiment is composed of 80% of the data in the extracted data, and the other 20% of the data in the extracted data other than the training data set (the data in the extracted data not selected into the training data set) is constructed as a test data set. By calculating the mean square error and mean absolute error percentage between the predicted energy efficiency data predicted by using the actual operation data in the test data set and the actual energy efficiency data directly extracted from the test data set, it is determined that the energy efficiency value predicted by the currently obtained energy efficiency soft measurement model is not much different from the actual value in the absence of the pump inlet pressure. Figure 3Schematic diagram of the evaluation result of the missing motor efficiency of the method for evaluating the energy efficiency of a pump according to an embodiment of the present application. Figure 3 , this embodiment takes the average value of 20 model training and test results, and obtains a mean square error of 0.0034 and a mean absolute error percentage of 4.5%. It can be seen that the measurement result accuracy of the energy efficiency soft measurement model obtained by the present invention is high and can meet the measurement accuracy requirements in actual engineering applications.

[0091] Embodiment 4

[0092] Based on the method for evaluating the energy efficiency of a pump described in the first embodiment, the embodiment of the present invention further provides a system for evaluating the energy efficiency of a pump. Figure 4 The overall structure diagram of the system for evaluating the energy efficiency of a pump according to the embodiment of the present application is shown in FIG. The structure and function of the system for evaluating the energy efficiency of a pump are described in detail below in conjunction with the embodiment of the present application.

[0093] The system for evaluating the energy efficiency of a pump according to the present invention at least comprises: a historical data acquisition module 41, a real-time data acquisition module 42, a model generation module 43 and an energy efficiency evaluation module 44. Specifically, the historical data acquisition module 41 is used to collect historical data of a plurality of pumps, wherein the historical data comprises historical operation data and corresponding historical energy efficiency data related to energy efficiency of each pump. The real-time data acquisition module 42 is used to acquire the operation data of the pump to be evaluated in the current period related to energy efficiency in real time, and determine the actual data type contained in the operation data of the current period. The model generation module 43 is used to extract the first type of operation data and the first type of energy efficiency data related to the actual data type determined by the real-time data acquisition module 42 from the historical data collected by the historical data acquisition module 41, construct a training data set, and train the first preset model using the training data set, so as to obtain an energy efficiency soft measurement model for evaluating the energy efficiency of the pump in the next period. The energy efficiency evaluation module 44 is used to obtain the corresponding energy efficiency evaluation result according to the operation data of the next period acquired by the real-time data acquisition module 42 and the energy efficiency soft measurement model generated by the model generation module 43.

[0094] The present invention proposes a method and system for evaluating the energy efficiency of a machine pump. The method first collects historical data of multiple machine pumps, and then uses the collected historical data, combined with the real-time data of the machine pump to be evaluated, to establish an energy efficiency soft measurement model based on machine learning technology for evaluating the energy efficiency of the machine pump in the next period; finally, based on the established energy efficiency soft measurement model, the operating data of the machine pump to be evaluated in the next period is used to obtain the energy efficiency evaluation result corresponding to the period. The present invention overcomes the defects of traditional methods that rely on expert experience knowledge to set key parameters and construct complex mechanism models. It does not require expert experience knowledge to construct a fluid mechanics model or other mechanism model based on physical knowledge, and obtains a data-driven energy efficiency soft measurement model. It realizes accurate evaluation of the energy efficiency of the machine pump under the condition of missing some machine pump operating parameters, and has the advantages of high intelligence and good online application performance.

[0095] The above description is only a specific implementation case of the present invention, and the protection scope of the present invention is not limited thereto. Any modification or replacement of the present invention by any technician familiar with the present technology within the technical specifications described in the present invention should be within the protection scope of the present invention.

[0096] Of course, the present invention may have many other embodiments. Without departing from the spirit and essence of the present invention, those skilled in the art may make various corresponding changes and modifications based on the present invention, but these corresponding changes and modifications should all fall within the scope of protection of the claims of the present invention.

[0097] It should be understood by those skilled in the art that the modules or steps of the present invention described above can be implemented by a general-purpose computing device, they can be concentrated on a single computing device, or distributed on a network composed of multiple computing devices, and optionally, they can be implemented by a program code executable by a computing device, so that they can be stored in a storage device and executed by the computing device, or they can be made into individual integrated circuit modules, or multiple modules or steps therein can be made into a single integrated circuit module for implementation. Thus, the present invention is not limited to any specific combination of hardware and software.

[0098] Although the embodiments disclosed in the present invention are as above, the above contents are only embodiments adopted for facilitating the understanding of the present invention and are not intended to limit the present invention. Any technician in the technical field to which the present invention belongs can make any modifications and changes in the form and details of the implementation without departing from the spirit and scope disclosed in the present invention, but the patent protection scope of the present invention shall still be subject to the scope defined in the attached claims.

Claims

1. A method for evaluating the energy efficiency of a pump, characterized in that: include: Collecting historical data of a plurality of pumps, wherein the historical data includes historical operation data related to energy efficiency and corresponding historical energy efficiency data of each pump; Acquire the operating data of the pump to be evaluated in the current period related to energy efficiency in real time, and determine the actual data type contained in the operating data of the current period; Extracting first-category operation data and first-category energy efficiency data related to the actual data type from the historical data, constructing a training data set, and using the training data set to train a first preset model, thereby obtaining an energy efficiency soft measurement model for evaluating the energy efficiency of the machine pump in the next period; According to the operation data of the next period, the energy efficiency soft measurement model is used to obtain the corresponding energy efficiency evaluation result.

2. The method according to claim 1, characterized in that The method further comprises: The collected historical data is first cleaned to remove erroneous data, and then the non-numerical data in the remaining data is uniformly normalized, and the numerical data in the remaining data is normalized, so as to use the uniformly normalized non-numerical data and normalized numerical data to train the first preset model.

3. The method according to claim 2, characterized in that The process of normalizing non-numeric data includes: Classifying the non-numerical data according to data sources to form multiple groups of data sequences corresponding to different data sources; Each group of data sequences is encoded according to different encoding methods to obtain encoded non-numerical data.

4. The method according to claim 3, characterized in that The encoding method is determined according to the statistical results of the number of source categories corresponding to different elements in each group of data sequences, including but not limited to: onehot encoding, mean encoding, label encoding, Bayesian Target encoding and Leave-one-out encoding, where: When the number of source categories of the corresponding group is less than or equal to 5, the onehot encoding is used, or, when the number of source categories of the corresponding group is greater than 5, the mean encoding or the label encoding is used.

5. The method according to any one of claims 2 to 4, characterized in that: The normalization processing method adopted is selected from one of maximum-minimum value normalization and Z-Score normalization.

6. The method according to any one of claims 2 to 5, characterized in that: Before performing unified normalization and normalization processing on the remaining data, the method further includes: Null value analysis is performed on the remaining data, and the corresponding null value data is assigned to zero according to the analysis result to eliminate the influence of the null value phenomenon on the model training result.

7. The method according to any one of claims 1 to 6, characterized in that: The first preset model is trained using a machine learning regression algorithm, and the machine learning regression algorithm is selected from one of a back propagation neural network BPNN, a lightweight and efficient gradient boosting tree LGBM, an extreme gradient boosting XGBoost, an extreme learning machine ELM and a KNN algorithm.

8. The method according to claim 7, characterized in that The method further comprises: When the back propagation neural network BPNN is used, the framework of the first preset model connects each layer in a node connection manner, and the layers include a fully connected input layer, an output layer, three hidden layers, three BN layers and a dropout layer, wherein, The input data of the input layer is the historical operation data in the training data set, and its dimension is the number of actual data types contained in the historical operation data, where: During the training process, the training of the first preset model is completed by configuring the initial learning rate, the number of iterations and the activation function for the back propagation neural network BPNN, and the activation function adopts the ReLu function.

9. The method according to any one of claims 1 to 8, characterized in that: The training data set is composed of 80% of the extracted data. After obtaining the energy efficiency soft measurement model, the method further includes: The data extracted but not included in the training data set is constructed as a test data set, and the test data set is used to perform error analysis on the energy efficiency soft measurement model, so as to determine the adaptability of the current energy efficiency soft measurement model to the actual application according to the analysis result.

10. The method according to claim 9, characterized in that The process of error analysis of the energy efficiency soft sensor model includes: The error analysis result of the energy efficiency soft measurement model is obtained by calculating the mean square error and mean absolute error percentage between the predicted energy efficiency data predicted by using the actual operation data in the test data set and the actual energy efficiency data directly extracted from the test data set.

11. The method according to any one of claims 1 to 10, characterized in that: The data types included in the historical operation data include but are not limited to: installation entity, manufacturer, rated flow, rated head, motor rated power, motor rated voltage, motor rated current, medium specific gravity, operating years, motor power factor, motor efficiency, inlet pressure, outlet pressure, mass flow, volume flow, operating voltage, operating current and operating efficiency.

12. A system for evaluating the energy efficiency of a pump, characterized in that: The system includes the following modules: A historical data acquisition module, which is used to collect historical data of multiple pumps, wherein the historical data includes historical operation data related to energy efficiency and corresponding historical energy efficiency data of each pump; A real-time data acquisition module, which is used to acquire in real time the operating data of the pump to be evaluated in the current period related to the energy efficiency, and determine the actual data type contained in the operating data in the current period; A model generation module, which is used to extract the first type of operation data and the first type of energy efficiency data related to the actual data type from the historical data, construct a training data set, and use the training data set to train the first preset model, so as to obtain an energy efficiency soft measurement model for evaluating the energy efficiency of the machine pump in the next period; The energy efficiency evaluation module is used to obtain corresponding energy efficiency evaluation results based on the operation data of the next period of time using the energy efficiency soft measurement model.

Images