Rotor temperature estimation method based on BP network nerve

Abstract

The invention discloses a BP network nerve-based rotor temperature estimation method. The method comprises the following steps of S1, calculating a required signal according to a rotor temperature; s2, calculating a BP neural network input layer signal; s3, calculating a BP neural network hidden layer signal according to the rotor temperature; and S4, calculating a BP neural network output layer signal according to the rotor temperature. According to the method, on the premise that hardware cost does not need to be additionally increased, the motor rotor temperature estimation requirements under different environment temperatures and variable load working conditions can be met.

Description

Technical Field

[0001] The present invention relates to the technical field of new energy motors, and in particular to a rotor temperature estimation method based on BP network neural network. Background Art

[0002] In recent years, new energy vehicles have significant advantages over traditional fuel vehicles in comprehensive energy consumption and exhaust emissions, reducing energy consumption and environmental pollution. Governments and OEMs around the world have increased their efforts in the development of new energy vehicles, and new energy vehicles have gradually gained favor among consumers.

[0003] Permanent magnet synchronous motors are the preferred main drive motors for new energy vehicles due to their advantages such as high efficiency, high power density, low speed and high torque. New energy vehicles work continuously under high temperature and high load conditions. The rotor magnets of permanent magnet synchronous motors are prone to overheating and irreversible demagnetization, causing power loss of the entire vehicle. In order to prevent the risk of power interruption of the entire vehicle caused by excessive motor rotor temperature, it is necessary to accurately monitor the motor rotor temperature in real time.

[0004] Due to the narrow structural space of the motor, the high cost of wireless temperature acquisition equipment, and the blockade of permanent magnet synchronous motor rotor temperature algorithm technology abroad, domestic OEMs usually conservatively limit the motor output power when the motor stator temperature reaches a certain level to indirectly ensure that the rotor does not overheat. This causes the motor performance to fail to fully exert its ultimate. Accurate estimation of the motor rotor temperature directly increases the motor stator temperature operating threshold. The increase in the stator temperature threshold can increase the motor current output capacity, and the current output capacity is strongly related to the volume size of the magnetic material. Under the premise that the rotor temperature is accurately estimated and the motor performance remains unchanged, the manufacturing cost can be reduced by reducing the volume size of the motor magnetic material. In addition, changes in the motor rotor temperature will cause changes in the magnetic flux of the motor permanent magnet. Changes in the motor magnetic flux will affect the actual torque output accuracy of the motor. Therefore, the motor rotor temperature estimation can compensate for the motor torque according to the magnetic flux to achieve precise control of the motor system output torque. Summary of the invention

[0005] The purpose of the present invention is to provide a rotor temperature estimation method based on BP network neural network to address the corresponding deficiencies of the prior art. The present invention can meet the requirements of motor rotor temperature estimation under different ambient temperatures and variable load conditions without increasing additional hardware costs.

[0006] The technical solutions to the above problems are as follows:

[0007] A rotor temperature estimation method based on BP network neural network includes the following steps:

[0008] S1, calculates the required signal based on the rotor temperature;

[0009] S2, calculate the BP neural network input layer signal;

[0010] S3, calculating the hidden layer signal of the BP neural network according to the rotor temperature;

[0011] S4, calculate the BP neural network output layer signal according to the rotor temperature.

[0012] In the present invention, the development of rotor temperature estimation based on neural network first establishes a rotor temperature estimation neural network model, develops a neural network training algorithm, and develops a neural network hyperparameter (number of network layers, number of neurons in each layer, etc.) optimization algorithm, then uses an experimental data set to train the neural network, optimize hyperparameters, and uses a test data set to verify accuracy. After the accuracy meets the requirements, the neural network model code is generated and hardware is deployed, and then bench testing and real vehicle verification are performed. There are many structures and types of neural networks. Here, a BP neural network is used to establish a rotor temperature estimation model, which has less calculation and higher estimation accuracy, so that it can be used preferentially in actual deployment.

[0013] The present invention can accurately estimate the motor rotor temperature, ensure the motor rotation safety, improve the motor current output capacity and torque output accuracy or reduce the motor manufacturing magnetic material cost. BRIEF DESCRIPTION OF THE DRAWINGS

[0014] Figure 1 It is a flow chart of the present invention.

[0015] Figure 2 Schematic diagram of the first hidden layer signal calculation.

[0016] Figure 3 Schematic diagram of the signal calculation for the second hidden layer.

[0017] Figure 4 Schematic diagram of the third hidden layer signal calculation.

[0018] Figure 5 Schematic diagram of BP neural network output layer signal calculation. DETAILED DESCRIPTION

[0019] The present invention is further described in detail below in conjunction with the accompanying drawings and specific embodiments. The examples described are only part of the embodiments of the present invention, not all of the embodiments. Based on the examples in the present invention, all other embodiments obtained by ordinary technicians in the field without creative work are within the scope of protection of the present invention.

[0020] like Figures 1 to 5 As shown, a rotor temperature estimation method based on BP network neural network of the present invention comprises the following steps:

[0021] S1, calculates the required signal according to the rotor temperature; the process of calculating the required signal according to the speed temperature in S1 is as follows:

[0022] S11, collect and obtain stator NTC sensor temperature, ambient temperature, cooling oil pump speed, motor speed, motor torque, bus voltage and current signals.

[0023] S12, calculate the electric energy input to the system = bus voltage * bus current / 1000.

[0024] S13, calculate system mechanical energy = 1000*motor torque*motor speed / 9550.

[0025] S14, calculate the difference between DC energy and mechanical energy = (bus voltage * bus current / 1000) -

[0026] (1000*motor torque*motor speed / 9550).

[0027] S2, calculate the BP neural network input layer signal. The process of calculating the BP neural network input layer signal in S2 is as follows:

[0028] S21, respectively normalizes the stator NTC sensor temperature, cooling oil pump speed, motor speed, motor torque, the difference between the electrical energy and mechanical energy of the input system, and the ambient temperature. The normalization in S21 is performed according to the following formula:

[0029]

[0030] In the above formula: is the value before normalization, x is the value after normalization, and x max 、x mi n are the maximum and minimum values ​​of the test data respectively.

[0031] S22, the normalized stator NTC sensor temperature, ambient temperature, cooling oil pump speed, motor speed, motor torque, and the difference between the electrical energy and mechanical energy of the input system are processed with different delays to obtain the signal values ​​at the first four moments, and the current value is not delayed.

[0032] In S22, the delay is performed according to the following process:

[0033] At time t, the stator NTC sensor temperature t: no delay;

[0034] At time t-1, the stator NTC sensor temperature: delay 1 time;

[0035] At time t-2, the stator NTC sensor temperature is t-2: delayed 2 times;

[0036] At time t-3, the stator NTC sensor temperature is t-3: delayed 3 times;

[0037] At time t-4, stator NTC sensor temperature t-4: delay 4 times.

[0038] The processing methods of the current and previous four moment values ​​of the cooling oil pump speed, motor speed, motor torque, and the difference between the electrical energy and mechanical energy input to the system are the same as above. According to the above, 26 signal values ​​are obtained, as follows:

[0039] Stator NTC sensor temperature t-4, stator NTC sensor temperature t-3, stator NTC sensor temperature t-2, stator NTC sensor temperature t-1, stator NTC sensor temperature t.

[0040] Ambient temperature.

[0041] Cooling oil pump speed t-4, cooling oil pump speed t-3, cooling oil pump speed t-2, cooling oil pump speed t-1, cooling oil pump speed t.

[0042] Motor speed t-4, motor speed t-3, motor speed t-2, motor speed t-1, motor speed t.

[0043] Motor torque t-4, motor torque t-3, motor torque t-2, motor torque t-1, motor torque t.

[0044] The electrical energy - mechanical energy t-4 of the input system, the electrical energy - mechanical energy t-3 of the input system, the electrical energy - mechanical energy t-2 of the input system, the electrical energy - mechanical energy t-1 of the input system, the electrical energy - mechanical energy t of the input system.

[0045] S3, calculate the hidden layer signal of the BP neural network according to the rotor temperature. The hidden layer signal of the BP neural network in S3 includes the first hidden layer signal, the second hidden layer signal, and the third hidden layer signal. The calculation of each hidden layer signal is as follows:

[0046] The calculation formula of the first hidden layer signal is as follows:

[0047]

[0048] u1=input 26 signal values×w1+b1

[0049] F(x1) refers to the first hidden layer signal, e is a constant, add1 is the coefficient in the activation function of the first hidden layer, w1 and b1 are the weight and bias of the first hidden layer, and w1 and b1 are obtained through neural network inverse training calculation.

[0050] The calculation formula of the second hidden layer signal is as follows:

[0051]

[0052] u2=F(x1)×w2+b2

[0053] F(x2) is the signal of the second hidden layer, e is a constant, add2 is the coefficient in the activation function of the second hidden layer, w2 and b2 are the weight and bias of the second hidden layer, and w2 and b2 are obtained through neural network inverse training calculation.

[0054] The calculation formula of the third hidden layer signal is as follows:

[0055]

[0056] u3=F(x2)×w3+b3

[0057] F(x3) is the third hidden layer signal, e is a constant, add3 is the coefficient in the activation function of the second hidden layer, w3 and b3 are the weight and bias of the third hidden layer, and w3 and b3 are obtained through neural network inverse training calculation.

[0058] S4, calculate the BP neural network output layer signal according to the rotor temperature. The calculation of the BP neural network output layer signal is: the third hidden layer signal is calculated by updating the output layer weight and deviation to obtain the real-time temperature of the motor rotor, and the calculation formula is:

[0059] F(y)=F(x3)×w1+b1

[0060] F(y) is the output temperature of the rotor.

[0061] The present invention can meet the requirements for estimating the motor rotor temperature under different ambient temperatures and variable load conditions without increasing the hardware cost, and has the characteristics of high real-time performance, high precision and strong adaptability.

Claims

1. A rotor temperature estimation method based on BP neural network, characterized in that: The following steps are involved: S1, calculates the required signal based on the rotor temperature; S2, calculate the BP neural network input layer signal; S3, calculating the hidden layer signal of the BP neural network according to the rotor temperature; S4, calculate the BP neural network output layer signal according to the rotor temperature.

2. The rotor temperature estimation method based on BP network neural network according to claim 1 is characterized in that: The process of calculating the required signal according to the speed and temperature in S1 is as follows: S11, collecting and obtaining stator NTC sensor temperature, ambient temperature, cooling oil pump speed, motor speed, motor torque, bus voltage and current signals; S12, calculate the electric energy input into the system = bus voltage * bus current / 1000; S13, calculate system mechanical energy = 1000*motor torque*motor speed / 9550; S14, calculate the difference between DC power energy and mechanical energy = (bus voltage*bus current / 1000)-(1000*motor torque*motor speed / 9550).

3. The rotor temperature estimation method based on BP neural network according to claim 2 is characterized in that: The process of calculating the BP neural network input layer signal in S2 is as follows: S21, normalizing the stator NTC sensor temperature, the cooling oil pump speed, the motor speed, the motor torque, the difference between the electrical energy and the mechanical energy input to the system, and the ambient temperature; S22, the normalized stator NTC sensor temperature, ambient temperature, cooling oil pump speed, motor speed, motor torque, and the difference between the electrical energy and mechanical energy of the input system are processed with different delays to obtain the signal values ​​at the first four moments, and the current value is not delayed.

4. The rotor temperature estimation method based on BP neural network according to claim 3 is characterized in that: The normalization process in S21 is performed according to the following formula: In the above formula: is the value before normalization, x is the value after normalization, and x max 、x min are the maximum and minimum values ​​of the test data respectively.

5. The rotor temperature estimation method based on BP neural network according to claim 3 is characterized in that: In S22, the delay is performed according to the following process: At time t, the stator NTC sensor temperature t: no delay; At time t-1, the stator NTC sensor temperature: delay 1 time; At time t-2, the stator NTC sensor temperature is t-2: delayed 2 times; At time t-3, the stator NTC sensor temperature is t-3: delayed 3 times; At time t-4, the stator NTC sensor temperature is t-4: delayed 4 times; The processing method of the current moment and the previous four moment values ​​of the cooling oil pump speed, motor speed, motor torque, and the difference between the electrical energy input to the system and the mechanical energy is the same as above.

6. The rotor temperature estimation method based on BP neural network according to claim 5 is characterized in that: The hidden layer signal of the BP neural network in S3 includes a first hidden layer signal, a second hidden layer signal, and a third hidden layer signal, wherein the calculation formula of the first hidden layer signal is as follows: u1=input 26 signal values×w1+b1 F(x1) refers to the first hidden layer signal, e is a constant, add1 is the coefficient in the activation function of the first hidden layer, w1 and b1 are the weight and bias of the first hidden layer; The calculation formula of the second hidden layer signal is as follows: u2=F(x1)×w2+b2 F(x2) is the second hidden layer signal, e is a constant, add2 is the coefficient in the activation function of the second hidden layer, w2 and b2 are the weight and bias of the second hidden layer; The calculation formula of the third hidden layer signal is as follows: u3=F(x2)×w3+b3 F(x3) is the third hidden layer signal, e is a constant, add3 is the coefficient in the activation function of the second hidden layer, and w3 and b3 are the weight and bias of the third hidden layer.

7. The rotor temperature estimation method based on BP network neural network according to claim 5 is characterized in that: The calculation of the output layer signal of the BP neural network is as follows: the third hidden layer signal is calculated by updating the output layer weights and deviations to obtain the real-time temperature of the motor rotor. The calculation formula is: F(y)=F(x3)×w1+b1 F(y) is the output temperature of the rotor.

Images