A pump pressure pulsation prediction system and method based on digital twin

Abstract

The present invention discloses a pump pressure pulsation prediction system and method based on digital twin, wherein the system includes: a pump unit physical model, a sensor module, an acquisition module, a construction module and a host computer; the sensor module is used to measure the real pump parameters of the pump unit physical model; the acquisition module is used to acquire the real pump parameters; the construction module is used to construct a pump unit virtual model based on the real pump parameters; the host computer uses the pump unit virtual model to predict pump pressure pulsation. The present invention adopts digital twin technology to create a virtual model of a physical object in a digital form to simulate its behavior in a real environment. It can simulate the operating flow state and pressure pulsation state of a pump station unit under different flow conditions. The present invention is simple to calculate and operate, has strong repeatability and high accuracy, and can effectively make up for the shortcomings of the physical model.

Description

Technical Field

[0001] The present invention relates to the field of pump pressure pulsation prediction, and in particular to a pump pressure pulsation prediction system and method based on digital twin. Background Art

[0002] At a time when renewable energy sources such as wind power, photovoltaics, and ocean energy are developing vigorously, it is very necessary to continue to develop hydropower and pumped storage with fast and large-capacity regulation capabilities. Since fluid machinery has the dual functions of power production and energy use, it is urgent to study advanced fluid machinery technology that adapts to the characteristics of future energy systems.

[0003] Fluid machinery is an indispensable core equipment in contemporary industrial production and energy conversion, including pumps, fans, compressors, valves, etc. At present, fluid machinery has been integrated with technologies such as the Internet of Things, big data, and artificial intelligence. With the improvement of green energy efficiency requirements, the fluid machinery industry will focus more on high efficiency, environmental protection, and intelligence. In the future, fluid machinery will achieve a higher degree of automation and intelligence. At the same time, the exploration of new fluid machinery materials and working principles will promote a qualitative leap in the technical level of the entire industry.

[0004] The healthy operation of pump units has always been a key research direction in the fluid machinery industry, and the operating status of pump units is closely related to the vibration and pressure pulsation of the units. By studying the relationship between the operating conditions and pressure pulsation of the pump under different settings, it is possible to effectively determine whether the pump unit is in a healthy operating state, conduct fault diagnosis in a timely manner at the early stage of abnormal pulsation, reduce damage to unit components, and prevent accidents caused by component damage. Studying pressure pulsation signals is of great significance for monitoring the healthy and normal operation of the unit and predicting unit failure operation. Summary of the invention

[0005] The purpose of the present invention is to provide a pump pressure pulsation prediction system based on digital twins in view of the deficiencies of the above-mentioned prior art. Based on the digital twin model, a three-dimensional modeling software is used to build a model of the pump station unit in equal dimensions, and a multi-condition calculation is performed on the pump station unit model through a numerical solution method. By setting monitoring points, the pressure pulsation at special positions of the unit is effectively monitored and collected, the monitoring data is processed and the prediction model is trained using virtual model data, and finally the trained prediction model is applied to the physical model using transfer learning, and finally the pressure pulsation signal prediction is realized. It makes up for the singleness and insufficiency of the physical model test and the insufficiency of the training model requiring a large amount of physical experimental data, and provides guidance for the safe and stable operation management of the pump station unit and avoiding the occurrence of abnormal pressure pulsation.

[0006] To achieve the above-mentioned object, the present invention provides a pump pressure pulsation prediction system based on digital twin, comprising: a pump unit physical model, a sensor module, an acquisition module, a construction module and a host computer;

[0007] The sensor module is used to measure the real pump parameters of the physical model of the pump unit;

[0008] The acquisition module is used to acquire the real pump parameters;

[0009] The construction module is used to construct a virtual model of the pump unit based on the real pump parameters;

[0010] The host computer uses the virtual model of the pump unit to predict pump pressure pulsation.

[0011] Preferably, the sensor module includes: a pressure sensor, a flow meter and a rotation speed sensor;

[0012] The pressure sensor is used to measure the pressure of the inlet and outlet, guide vanes and impeller of the axial flow pump;

[0013] The flow meter is used to measure the flow of the axial flow pump;

[0014] The rotation speed sensor is used to measure the rotation speed of the axial flow pump.

[0015] Preferably, the acquisition module includes: an acquisition card and a data center;

[0016] The acquisition card is used to acquire the real pump parameters in real time;

[0017] The data center is used to store the real-time data collected by the acquisition card, the historical data and maintenance records of the axial flow pump.

[0018] Preferably, the host computer includes: an information portal module, a pressure pulsation prediction module and a control system;

[0019] The information portal module provides an administrator login platform and a human-computer interaction platform;

[0020] The pressure pulsation prediction module is used to predict pump pressure pulsation based on the pump unit virtual model;

[0021] The control system is used to control the opening and closing of valves, the setting of flow rate and the setting of rotation speed in the physical model of the pump unit when the pressure pulsation predicts abnormal pressure pulsation.

[0022] Preferably, the workflow of the construction module includes: establishing a numerical model of the axial flow pump through three-dimensional modeling software, and then constructing the virtual model of the pump unit based on the numerical model of the axial flow pump, the performance parameters of each component of the axial flow pump, and the fluid calculation model, the structural calculation model, and the degradation behavior and wear mechanism of the axial flow pump.

[0023] Preferably, the established numerical model is meshed by using a meshing technique, and a steady-state calculation is performed on the meshed numerical model to ensure that the convergence accuracy of the calculated residual value is less than 1.0×10 -5 On the basis of steady calculation, monitoring points are set for the numerical model, including the inlet flow channel monitoring point, the outlet flow channel monitoring point, the impeller monitoring point, and the guide vane monitoring point. The sampling time and sampling frequency are also set, and the instantaneous pressure data within the sampling time period is obtained through unsteady calculation.

[0024] Preferably, the pressure pulsation prediction module predicts pump pressure pulsation by training an optimized prediction model; the steps include: first decomposing and filtering the obtained pressure pulsation signal of the virtual model of the pump unit through variational mode decomposition, reconstructing the intrinsic mode function components, and extracting frequency domain data from the signal; introducing the sparrow optimization algorithm that integrates Osprey and Cauchy variations to optimize the parameters of variational mode decomposition; and then using the long short-term memory network model to model and predict the signal.

[0025] Preferably, after the prediction model is constructed, the physical model of the pump unit is used as the target domain, and the virtual model of the pump unit is used as the source domain;

[0026] A model-based migration method is used to realize the shared parameters of the target domain model and the source domain pseudo model, and the prediction model trained in the source domain model is applied to the target domain model. By updating and reusing its parameters, the pressure pulsation prediction in the target domain model is realized.

[0027] Preferably, the model-based migration method includes: fine-tuning of the deep neural network, the fine-tuning strategy re-trains and optimizes the migration parameters on the target domain model to achieve pressure pulsation prediction.

[0028] The present invention also provides a pump pressure pulsation prediction method based on digital twins, which is applied to the above system and comprises the following steps:

[0029] S1. Measure the real pump parameters of the physical model of the pump unit;

[0030] S2. Collecting the real pump parameters;

[0031] S3. Based on the real pump parameters, construct a virtual model of the pump unit;

[0032] S4. Utilize the virtual model of the pump unit to predict pump pressure pulsation.

[0033] Compared with the prior art, the present invention has the following beneficial effects:

[0034] The present invention uses digital twin technology to create a virtual model of a physical object in a digital form to simulate its behavior in a real environment. It can simulate the flow state and pressure pulsation state of the pump station unit under different flow conditions. The present invention is simple to calculate and operate, has strong repeatability and high accuracy, can effectively make up for the shortcomings of the physical model, and can effectively predict the pressure pulsation state of the pump station unit in actual operation, and improve the safety and stability of the pump station unit operation. The pressure pulsation prediction platform integrates control systems, information portals, and pressure pulsation prediction modules, combining automation and intelligence with pumps, greatly saving human resources. BRIEF DESCRIPTION OF THE DRAWINGS

[0035] In order to more clearly illustrate the technical solution of the present invention, the following briefly introduces the drawings required for use in the embodiments. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative labor.

[0036] Figure 1 A schematic diagram of a method flow of an embodiment of the present invention;

[0037] Figure 2 A schematic diagram of a process for constructing a virtual model of a pump unit according to an embodiment of the present invention;

[0038] Figure 3 This is a schematic diagram of the structure of the host computer according to an embodiment of the present invention;

[0039] Figure 4 A schematic diagram of a pressure pulsation prediction method according to an embodiment of the present invention;

[0040] Figure 5 Schematic diagram of the workflow of the prediction model according to an embodiment of the present invention. DETAILED DESCRIPTION

[0041] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0042] In order to make the above-mentioned objects, features and advantages of the present invention more obvious and easy to understand, the present invention is further described in detail below with reference to the accompanying drawings and specific embodiments.

[0043] Embodiment 1

[0044] like Figure 1As shown, it is a schematic diagram of the system structure of the present invention, including: a physical model of a pump unit, a sensor module, an acquisition module, a construction module and a host computer; the sensor module is used to measure the real pump parameters of the physical model of the pump unit; the acquisition module is used to acquire the real pump parameters; the construction module is used to construct a virtual model of the pump unit based on the real pump parameters; the host computer uses the virtual model of the pump unit to predict the pump pressure pulsation.

[0045] The following will explain in detail how the present invention solves technical problems in real life in conjunction with this embodiment.

[0046] Firstly, an axial flow pump device of any model is selected as the physical model of the pump unit, and then the device is connected to the sensor module, which includes a pressure sensor, a flow meter and a speed sensor. The pressure sensor is used to measure the pressure of the inlet and outlet, guide vanes and impeller of the axial flow pump; the flow meter is used to measure the flow rate of the axial flow pump; and the speed sensor is used to measure the speed of the axial flow pump.

[0047] The acquisition card in the acquisition module uses a USB-6009 data acquisition card with a sampling frequency of 48KS / s. The acquisition card is used to collect the real pump parameters measured by the sensor module in real time. In addition, the acquisition module also includes a data center for storing the real-time data collected by the acquisition card, the historical data and maintenance records of the axial flow pump equipment, which are used to provide reference for the subsequent construction of the digital twin model. Specifically, the data center stores the collected data for real-time prediction of the synchronous operation of the digital twin and simulation, and can also be used to mine the rules of historical data. The data center uses NavicatformySQL, and through the joint use of python and mySQL, the database stores data in real time.

[0048] After the real pump parameters are collected, they are transmitted to the construction module through the data center, and the construction module builds a virtual model of the pump unit. The premise of using digital twin technology for pump-related services is to establish a high-precision digital twin model.

[0049] Afterwards, if Figure 2 As shown, the construction module establishes a numerical model of the axial flow pump equipment through three-dimensional modeling software, and then constructs a virtual model of the pump unit based on the three-dimensional model of the axial flow pump and the performance parameters of each component of the axial flow pump. After the virtual model of the pump unit is built, the data obtained by the model is analyzed and corrected with the real pump parameters collected by the sensor module.

[0050] In this embodiment, in order to improve the accuracy of the numerical results, the meshing technology is used to mesh the established numerical model. The CFD calculation software is used to perform steady-state calculations on the meshed numerical model to ensure that the convergence accuracy of the calculated residual value is less than 1.0×10 -5, based on the steady calculation, the monitoring points of the numerical model are set, including the inlet flow channel monitoring point, the outlet flow channel monitoring point, the impeller monitoring point, and the guide vane monitoring point, and the sampling time and sampling frequency are set, and the instantaneous pressure data within the sampling time period is obtained through unsteady calculation. The virtual model of the pump unit is constructed by 3D modeling software according to the actual size of the actual operating physical model of the pump unit; the constructed virtual model of the pump unit is connected to the physical model of the pump unit through the data center, and the data center is also used to store the measurement parameters of the virtual model of the pump unit.

[0051] like Figure 3 As shown, the host computer includes an information portal module, a pressure pulsation prediction module and a control system. The information portal module provides an administrator login platform and a human-computer interaction platform; the control system is used to control the opening and closing of valves, flow rate setting, and speed setting in the physical model of the pump unit when the pressure pulsation predicts abnormal pressure pulsation.

[0052] After the virtual model of the pump unit is built, the pressure pulsation prediction module is used to predict the pressure pulsation of the physical model of the pump unit through the virtual model of the pump unit. The method flow is as follows: Figure 4 The specific steps include:

[0053] First, a fixed flow rate Q is set, and the acquisition module is used to collect the instantaneous pressure data of the digital pump unit virtual model (source domain model) and the pump unit physical model (target domain model) at each monitoring point during the sampling time period under the settings of 1.0Q, 0.4Q, 0.6Q, and 0.8Q, respectively, to form the source domain pressure pulsation signal and the target domain pressure pulsation signal, respectively. The LabVIEW program can realize real-time data acquisition through the acquisition card and communication equipment through its own DAQ acquisition function.

[0054] Afterwards, if Figure 5 As shown in the figure, an optimized prediction model OCSSA-VMD-LSTM (Sparrow optimization algorithm integrating Osprey and Cauchy variation-Variational mode decomposition-Long short-term memory network) is trained, and the pressure pulsation signal of the obtained source domain model is first decomposed and filtered by variational mode decomposition, the intrinsic mode function component is reconstructed, the frequency domain data is extracted from the signal, and then the signal is modeled and predicted using the long short-term memory network model. The sparrow optimization algorithm integrating Osprey and Cauchy variation is introduced to optimize the parameters of variational mode decomposition, which solves the limitation of arbitrary selection on the prediction performance of the model and improves the prediction accuracy of the model.

[0055] Specifically, the specific steps of variational mode decomposition construction are:

[0056] 1) Initialize the original parameters, such as the modal function set Lagrange multipliers Instantaneous frequency set And the number of cycles n=0.

[0057] 2) Set n=n+1 and enter the loop.

[0058] 3) When the number of decompositions is less than the specified number of decompositions k, update the relevant parameters according to the update formula:

[0059]

[0060] Among them, u k represents each mode function; f is the signal; ω is the frequency; ω k is the instantaneous frequency; u i is the modal function of the i-th layer; ε is the Lagrange multiplier; α represents the penalty factor.

[0061]

[0062] The Lagrange multiplier ε is updated as follows:

[0063]

[0064] Among them, τ represents the noise tolerance; ε k represents the updated Lagrange multiplier.

[0065] Given precision θ>0, stop iteration when the following equation is satisfied, otherwise return to step 2).

[0066]

[0067] Among them, K represents the number of decomposed modes; θ represents the given discrimination accuracy; and α is used to discriminate the smoothness and decomposition accuracy of the signal. These two parameters have an important influence on the decomposition results of variational mode decomposition, so the sparrow optimization algorithm integrating Osprey and Cauchy mutation is used to optimize the parameter mode number K and penalty factor α of variational mode decomposition. Determine the optimal parameter combination [K, α].

[0068] The sparrow optimization algorithm combining Osprey and Cauchy mutation is used to optimize the variational mode decomposition parameters. The specific steps are as follows:

[0069] Initialize parameters and ranges, calculate the initial population fitness value, and construct the fitness function of all sparrow populations. The discoverer provides the foraging direction for the joiner. The discoverer's position update is as follows:

[0070]

[0071] Where t represents the current iteration number, j ranges from {1, 2, ...d}, Iter represents the value of the jth dimension of the i-th sparrow at the t-th iteration; maxrepresents the maximum number of iterations; β∈[0,1] is a random number, ST∈[0.5,1] is a safety value; N2∈[0,1] represents the warning value range; M represents a 1×d matrix, and all the elements in it are 1, Q represents a random number, and it obeys the normal distribution; Indicates the finder's position after iterative updates.

[0072] In the process of searching for food, the position of the joiner changes with the position of the discoverer until the required food is found. The joiner position update description is as follows:

[0073]

[0074] Where n is in the range of {1,2,...,d}, Y p is the optimal position occupied by the discoverer, Y worst represents the worst position of the current global, A and L both represent a 1×d matrix, and A + =A T (AA T ) -1 ; When i>n / 2, it indicates that the i-th joiner with a poor fitness value is most likely to starve.

[0075] Based on the sparrow algorithm, the global exploration strategy of the osprey optimization algorithm in the first stage is used to replace the original sparrow algorithm's discoverer position update formula. The discoverer position update method in the sparrow algorithm is updated based on the osprey's movement simulation method toward the fish. The global exploration strategy formula of the osprey optimization algorithm in the first stage is as follows:

[0076]

[0077] In the formula, x i，j Represents the value of the jth dimension of the i-th sparrow; It represents the position of the finder after iterative update; SF is the finder position selected by the joiner, r is a random number between [0, 1], and the value of I is one of {1, 2}.

[0078] The Cauchy mutation strategy is used to replace the joiner position update formula of the original sparrow algorithm.

[0079] x newbest =x best +x best ×Cauchy(0,1)

[0080] In the formula, x best is the position of the joiner before the iteration update, x newbest is the position of the joiner after iterative update; Cauchy(0,1) is the standard Cauchy distribution function.

[0081] Finally, the long short-term memory network model is used to model and predict the signal. The specific steps are as follows:

[0082] Set the switching value of the forget gate, which determines whether part of the information from the previous state needs to be discarded in the input of this step:

[0083] f t =σ(W f x t +U f h t-1 +b f )

[0084] In the formula, f t is the forget gate; x t is the current input vector; h t is the current hidden layer vector; b f , U f , W f They represent the loop weights of bias, input weight, and forget gate respectively; σ represents the sigmoid function.

[0085] The information stored in the storage unit is updated using the switch values ​​of the input gate and the forget gate:

[0086] i t =σ(W i x t +U i h t-1 +b i )

[0087]

[0088] In the formula, i t is the input gate, is the candidate vector of memory cells; h t-1 is the hidden state of the previous time state; W i , U i and b i Represents the switch value of the input gate, the weight matrix of the input gate, and the bias vector used to calculate the current time step; tanh is the activation function, W c , U c 、b c is the switch value of the input gate in the memory state, the weight matrix of the input gate, and the bias vector.

[0089] Update the cell state by combining the input gate and forget gate information:

[0090]

[0091] In the formula, C tC is the memory state; t-1 It is the memory status of the previous moment.

[0092] o t =σ(W o [h t-1 , x t ]+b o )

[0093] h t =o t *tanh(C t )

[0094] In the formula, o t is the output gate; h t is the hidden state; W o and b o are the switch value and bias vector used to calculate the output gate of the current time step.

[0095] After the prediction model is built, the model-based migration method is used to realize the shared parameters of the target domain model and the source domain model. The above model trained in the source domain model is applied to the target domain model. By updating and reusing its parameters, the pressure pulsation prediction in the target domain is realized.

[0096] The model-based migration method is used to fine-tune the deep neural network. This method updates and reuses the parameters of the model trained in the source domain to make it work in the target domain. Specifically, the fine-tuning strategy retrains and optimizes the migration parameters on the target domain data to achieve pressure pulsation prediction. This method can effectively migrate the network parameters in the source domain to the target domain to reduce the dependence of the model on the data in the target domain.

[0097] Embodiment 2

[0098] This embodiment also provides a pump pressure pulsation prediction method based on digital twins, the steps comprising:

[0099] S1. Measure the real pump parameters of the physical model of the pump unit.

[0100] Firstly, an axial flow pump device of any model is selected as the physical model of the pump unit, and then the device is connected to sensors, including pressure sensor, flow meter and speed sensor; among them, the pressure sensor is used to measure the pressure of the inlet and outlet, guide vane and impeller of the axial flow pump; the flow meter is used to measure the flow rate of the axial flow pump; the speed sensor is used to measure the speed of the axial flow pump.

[0101] S2. Collect real pump parameters.

[0102] A USB-6009 data acquisition card with a sampling frequency of 48KS / s is used as the acquisition card. The real pump parameters measured by the sensor are collected in real time using this acquisition card. Afterwards, a data center is established to store the real-time data collected by the acquisition card, the historical data and maintenance records of the axial flow pump equipment, which are used to provide reference for the subsequent construction of the digital twin model. Specifically, the data center stores the collected data for real-time prediction of the synchronous operation of the digital twin and simulation, and can also be used to mine the rules of historical data. The data center uses Navicat for mySQL, and through the joint use of python and mySQL, the database stores data in real time.

[0103] After the real pump parameters are collected, they are transmitted to the host computer through the data center, and the host computer builds a virtual model of the pump unit. The premise of using digital twin technology for pump-related services is to establish a high-precision digital twin model.

[0104] S3. Build a virtual model of the pump unit based on the real pump parameters.

[0105] A numerical model of the axial flow pump equipment is established through three-dimensional modeling software, and then a virtual model of the pump unit is constructed based on the three-dimensional model of the axial flow pump and the performance parameters of each component of the axial flow pump. After the virtual model of the pump unit is completed, the data obtained by the model is analyzed and corrected with the actual pump parameters collected by the sensor module.

[0106] In this embodiment, in order to improve the accuracy of the numerical results, the meshing technology is used to mesh the established numerical model. The CFD calculation software is used to perform steady-state calculations on the meshed numerical model to ensure that the convergence accuracy of the calculated residual value is less than 1.0×10 -5 , based on the steady calculation, the monitoring points of the numerical model are set, including the inlet flow channel monitoring point, the outlet flow channel monitoring point, the impeller monitoring point, and the guide vane monitoring point, and the sampling time and sampling frequency are set, and the instantaneous pressure data within the sampling time period is obtained through unsteady calculation. The virtual model of the pump unit is constructed by 3D modeling software according to the actual size of the actual operating physical model of the pump unit; the constructed virtual model of the pump unit is connected to the physical model of the pump unit through the data center, and the data center is also used to store the measurement parameters of the virtual model of the pump unit.

[0107] S4. Use the virtual model of the pump unit to predict the pump pressure pulsation.

[0108] First, a fixed flow rate Q is set, and the acquisition module is used to collect the instantaneous pressure data of the digital pump unit virtual model (source domain model) and the pump unit physical model (target domain model) at each monitoring point during the sampling time period under the settings of 1.0Q, 0.4Q, 0.6Q, and 0.8Q, respectively, to form the source domain pressure pulsation signal and the target domain pressure pulsation signal, respectively. The LabVIEW program can realize real-time data acquisition through the acquisition card and communication equipment through its own DAQ acquisition function.

[0109] Afterwards, an optimized prediction model OCSSA-VMD-LSTM (Sparrow optimization algorithm integrating Osprey and Cauchy variation-Variational mode decomposition-Long short-term memory network) was trained. The pressure pulsation signal of the source domain model was first decomposed and filtered by variational mode decomposition, the intrinsic mode function components were reconstructed, the frequency domain data was extracted from the signal, and then the signal was modeled and predicted using the long short-term memory network model. The Sparrow optimization algorithm integrating Osprey and Cauchy variation was introduced to optimize the parameters of variational mode decomposition, which solved the limitation of arbitrary selection on the model prediction performance and improved the prediction accuracy of the model.

[0110] Specifically, the specific steps of variational mode decomposition construction are:

[0111] 1) Initialize the original parameters, such as the modal function set Lagrange multipliers Instantaneous frequency set And the number of cycles n=0.

[0112] 2) Set n=n+1 and enter the loop.

[0113] 3) When the number of decompositions is less than the specified number of decompositions k, update the relevant parameters according to the update formula:

[0114]

[0115] Among them, u k represents each mode function; f is the signal; ω is the frequency; ω k is the instantaneous frequency; u i is the modal function of the i-th layer; ε is the Lagrange multiplier; α represents the penalty factor.

[0116]

[0117] The Lagrange multiplier ε is updated as follows:

[0118]

[0119] Among them, τ represents the noise tolerance; ε k represents the updated Lagrange multiplier.

[0120] Given precision θ>0, stop iteration when the following equation is satisfied, otherwise return to step 2).

[0121]

[0122] Among them, K represents the number of decomposed modes; θ represents the given discrimination accuracy; and α is used to discriminate the smoothness and decomposition accuracy of the signal. These two parameters have an important influence on the decomposition results of variational mode decomposition, so the sparrow optimization algorithm integrating Osprey and Cauchy mutation is used to optimize the parameter mode number K and penalty factor α of variational mode decomposition. Determine the optimal parameter combination [K, α].

[0123] The sparrow optimization algorithm combining Osprey and Cauchy mutation is used to optimize the variational mode decomposition parameters. The specific steps are as follows:

[0124] Initialize parameters and ranges, calculate the initial population fitness value, and construct the fitness function of all sparrow populations. The discoverer provides the foraging direction for the joiner. The discoverer's position update is as follows:

[0125]

[0126] Where t represents the current iteration number, j ranges from {1, 2, ...d}, Iter represents the value of the jth dimension of the i-th sparrow at the t-th iteration; max represents the maximum number of iterations; β∈[0,1] is a random number, ST∈[0.5,1] is a safety value; N2∈[0,1] represents the warning value range; M represents a 1×d matrix, and all the elements in it are 1, Q represents a random number, and it obeys the normal distribution; Indicates the finder's position after iterative updates.

[0127] In the process of searching for food, the position of the joiner changes with the position of the discoverer until the required food is found. The joiner position update description is as follows:

[0128]

[0129] Where n is in the range of {1,2,...,d}, Y p is the optimal position occupied by the discoverer, Y worst represents the worst position of the current global, A and L both represent a 1×d matrix, and A + =A T (AA T ) -1 ; When i>n / 2, it indicates that the i-th joiner with a poor fitness value is most likely to starve.

[0130] Based on the sparrow algorithm, the global exploration strategy of the osprey optimization algorithm in the first stage is used to replace the original sparrow algorithm's discoverer position update formula. The discoverer position update method in the sparrow algorithm is updated based on the osprey's movement simulation method toward the fish. The global exploration strategy formula of the osprey optimization algorithm in the first stage is as follows:

[0131]

[0132] In the formula, x i，j Represents the value of the jth dimension of the i-th sparrow; It represents the position of the finder after iterative update; SF is the finder position selected by the joiner, r is a random number between [0, 1], and the value of I is one of {1, 2}.

[0133] The Cauchy mutation strategy is used to replace the joiner position update formula of the original sparrow algorithm.

[0134] x newbest =x best +x best ×Cauchy(0,1)

[0135] In the formula, x best is the position of the joiner before the iteration update, x newbest is the position of the joiner after iterative update; Cauchy(0,1) is the standard Cauchy distribution function.

[0136] Finally, the long short-term memory network model is used to model and predict the signal. The specific steps are as follows:

[0137] Set the switching value of the forget gate, which determines whether part of the information from the previous state needs to be discarded in the input of this step:

[0138] f t =σ(W f x t +U f h t-1 +b f )

[0139] In the formula, f t is the forget gate; x t is the current input vector; h t is the current hidden layer vector; b f , U f , W f They represent the loop weights of bias, input weight, and forget gate respectively; σ represents the sigmoid function.

[0140] The information stored in the storage unit is updated using the switch values ​​of the input gate and the forget gate:

[0141] i t =σ(W i x t +U i h t-1 +b i )

[0142]

[0143] In the formula, i t is the input gate, is the candidate vector of memory cells; h t-1 is the hidden state of the previous time state; W i , U i and b i Represents the switch value of the input gate, the weight matrix of the input gate, and the bias vector used to calculate the current time step; tanh is the activation function, W c , U c 、b c is the switch value of the input gate in the memory state, the weight matrix of the input gate, and the bias vector.

[0144] Update the cell state by combining the input gate and forget gate information:

[0145]

[0146] In the formula, C t C is the memory state; t-1 It is the memory status of the previous moment.

[0147] Finally, the output gate is responsible for updating the data to be output:

[0148] o t =σ(W o [h t-1 , x t ]+b o )

[0149] h t =o t *tanh(C t )

[0150] In the formula, o t is the output gate; h t is the hidden state; W o and b o are the switch value and bias vector used to calculate the output gate of the current time step.

[0151] After the prediction model is built, the model-based migration method is used to realize the shared parameters of the target domain model and the source domain model. The above model trained in the source domain model is applied to the target domain model. By updating and reusing its parameters, the pressure pulsation prediction in the target domain is realized.

[0152] The model-based migration method is used to fine-tune the deep neural network. This method updates and reuses the parameters of the model trained in the source domain to make it work in the target domain. Specifically, the fine-tuning strategy retrains and optimizes the migration parameters on the target domain data to achieve pressure pulsation prediction. This method can effectively migrate the network parameters in the source domain to the target domain to reduce the dependence of the model on the data in the target domain.

[0153] The embodiments described above are only descriptions of the preferred embodiments of the present invention and are not intended to limit the scope of the present invention. Without departing from the design spirit of the present invention, various modifications and improvements made to the technical solutions of the present invention by ordinary technicians in this field should all fall within the protection scope determined by the claims of the present invention.

Claims

1. A pump pressure pulsation prediction system based on digital twin, characterized in that: include: Pump unit physical model, sensor module, acquisition module, construction module and host computer; The sensor module is used to measure the real pump parameters of the physical model of the pump unit; The acquisition module is used to acquire the real pump parameters; The construction module is used to construct a virtual model of the pump unit based on the real pump parameters; The host computer uses the virtual model of the pump unit to predict the pump pressure pulsation; the host computer includes: an information portal module, a pressure pulsation prediction module and a control system; the information portal module provides an administrator login platform and a human-computer interaction platform; the pressure pulsation prediction module is used to predict the pump pressure pulsation based on the virtual model of the pump unit; the control system is used to control the opening and closing of the valve, the setting of the flow rate, and the setting of the speed in the physical model of the pump unit when the pressure pulsation predicts an abnormal pressure pulsation; the pressure pulsation prediction module predicts the pump pressure pulsation by training an optimized prediction model; the steps include: first decomposing and filtering the obtained pressure pulsation signal of the virtual model of the pump unit through variational mode decomposition, reconstructing the intrinsic mode function component, and extracting frequency domain data from the signal; introducing the sparrow optimization algorithm that integrates osprey and Cauchy variation to optimize the parameters of variational mode decomposition; and then using the long short-term memory network model to model and predict the signal; after the prediction model is built, the physical model of the pump unit is used as the target domain, and the virtual model of the pump unit is used as the source domain; A model-based migration method is used to realize the shared parameters of the target domain model and the source domain model, and the prediction model trained in the source domain model is applied to the target domain model. By updating and reusing its parameters, the pressure pulsation prediction in the target domain model is realized.

2. The pump pressure pulsation prediction system based on digital twin according to claim 1 is characterized in that: The sensor module includes: a pressure sensor, a flow meter and a rotation speed sensor; The pressure sensor is used to measure the pressure of the inlet and outlet, guide vanes and impeller of the axial flow pump; The flow meter is used to measure the flow of the axial flow pump; The rotation speed sensor is used to measure the rotation speed of the axial flow pump.

3. The pump pressure pulsation prediction system based on digital twin according to claim 1 is characterized in that: The acquisition module includes: an acquisition card and a data center; The acquisition card is used to acquire the real pump parameters in real time; The data center is used to store the real-time data collected by the acquisition card, the historical data and maintenance records of the axial flow pump.

4. The pump pressure pulsation prediction system based on digital twin according to claim 1, characterized in that: The workflow of the construction module includes: establishing a numerical model of the axial flow pump through three-dimensional modeling software, and then constructing a virtual model of the pump unit based on the numerical model of the axial flow pump, the performance parameters of each component of the axial flow pump, and the fluid calculation model, the structural calculation model, and the degradation behavior and wear mechanism of the axial flow pump.

5. The pump pressure pulsation prediction system based on digital twin according to claim 4 is characterized in that: The established numerical model is meshed using meshing technology, and the meshed numerical model is subjected to steady-state calculation to ensure that the convergence accuracy of the calculated residual value is less than 1.0×10 -5 On the basis of steady calculation, monitoring points are set for the numerical model, including the inlet flow channel monitoring point, the outlet flow channel monitoring point, the impeller monitoring point, and the guide vane monitoring point. The sampling time and sampling frequency are also set, and the instantaneous pressure data within the sampling time period is obtained through unsteady calculation.

6. The pump pressure pulsation prediction system based on digital twin according to claim 1, characterized in that: The model-based migration method includes: fine-tuning of the deep neural network, and the fine-tuning strategy re-trains and optimizes the migration parameters on the target domain model to achieve pressure pulsation prediction.

7. A pump pressure pulsation prediction method based on digital twin, the method is applied to the system according to any one of claims 1 to 6, characterized in that the steps include: S1. Measure the real pump parameters of the physical model of the pump unit; S2. Collecting the real pump parameters; S3. Based on the real pump parameters, construct a virtual model of the pump unit; S4. Utilize the virtual model of the pump unit to predict pump pressure pulsation.

Images