CN104598686A - Water pump motor modeling and optimizing method based on electromagnetic calculation and neural network - Google Patents

Abstract

The invention discloses a water pump motor modeling and optimization method based on electromagnetic calculation and neural network. According to the original structural parameters of the three-phase asynchronous motor used in the drainage pump, the output performance of the motor is calculated through the electromagnetic calculation method of the equivalent magnetic circuit of the motor. Parameters, including output torque, efficiency and motor stator starting current, etc. For pump performance, the BP neural network method is used to establish a hydraulic model to quickly fit the flow field model that is difficult to accurately calculate. Combining the motor mechanism model and the hydraulic fitting model to form an "integrated pump model". The simulated annealing genetic algorithm is used as an optimization method to optimize the design of some structural parameters of the motor. The main function of this method is to intelligently search the optimal motor structure design parameters under the premise of satisfying the input shaft power of the water pump, accurately match the performance of the motor and the load characteristics of the water pump, and reduce the margin coefficient of the power of the supporting motor to reduce the power of the supporting motor.

Description

Modeling and optimization method of water pump motor based on electromagnetic calculation and neural network

technical field

The invention relates to the field of water pump motor neural network modeling methods, in particular to a water pump motor modeling and optimization method based on electromagnetic calculation and neural network.

Background technique

A large amount of production wastewater is produced during the underground mining of mines, accompanied by a large amount of seepage groundwater. In order to ensure safe production in the mine, a large drainage pump must be used to discharge the water out of the mine in a timely and reliable manner. The long-term safe production of the mine requires that these drainage pumps must continue to run along with the entire production process of the mine. Therefore, the mine drainage pump system is one of the key equipment for the safety of mine production, and it is also one of the main electrical equipment in the mine. Electricity can account for up to 40% of the total electricity consumption of mines. The high-efficiency and energy-saving operation of the drainage pump is of great significance to the mine's energy saving, consumption reduction and cost control.

The drainage pump system is a structure that uses a motor to drive the pump impeller to rotate to generate centrifugal force to discharge the liquid along the pipeline. The power selection of the supporting motor of the water pump is usually determined by multiplying the shaft power at the design point of the water pump by an appropriate reserve factor. If the configuration of the motor and pump is unreasonable, the phenomenon of large horses and small carts will occur, and the energy of the motor will not be effectively utilized. At the same time, the efficiency of the motor used by the water pump itself is not high, and a large amount of unnecessary energy loss will also be generated. Therefore, the energy-saving direction of the drainage pump system mainly includes the performance optimization of the matching motor of the drainage pump and the power matching of the drainage pump and the matching motor.

The current work on the energy saving of the drainage pump system has the following problems:

(1) The mechanism model is used to calculate and analyze the hydraulic performance of the pump. Since the hydraulic mechanism model of the pump is a three-dimensional nonlinear multivariate dynamic model, the CFD flow field flow state calculation method must be used for simulation calculation and analysis. However, the flow field model of the water pump cannot be effectively combined with the electromagnetic model of the motor, which is not conducive to the overall performance analysis of the system;

(2) The optimization work of the motor is aimed at the performance index of the motor itself, without simultaneously examining how to reduce the power of the matching motor and improve the coupling matching degree between the performance of the motor and the load characteristics of the pump.

SUMMARY OF THE INVENTION The object of the present invention is to provide a water pump motor modeling and optimization method based on electromagnetic calculation and neural network, so as to solve the problems existing in the prior art.

In order to achieve the above object, the technical scheme adopted in the present invention is:

The water pump motor modeling and optimization method based on electromagnetic calculation and neural network is characterized in that it includes the following steps:

(1) According to the mechanism model of the submersible three-phase asynchronous motor for water pumps, write the motor forward electromagnetic calculation program;

(2) Using the BP neural network method to carry out data fitting modeling on the hydraulic model of the pump;

(3) The output result of the electromagnetic calculation of the motor is used as the input data of the hydraulic model of the pump, and the integrated model of the pump is formed by combining two parts of the program;

(4) Reversely fit the test data of the water pump, and realize the function of optimizing the design of the equipped motor according to the demand data obtained from the reverse fitting of the pump performance, the set margin coefficient target and the motor performance target. Form a pump motor reverse optimization system;

(5) Using the simulated annealing genetic algorithm, according to the demand data obtained from the reverse fitting of the pump performance, the set margin coefficient target and the motor performance target, and with the motor forward electromagnetic calculation program, the computer automatic Optimized design.

The described water pump motor modeling and optimization method based on electromagnetic calculation and neural network is characterized in that: in the step (1), the motor forward electromagnetic calculation program:

For the submersible motor used by the water pump, the mechanism modeling method is used to program the calculation process to form a motor electromagnetic calculation program that can operate independently;

The input data of the program includes: output power kW, power frequency Hz, line voltage V, stator connection, number of poles, pitch, number of stator and rotor slots, stator outer and inner diameter mm, air gap mm, rotor inner diameter mm, stator slot type Size mm, rotor slot size mm), number of turns per circle, number of parallel branches, number of parallel windings, slot insulation thickness mm, wire diameter, double-sided paint film mm, iron core length mm, lamination coefficient, and straight line part of the coil Protrusion length mm, average diameter of end ring mm, area mm^2, stray loss kW, iron loss coefficient, leakage reactance coefficient, etc.

The calculated output data include: efficiency, power factor, rated torque, maximum torque, starting torque, starting current, slot full rate;

On the basis of the electromagnetic calculation of the equivalent magnetic circuit method of the traditional three-phase asynchronous motor, according to the operating conditions of the submersible motor submerged in water and the structural characteristics of both the stator and the rotor using closed slots, the mechanical loss of the motor and the leakage permeance of the upper part of the stator and rotor slots In terms of the calculation method, it adopts a calculation method different from that of ordinary asynchronous motors;

The mechanical loss of the water pump motor is divided into three parts for calculation: the friction loss between the rotor and the cooling water, the friction loss of the thrust bearing and the friction loss of the guide bearing;

For the leakage permeance of the upper part of the closed slot of the stator and rotor of the pump motor, the method of equivalent slot width is used for calculation.

The water pump motor modeling and optimization method based on electromagnetic calculation and neural network is characterized in that: in the step (2), the BP neural network:

Two BP neural network 1 and BP neural network 2 with similar structures and different functions are established, which are respectively used to combine the motor forward electromagnetic calculation program to form an integrated model of the machine and pump, and combine the motor electromagnetic calculation and optimization algorithm to form the reverse direction of the pump motor. Optimal design system;

The pump hydraulic BP neural network model 1, which is used to combine the positive electromagnetic calculation program of the motor to form the integrated model of the machine and pump, adopts double input and double output, three-layer structure, 8 neurons in the hidden layer, gradient descent method to correct errors, and training termination conditions In order to train the structural characteristics with an error of less than 2%, the input shaft power and pump efficiency of the pump are used as the model input, and the head and flow data of the pump are output;

It is used to combine the motor forward electromagnetic calculation program and simulated annealing genetic algorithm to form the water pump hydraulic BP neural network model 2 of the reverse optimization system of the pump motor. It adopts three-input and single-output, three-layer structure, hidden layer 8 neurons, and gradient descent method. The error is corrected, and the training termination condition is the structural feature that the training error is less than 2%. The pump design flow rate, head, and target pump efficiency are used as inputs to calculate and output the predicted shaft power of the pump.

The water pump motor modeling and optimization method based on electromagnetic calculation and neural network is characterized in that: in the step (3), the machine-pump integration model:

After inputting the structural parameters of the motor for the forward electromagnetic calculation of the motor, the calculated result is transmitted to the BP neural network 1 that has been trained and converged by the measured samples, and finally outputs the output flow and head of the water pump under the current motor power state;

The input structure parameters used by the motor forward electromagnetic program, and the data used by the pump hydraulic BP neural network model 1 for neuron training, are stored in the folder in the form of txt files, and the program automatically reads them;

The machine-pump integration model can be calculated independently of the reverse optimization system of the pump motor described in step (4).

The water pump motor modeling and optimization method based on electromagnetic calculation and neural network is characterized in that: in the step (4), the reverse optimization system of the water pump motor:

Combining the motor forward electromagnetic calculation program, BP neural network 2 and simulated annealing genetic algorithm, the reverse optimization system of the water pump motor is formed;

After the BP neural network 2 is trained and converged according to the measured samples, when a set of flow rate, head and target pump efficiency data are input, it will fit and give the shaft power that the pump needs to input in this state, with the artificially given margin coefficient, the optimal output power of the motor that the pump should be equipped with in this state can be obtained; use this power as the power optimization design target of the motor, and cooperate with other indicators of efficiency and torque to perform multi-objective optimization of the pump motor Design; the optimization target is set as the motor efficiency, power factor, maximum torque, starting torque, starting current and the backup coefficient between the pump; the motor optimization variable selection: iron core length, number of conductors per slot, stator and rotor slots Some internal dimension parameters and rated output power;

In the multi-objective optimization design of water pump motors, it is generally difficult to find the optimal solution that satisfies all the objective requirements, so the non-inferior solutions that meet part of the optimization objectives obtained at the end of the optimization are output in txt format as the optimization results for manual selection;

The reverse optimization system of the water pump motor can be run independently from the integrated model of the machine and pump described in step (3) for calculation.

The invention relates to the technical field of three-phase asynchronous submersible motor optimization design and large-scale drainage pump water conservancy performance based on neural network modeling technology, through the forward electromagnetic calculation of the motor equipped with the water pump and the neural network fitting modeling of the hydraulic performance of the water pump, the pump is established Integrated model. At the same time, combined with the optimization algorithm, a reverse optimization design system for the pump motor is formed.

Compared with the prior art, the present invention has the beneficial effects that: the present invention adopts the neural network fitting method to model the water pump, and the established model is suitable for analyzing and calculating the overall performance of the system in combination with the motor model. Combined with the optimization algorithm, according to the principle of high matching of the power of the pump, the intelligent reverse optimization of the motor structure design parameters is carried out.

Description of drawings

Fig. 1 is a logical relationship diagram of the realization of the machine-pump integrated calculation and the reverse optimization function of the water pump motor in the present invention.

Figure 2 shows the BP neural network structure used in the pump hydraulic performance fitting model.

Figure 3 is the BP neural network structure used in the reverse optimization design system of the pump motor to reversely fit the test data of the pump.

Figure 4 is the flow chart of the overall performance of the computerized pump system of the machine-pump integration model.

Figure 5 is the process of optimizing the calculation of the equipped motor by the reverse optimization design system of the water pump motor.

Detailed ways

The modeling and optimization method of water pump motor based on electromagnetic calculation and neural network includes the following steps:

(1) According to the mechanism model of the submersible three-phase asynchronous motor for water pumps, write the motor forward electromagnetic calculation program;

(2) Using the BP neural network method to carry out data fitting modeling on the hydraulic model of the pump;

(3) The output result of the electromagnetic calculation of the motor is used as the input data of the hydraulic model of the pump, and the integrated model of the pump is formed by combining two parts of the program;

(4) Reversely fit the test data of the water pump, realize the optimal design function of the motor used according to the power demand of the water pump and the safety margin coefficient, and form a reverse optimization system of the water pump motor, as shown in Figure 5;

(5) Using the simulated annealing genetic algorithm, according to the demand data obtained from the reverse fitting of the pump performance, the set margin coefficient target and the motor performance target, and with the motor forward electromagnetic calculation program, the computer automatic Optimized design.

In step (1), the motor forward electromagnetic calculation program:

For the submersible motor used by the water pump, the mechanism modeling method is used to program the calculation process to form a motor electromagnetic calculation program that can operate independently;

The input data of the program includes: output power kW, power frequency Hz, line voltage V, stator connection, number of poles, pitch, number of stator and rotor slots, stator outer and inner diameter mm, air gap mm, rotor inner diameter mm, stator slot type Size mm, rotor slot size mm), number of turns per circle, number of parallel branches, number of parallel windings, slot insulation thickness mm, wire diameter, double-sided paint film mm, iron core length mm, lamination coefficient, and straight line part of the coil Protrusion length mm, average diameter of end ring mm, area mm^2, stray loss kW, iron loss coefficient, leakage reactance coefficient, etc.

The calculated output data include: efficiency, power factor, rated torque, maximum torque, starting torque, starting current, slot full rate;

On the basis of the electromagnetic calculation of the equivalent magnetic circuit method of the traditional three-phase asynchronous motor, according to the operating conditions of the submersible motor submerged in water and the structural characteristics of both the stator and the rotor using closed slots, the mechanical loss of the motor and the leakage permeance of the upper part of the stator and rotor slots In terms of the calculation method, it adopts a calculation method different from that of ordinary asynchronous motors;

The mechanical loss of the water pump motor is divided into three parts for calculation: the friction loss between the rotor and the cooling water, the friction loss of the thrust bearing and the friction loss of the guide bearing;

For the leakage permeance of the upper part of the closed slot of the stator and rotor of the pump motor, the method of equivalent slot width is used for calculation.

In step (2), the BP neural network:

Two BP neural network 1 and BP neural network 2 with similar structures and different functions are established, which are respectively used to combine the motor forward electromagnetic calculation program to form an integrated model of the machine and pump, and combine the motor electromagnetic calculation and optimization algorithm to form the reverse direction of the pump motor. Optimal design system;

The pump hydraulic BP neural network model 1, which is used to combine the positive electromagnetic calculation program of the motor to form the integrated model of the machine and pump, adopts double input and double output, three-layer structure, 8 neurons in the hidden layer, gradient descent method to correct errors, and training termination conditions For the structural characteristics of the training error less than 2%, as shown in Figure 2, the input shaft power and pump efficiency of the water pump are used as the model input, and the head and flow data of the water pump are output;

It is used to combine the motor forward electromagnetic calculation program and simulated annealing genetic algorithm to form the water pump hydraulic BP neural network model 2 of the reverse optimization system of the pump motor. It adopts three-input and single-output, three-layer structure, hidden layer 8 neurons, and gradient descent method. The error is corrected, and the training termination condition is the structural feature that the training error is less than 2%. As shown in Figure 3, the pump design flow, head, and target pump efficiency are used as inputs to calculate the output and predict the required shaft power of the pump.

In step (3), the machine-pump integration model:

After inputting the motor structure parameters to carry out the motor forward electromagnetic calculation, the calculation result is transmitted to the BP neural network 1 which has been trained and converged by the measured samples, and finally outputs the output flow and head of the water pump under the current motor power state, as shown in Figure 4 Show;

The input structure parameters used by the motor forward electromagnetic program, and the data used by the pump hydraulic BP neural network model 1 for neuron training, are stored in the folder in the form of txt files, and the program automatically reads them;

The machine-pump integration model can be calculated independently of the reverse optimization system of the pump motor in step (4).

In step (4), the pump motor reverse optimization system shown in Figure 5:

Combining the motor forward electromagnetic calculation program, BP neural network 2 and simulated annealing genetic algorithm, the reverse optimization system of the water pump motor is formed;

After the BP neural network 2 is trained and converged according to the measured samples, when a set of flow rate, head and target pump efficiency data are input, it will fit and give the shaft power that the pump needs to input in this state, with the artificially given margin coefficient, the optimal output power of the motor that the pump should be equipped with in this state can be obtained; use this power as the power optimization design target of the motor, and cooperate with other indicators of efficiency and torque to perform multi-objective optimization of the pump motor Design; the optimization target is set as the motor efficiency, power factor, maximum torque, starting torque, starting current and the backup coefficient between the pump; the motor optimization variable selection: iron core length, number of conductors per slot, stator and rotor slots Some internal dimension parameters and rated output power;

In the multi-objective optimization design of water pump motors, it is generally difficult to find the optimal solution that satisfies all the objective requirements, so the non-inferior solutions that meet part of the optimization objectives obtained at the end of the optimization are output in txt format as the optimization results for manual selection;

The reverse optimization system of the water pump motor can be run independently from the integrated model of the machine and pump in step (3) for calculation.

The pump motor modeling and optimization system based on electromagnetic calculation and neural network can establish a machine-pump integration model through the forward electromagnetic calculation of the pump motor and the neural network fitting of the hydraulic performance of the pump. At the same time, combined with the optimization algorithm, according to the principle of high power matching between the pump and the pump, the intelligent reverse optimization of the motor for the pump is carried out.

Among them, the specific machine-pump integration modeling process is as follows:

Step 1: Input the original structural parameters of the three-phase asynchronous motor used in the pump system of the target machine, and calculate the output parameters such as motor output torque, power, and state parameters such as current and efficiency through the motor forward electromagnetic calculation program.

Step 2: According to the hydraulic performance test data of the target machine pump system, train the BP neural network model 1 of the water pump until the neural network has the same input and output characteristics as the measured data of the water pump.

Step 3: Transmit the result of the motor forward electromagnetic calculation to the pump BP neural network model 1, and calculate and output the final output flow and head performance of the pump system under the current motor power state. Simultaneously output the state parameters of the motor and some state parameters of the water pump.

The process of intelligent reverse optimization of the motors used for the above water pumps is as follows:

Step 1: According to the test data of the hydraulic performance test of the object machine pump system, reverse train the BP neural network 2 until it converges.

Step 2: Input the pump design flow rate, head and target pump efficiency into BP neural network 2, and calculate the shaft power that needs to be input from the motor under the condition of the flow rate and head requirements of the water pump.

Step 3: Set the performance target values of the optimal design of the motor and the reduction target value of the machine-pump backup coefficient.

Step 4: Divide the structural parameters of the motor into two groups of parameters that do not participate in the optimization and those that participate in the optimization. Run the optimization algorithm to adjust the parameters participating in the optimization, and cooperate with the parameters not participating in the optimization to form multiple new motor design schemes.

Step 5: Input all new motor design schemes to the motor forward electromagnetic calculation program, and calculate the output performance of each scheme.

Step 6: Verify that all new motor designs satisfy all optimization objectives. Yes, output this scheme parameter. No, regenerate a set of motor designs following the simulated annealing genetic algorithm rules. Repeat step 5 until the optimization termination condition is satisfied.

Claims (5)

1. calculate and the pump motor modeling and optimization method of neural network based on electromagnetism, it is characterized in that: comprise the following steps:

(1), according to the mechanism model of water pump submersible three phase asynchronous motor, motor forward electromagnetic computation program is write;

(2), adopt BP neural net method, data fitting modeling is carried out to water pump hydraulic model;

(3), using the input data of the Output rusults of motor electromagnetic calculating as water pump hydraulic model, in conjunction with the pump integrated model of two parts program forming machine;

(4), to parallel water pump data carry out reverse matching, according to pump power demand, safety allowance coefficient, realize being optimized design function to institute's adapted motor, form pump motor reverse optimization system;

(5), adopt Genetic Simulated Annealing Algorithm, the demand data drawn according to the reverse matching of pump performance, the allowance coefficient target of setting and motor performance target, coordinate motor forward electromagnetic computation program, computing machine Automatic Optimal Design is carried out to adapted motor.

2. according to claim 1ly to calculate and the pump motor modeling and optimization method of neural network based on electromagnetism, it is characterized in that: in described step (1), motor forward electromagnetic computation program:

To the submersible motor of water pump adapted, adopt modelling by mechanism method, computation process sequencing being formed can the motor electromagnetic calculation procedure of independent operating;

The input data of program comprise: output power kW, supply frequency Hz, line voltage V, stator connection, number of poles, pitch, rotor groove number, stator outer interior diameter mm, air gap mm, rotor internal diameter mm, stator slot molded dimension mm, slot shape of rotor size mm), often enclose the number of turn, parallel branch number around radical, slot insulation thickness mm, wire diameter, bilateral paint film mm, the long mm of iron core, stacking factor, coil straight line portion stretch out long mm, end ring mean diameter mm, area mm^2, stray loss kW, iron loss factor, leakage reactance coefficient etc.;

Export data as calculated to comprise: efficiency, power factor, nominal torque, torque capacity, detent torque, starting current, copper factor;

On the basis that traditional threephase asynchronous magnetic equivalent circuit method electromagnetism calculates, the design feature that service condition in water and rotor all adopt closed slot is immersed according to submersible motor, on the mechanical loss of motor and the computing method of rotor groove top leakage permeance, have employed the computing method different from common asynchronous moter;

Pump motor mechanical loss is divided into three parts to calculate: the frictional dissipation of the frictional dissipation of rotor and chilled water, the frictional dissipation of thrust bearing and guide bearing;

For the groove top leakage permeance of pump motor rotor closed slot, the method for equivalent slot opening is adopted to calculate.

3. according to claim 1ly to calculate and the pump motor modeling and optimization method of neural network based on electromagnetism, it is characterized in that: in described step (2), BP neural network:

Set up two kinds of structure proximates, the BP neural network 1 that function is different and BP neural network 2, be respectively used in conjunction with motor forward electromagnetic computation program, the pump integrated model of forming machine, and in conjunction with motor electromagnetic calculation and optimization algorithm, form pump motor reverse optimization design system;

For in conjunction with motor forward electromagnetic computation program, the water pump waterpower BP neural network model 1 of the pump integrated model of forming machine, adopt two-output impulse generator, three-decker, hidden layer 8 neuron, gradient descent method round-off error, training end condition is that training error is less than 2% " architectural feature; with water pump input shaft power, pump efficiency for mode input, export water lift of pump and data on flows;

For in conjunction with motor forward electromagnetic computation program and Genetic Simulated Annealing Algorithm, form the water pump waterpower BP neural network model 2 of pump motor reverse optimization system, adopt triple input single output, three-decker, hidden layer 8 neuron, adopts gradient descent method round-off error, and training end condition is that training error is less than 2% " architectural feature; with pumps design flow, lift, target pump efficiency for input, calculate prediction of output water pump demand shaft power.

4. according to claim 1ly to calculate and the pump motor modeling and optimization method of neural network based on electromagnetism, it is characterized in that: in described step (3), motor-and-pump integrated model:

After input electric machine structure parameter carries out the calculating of motor forward electromagnetism, the result of calculating is transferred to through actual measurement sample training and the BP neural network 1 of having restrained, finally exports delivery rate, the lift of water pump under current motor power rating;

The input structure parameter that motor forward electromagnetic program uses, and water pump waterpower BP neural network model 1 is used for carrying out the data of neuron training, all leave in file with the form of txt document, program reads automatically;

Motor-and-pump integrated model can independent of pump motor reverse optimization system described in step (4), and isolated operation calculates.

5. according to claim 1ly to calculate and the pump motor modeling and optimization method of neural network based on electromagnetism, it is characterized in that: in described step (4), pump motor reverse optimization system:

In conjunction with motor forward electromagnetic computation program, BP neural network 2 and Genetic Simulated Annealing Algorithm, form pump motor reverse optimization system;

BP neural network 2 carries out training according to actual measurement sample and after restraining, when input one group of flow, lift and target pump efficiency data, matching is provided the shaft power that water pump needs to input in this case, coordinate artificially given allowance coefficient, water pump can be drawn in this case, the optimum output power that the motor of its adapted should have; Power-optimized designs target using this power as motor, coordinates efficiency, other indexs of torque, carries out pump motor multi-objective optimization design of power; Optimization aim is set as the reserve factor between the efficiency of motor, power factor, torque capacity, detent torque, starting current and pump; Motor optimized variable is chosen: iron core long, every groove conductor number, rotor slot part inside dimension parameter and output rating;

Pump motor multi-objective optimization design of power is generally difficult to find the optimum solution meeting all target calls, then exported as optimum results, for artificial selection using txt form by the noninferior solution meeting part optimization aim obtained when optimizing and stop;

Pump motor reverse optimization system can independent of the described motor-and-pump integrated model of step (3), and isolated operation calculates.