阅读说明：本技术 一种提高电动汽车用永磁同步电机效率的方法 (Method for improving efficiency of permanent magnet synchronous motor for electric automobile ) 是由 谢芳 安超晨 于飞 于 2021-09-03 设计创作，主要内容包括：本发明公开一种提高电动汽车用永磁同步电机效率的方法,通过对电流的协调控制,提高电动汽车用内置式永磁同步电机的运行效率,首先,建立电机效率的数学模型,定性分析电机在不同运行区域下效率最优时的直轴电流和交轴电流分配规律,设计了基于电流分配模型的矢量控制系统；其次,搭建AVL测功机台架系统实验平台,通过实测数据建立样本空间；其次,引入深度置信网络的电流回归模型,采用“预训练—微调参数”的两阶段学习训练网络参数；最后,将此回归模型嵌入到控制系统中,实现电流的协调控制。该方法配比电流精确,可以提高永磁同步电机在整个调速范围的运行效率。(The invention discloses a method for improving the efficiency of a permanent magnet synchronous motor for an electric vehicle, which improves the operating efficiency of a built-in permanent magnet synchronous motor for the electric vehicle by coordinating and controlling current, firstly, a mathematical model of the motor efficiency is established, the distribution rule of direct axis current and quadrature axis current when the efficiency of the motor is optimal in different operating areas is qualitatively analyzed, and a vector control system based on a current distribution model is designed; secondly, an AVL dynamometer rack system experiment platform is set up, and a sample space is established through actually measured data; secondly, introducing a current regression model of a deep confidence network, and learning and training network parameters by adopting two stages of pre-training-fine tuning parameters; and finally, embedding the regression model into a control system to realize the coordination control of the current. The method has accurate proportioning current and can improve the operation efficiency of the permanent magnet synchronous motor in the whole speed regulation range.)

1. A method for improving the efficiency of a permanent magnet synchronous motor for an electric automobile is characterized by comprising the following steps: the method comprises the following steps:

the method comprises the following steps: establishing motor efficiency and direct axis current i according to a stator voltage equation, an electromagnetic torque equation, motor input power, copper power consumption, iron power consumption and motor output power under the condition of steady-state operation of the permanent magnet synchronous motor_sdAnd quadrature axis current i_sqThe mathematical relationship of (1);

step two: solving the direct-axis current i when the efficiency of the permanent magnet synchronous motor is optimal in the constant torque and constant power operation area according to the actual operation condition of the permanent magnet synchronous motor by utilizing the mathematical relation established in the step one_sdAnd quadrature axis current i_sqAccording toSolving direct axis current i under different operating conditions_sdQuadrature axis current i_sqDeducing a current schematic diagram of the established analytical model;

step three: establishing a vector control system according to the optimal current in the step two, wherein the vector control system comprises a current distribution module, a PI (proportion integration) regulation module, a coordinate transformation module and an SVPWM (space vector pulse width modulation) module, the given rotating speed and torque of the permanent magnet synchronous motor are used as input values, and a current distribution model carries out current distribution according to the input values to obtain reference direct axis currentQuadrature axis currentOutput reference current direct axis currentQuadrature axis currentThe control signal of the inverter is generated by PI regulation and coordinate transformation and then is used as module input generated by SVPWM, so that the corresponding control effect is achieved;

step four: selecting a current distribution analytical model as a control i_sdAnd i_sqThe method comprises the following steps that a control model of the ratio is used for measuring current discrete data and various system parameters influencing current distribution when a permanent magnet synchronous motor runs under all working conditions on an AVL experiment platform and using the current discrete data and the system parameters as a sample library for building a deep belief network current regression model;

step five: preprocessing actual measurement data of the permanent magnet synchronous motor under the operating conditions of a constant torque area and a constant power area, wherein the preprocessing result is used as a characteristic set in a sample library of current distribution of a deep confidence network regression algorithm; inputting the feature set, searching a mapping relation between an input feature x and an output feature y by using a Restricted Boltzmann Machine (RBM), learning network parameters in a training deep belief network regression model by adopting two stages of pre-training and fine-tuning parameters, and obtaining well-trained network parametersThe deep confidence network current regression model finally obtains the output result, namely the direct axis currentCurrent of sum and quadrature axis

Step six: and replacing the trained deep belief network current regression model with the current distribution model in the vector control system in the third step, finally determining the control model for improving the efficiency of the permanent magnet synchronous motor, and verifying the effectiveness of the model through an AVL (amplitude versus voltage) experimental platform.

2. The method for improving the efficiency of the permanent magnet synchronous motor for the electric automobile according to claim 1, characterized in that: in the first step: according to a stator voltage equation, an electromagnetic torque equation, motor input power, copper power consumption, iron power consumption and motor output power under the condition of steady-state operation of the permanent magnet synchronous motor, establishing a mathematical relation among motor efficiency, torque current and exciting current as follows:

3. the method for improving the efficiency of the permanent magnet synchronous motor for the electric automobile according to claim 1, characterized in that: in the third step: the current distribution vector control system carries out formula calculation according to different operation conditions of the motor to obtain reference currents under different conditionsReference current of outputThrough PI regulation module and coordinate transformation module, and then as SVPWM generationTo generate inverter control signals.

4. The method for improving the efficiency of the permanent magnet synchronous motor for the electric automobile according to claim 1, characterized in that: the concrete implementation process of the step five is as follows:

(1) carrying out normalization preprocessing on the measured data of the permanent magnet synchronous motor under the operating conditions of a constant torque area and a constant power area by adopting a z-score normalization method;

wherein the content of the first and second substances,σ_xthe mean value and the standard deviation of a certain characteristic x in the measured data are obtained; x' is a value subjected to normalization processing;

according to the second step, the efficiency of the permanent magnet synchronous motor in the constant torque and constant power operation area is optimized_sdAnd i_sqIs selected to the electromagnetic torque T in the constant torque region_eN, permanent magnet flux linkage psi_fD-axis inductance L_dQ-axis inductor L_qStator phase current I_sAnd electromagnetic torque T in constant power region_eN, permanent magnet flux linkage psi_fD-axis inductance L_dQ-axis inductor L_qCopper power consumption P_cuIron loss power P_feInput power P_inOutput power P_outMaximum stator voltage U_maxMaximum stator current I_maxD-axis voltage u_sdQ-axis voltage u_sqAs an input feature set;

(2) pre-training: taking the feature set as input, training a restricted Boltzmann machine RBM from bottom to top, and at each layer, obtaining a parameter w_ijConstructing according to the data obtained by the calculation of the previous layer, calculating the state of the unit according to the formulas (1) and (2), and updating the weight by adopting a single-step comparison bifurcation algorithm;

wherein, w_ijIs the connection weight between neurons i, j; a is_i、x_iBias and state of the ith neuron in the visible layer, respectively; b_j、x_jBias and state of the jth neuron in the hidden layer respectively; σ (x) is a sigmiod activation function in a neural network;

(3) fine adjustment of parameters: after the training is finished, the weight of the whole network is finely adjusted from top to bottom by using a wake-sleep algorithm until the set training times or the error meets the requirement, and finally the output result, namely the direct-axis current is obtainedCurrent of sum and quadrature axis

Technical Field

The invention relates to the technical field of permanent magnet synchronous motors, in particular to a method for improving the efficiency of a permanent magnet synchronous motor for an electric automobile.

Background

Because the driving conditions of the electric automobile are complex, the electric automobile is often started, stopped and accelerated and decelerated, and the driving comfort of a driver and the riding comfort of passengers are considered, so that the electric automobile has high requirements on a driving system of the electric automobile. The embedded permanent magnet synchronous motor is widely applied to the field of electric automobiles due to the advantages of high efficiency, high power density and the like. However, the complicated operating conditions of the electric vehicle may greatly reduce the motor efficiency. Therefore, the research on the efficiency optimization of the driving motor of the electric automobile has important significance for saving energy and reducing environmental pollution. The existing method for improving the efficiency mainly comprises the design and control strategy of a motor body. The design of the motor body is to realize energy conservation by using new materials and improved technologies. However, optimization of the control strategy is a cost effective method when determining the motor body.

The existing motor efficiency optimization control strategy methods are many and mainly divided into two categories, namely an efficiency optimization method of a motor loss model and an input power minimum strategy. The control efficiency of the loss model control method is globally optimal, the control speed is high, but the method is too dependent on an accurate motor mathematical model and cannot adapt to variable electric automobile operation conditions. The input power minimum strategy is to minimize the input power of the power supply under the same working condition by adjusting the dq axis current, and the algorithm has larger calculated amount and slower optimization speed and can generate phenomena such as oscillation and the like.

The prior art related to the present invention is: an efficiency optimization algorithm based on the motor loss model; based on a hybrid fuzzy search efficiency optimization method; predicting a current control strategy based on minimum loss of iron loss online calculation; and optimizing and controlling the efficiency based on an online search method. The prior art has the following disadvantages: the method is too dependent on accurate motor parameters, has a plurality of influence factors, and cannot meet the requirement of the electric automobile on complex operation conditions; the algorithm has the advantages of large calculation amount, low optimization speed, complex control system, difficult parameter setting and easy generation of torque jitter.

Disclosure of Invention

The technical problem to be solved by the invention is as follows: the efficiency optimization of the permanent magnet synchronous motor for the electric automobile under different working conditions is realized by the coordinated control of current, the continuous accurate prediction of multi-input and multi-output can be realized under any working condition point by utilizing discontinuous discrete measured data, the structure of a system is simplified, the operand is reduced, the accurate current input quantity is provided for a PI regulator, the method is accurate in current proportioning, and the operating efficiency of the permanent magnet synchronous motor in the whole speed regulation range can be improved.

The technical scheme adopted by the invention is as follows: the efficiency method of the permanent magnet synchronous motor for the electric automobile based on the deep belief network regression algorithm comprises the following steps of collecting measured data as a sample library, constructing a deep belief network regression model, embedding the deep belief network regression model into a vector control system, and realizing efficiency optimization of the motor under different working conditions through coordination control of current:

the method comprises the following steps: establishing motor efficiency and direct axis current i according to a stator voltage equation, an electromagnetic torque equation, motor input power, copper power consumption, iron power consumption and motor output power under the condition of steady-state operation of the permanent magnet synchronous motor_sdAnd quadrature axis current i_sqThe mathematical relationship of (1);

step two: solving the direct-axis current i when the efficiency of the permanent magnet synchronous motor is optimal in the constant torque and constant power operation area according to the actual operation condition of the permanent magnet synchronous motor by utilizing the mathematical relation established in the step one_sdAnd quadrature axis current i_sqAccording to the direct axis current i under different operation conditions_sdQuadrature axis current i_sqDeducing a current schematic diagram of the established analytical model;

step three: establishing a vector control system according to the optimal current in the step two, wherein the vector control system comprises a current distribution module, a PI (proportion integration) regulation module, a coordinate transformation module and an SVPWM (space vector pulse width modulation) module, the given rotating speed and torque of the permanent magnet synchronous motor are used as input values, and a current distribution model carries out current distribution according to the input valuesObtaining a reference direct axis currentQuadrature axis currentOutput reference current direct axis currentQuadrature axis currentThe control signal of the inverter is generated by PI regulation and coordinate transformation and then is used as module input generated by SVPWM, so that the corresponding control effect is achieved;

step four: selecting a current distribution analytical model as a control i_sdAnd i_sqThe method comprises the following steps that a control model of the ratio is used for measuring current discrete data and various system parameters influencing current distribution when a permanent magnet synchronous motor runs under all working conditions on an AVL experiment platform and using the current discrete data and the system parameters as a sample library for building a deep belief network current regression model;

step five: preprocessing actual measurement data of the permanent magnet synchronous motor under the operating conditions of a constant torque area and a constant power area, wherein the preprocessing result is used as a characteristic set in a sample library of current distribution of a deep confidence network regression algorithm; inputting the feature set, searching a mapping relation between an input feature x and an output feature y by using a Restricted Boltzmann Machine (RBM), learning network parameters in a training deep confidence network regression model by adopting two stages of pre-training and fine-tuning parameters to obtain the trained deep confidence network current regression model, and finally obtaining an output result, namely a direct current axisCurrent of sum and quadrature axis

Step six: and replacing the trained deep belief network current regression model with the current distribution model in the vector control system in the third step, finally determining the control model for improving the efficiency of the permanent magnet synchronous motor, and verifying the effectiveness of the model through an AVL (amplitude versus voltage) experimental platform.

In the first step: according to a stator voltage equation, an electromagnetic torque equation, motor input power, copper power consumption, iron power consumption and motor output power under the condition of steady-state operation of the permanent magnet synchronous motor, establishing a mathematical relation among motor efficiency, torque current and exciting current as follows:

in the third step: the current distribution vector control system carries out formula calculation according to different operation conditions of the motor to obtain reference currents under different conditionsReference current of outputAnd the control signal is used as the module input generated by SVPWM after passing through a PI regulation module and a coordinate transformation module to generate an inverter control signal.

The concrete implementation process of the step five is as follows:

(1) carrying out normalization preprocessing on the measured data of the permanent magnet synchronous motor under the operating conditions of a constant torque area and a constant power area by adopting a z-score normalization method;

wherein the content of the first and second substances,σ_xthe mean value and the standard deviation of a certain characteristic x in the measured data are obtained; x' is a value subjected to normalization processing;

according to the efficiency of the permanent magnet synchronous motor in the constant torque and constant power operation area obtained in the step twoTime of priority i_sdAnd i_sqIs selected to the electromagnetic torque T in the constant torque region_eN, permanent magnet flux linkage psi_fD-axis inductance L_dQ-axis inductor L_qStator phase current I_sAnd electromagnetic torque T in constant power region_eN, permanent magnet flux linkage psi_fD-axis inductance L_dQ-axis inductor L_qCopper power consumption P_cuIron loss power P_feInput power P_inOutput power P_outMaximum stator voltage U_maxMaximum stator current I_maxD-axis voltage u_sdQ-axis voltage u_sqAs an input feature set;

(2) pre-training: taking the feature set as input, training a restricted Boltzmann machine RBM from bottom to top, and at each layer, obtaining a parameter w_ijConstructing according to the data obtained by the calculation of the previous layer, calculating the state of the unit according to the formulas (1) and (2), and updating the weight by adopting a single-step comparison bifurcation algorithm;

wherein, w_ijIs the connection weight between neurons i, j; a is_i、x_iBias and state of the ith neuron in the visible layer, respectively; b_j、x_jBias and state of the jth neuron in the hidden layer respectively; σ (x) is a sigmiod activation function in a neural network;

(3) fine adjustment of parameters: after the training is finished, the weight of the whole network is finely adjusted from top to bottom by using a wake-sleep algorithm until the set training times or the error meets the requirement, and finally the output result, namely the direct-axis current is obtainedCurrent of sum and quadrature axis

Compared with the prior art, the invention has the advantages that:

(1) according to the invention, from the angle of deep learning, a vector control system of a current prediction model based on a deep learning belief network regression algorithm is modeled, so that the excitation current and the torque current of the permanent magnet synchronous motor are subjected to regression prediction, and the operating efficiency of the whole speed regulation range of the permanent magnet synchronous motor is improved through the coordinated control of the currents.

(2) The method utilizes discontinuous discrete measured data for modeling, realizes continuous and accurate prediction of multiple inputs and multiple outputs at any working condition point, simplifies the structure of the system, reduces the calculation amount, provides accurate current input amount for the PI regulator, is accurate in proportioning current, and effectively reduces the influence of parameter change in the actual operation of the motor.

Drawings

FIG. 1 is a schematic current diagram of an analytical model developed according to a formula;

FIG. 2 is a diagram of a vector control system based on a current sharing model;

FIG. 3 is a diagram of a deep confidence network current regression algorithm;

FIG. 4 is a comparison of the current measured curve in the constant torque zone and the current regression curve through the deep confidence network regression model;

FIG. 5 is a comparison graph of the current measured curve in the constant power region and the current regression curve through the depth confidence network regression model;

FIG. 6 is a comparison graph of motor efficiency of a vector control system of a constant torque zone conventional vector control and embedded depth confidence network current regression model;

FIG. 7 is a comparison graph of motor efficiency of a vector control system of a conventional vector control and embedded depth confidence network current regression model in a constant power region;

FIG. 8 is an efficiency map of a vector control system with embedded deep belief network current regression model over the entire speed regulation range of the motor;

FIG. 9 is a flow chart of an implementation of the method of the present invention.

Detailed Description

The invention is further described with reference to the following figures and detailed description.

As shown in fig. 9, the specific implementation steps of the present invention are as follows:

the method comprises the following steps: in the directional vector control of a rotor magnetic field of the permanent magnet synchronous motor, a stator voltage equation (1) and an electromagnetic torque equation (2) of the motor are in steady operation. The motor loss mainly comprises two parts of mechanical loss and electrical loss, and the mechanical loss is usually uncontrollable; the electrical losses include stator copper losses (3) and core losses (4). The efficiency of the motor can be represented by the ratio of the output power (5) of the motor to the input power of the motor, i.e. equation (6); as can be seen from the above equation, for a given operating condition (given torque T)_eAnd a rotational speed omega_r) The optimized control of the motor efficiency can be realized by coordinately controlling the torque current and the exciting current.

T_e＝1.5n_pi_q[i_d(L_d-L_q)+ψ_f] (2)

P_fe＝C_feω_r[(L_di_sd+ψ_f)²+(L_qi_sd)²] (4)

P_out＝ω_rT_e (5)

In the formula: u. of_sd、u_sqDq-axis components of the stator voltage, respectively; i.e. i_sd、i_sqRespectively stator currentA dq axis component; r is the resistance of the stator; psi_d、ψ_qIs the dq axis component of the stator flux linkage; omega_eIs the electrical angular velocity; l is_d、L_qDq-axis inductance components, respectively; psi_fIs a permanent magnet flux linkage; t is_eIs an electromagnetic torque, n_pIs the number of pole pairs; c_feIs the iron loss factor.

Step two: and establishing a current distribution model of the motor in different operation areas. When the rotating speed is lower than the basic rotating speed, the principle of the maximum current ratio is selected to control the stator current, and a stator current vector (7) is determined, so that the copper consumption of the motor is minimized, and the running efficiency of the motor is improved. Solving the optimal current distribution in a constant torque operation area according to a Lagrange multiplier method and taking an equation (2) as a constraint condition, wherein the calculation result is an equation (10); in the actual operation of the motor, factors such as magnetic saturation are considered, and the rated direct-axis current i is often used_sdratedAs(constant power region) the motor is limited by current and voltage in the constant power region, and the input efficiency P (i) of the motor is limited under the conditions of voltage limitation and current limitation_sd,i_sq) Obtaining a minimum value; defining Lagrangian as formula (12), and applying formula (12) to i_sdAnd i_sqAnd (3) solving the partial derivatives to be zero, substituting the partial derivative equation into the formula (2), and solving the approximate result as the formula (14).

From the above formula, i under different working conditions_sd、i_sqWhen ω is_r<ω_rbaseHandle i_sdratedThe point coordinates can be obtained by substituting the equation (7) into the equation (8), and the current vector curve at the time of the optimum efficiency is o-p₁. When ω is_r≥ω_rbaseA-B and A '-B' are two conditions when the torque is different, and the upper part of the curve is a selectable current scheme within a constraint range. If the minimum point of equation (12) is within the constraint range, i.e., A '-B' and o-p₁The intersection point of the two-dimensional model meets the voltage and current constraints, and the extreme point is the optimal solution; if the minimum value point is beyond the constraint range, the abscissa of the point B on the voltage ring is closest to the minimum value point, namelyAnd the point B is the optimal solution.

In the formula: i is_sIs the stator current; i is_maxIs the maximum stator current; u shape_maxIs the maximum stator voltage; u. of_sd、u_sqDq-axis components of the stator voltage, respectively; i.e. i_sd、i_sqDq-axis components of the stator current, respectively; r is the resistance of the stator; psi_d、ψ_qIs the dq axis component of the stator flux linkage; omega_rIs the mechanical angular velocity; l is_d、L_qDq-axis inductance components, respectively; psi_fIs a permanent magnet flux linkage; t is_eIs an electromagnetic torque, n_pIs the number of pole pairs.

Wherein the content of the first and second substances,

step three: analyzing the general rule of influencing current distribution according to different operating conditions, and constructing a current distribution vector control system as shown in fig. 2, wherein the control effect of the control system mainly depends on a current distribution module in a controller, input values such as rotating speed and torque are given according to the requirements of a motor, a current distribution model carries out current distribution according to the input values to obtain reference instruction currentOutput reference current commandThe control signal of the inverter is generated by PI regulation and coordinate transformation and then used as module input generated by SVPWM, so that the corresponding control effect is achieved, and the control of the high-performance motor is realized.

Step four: and (3) building an experimental platform based on an AVL dynamometer rack system, and testing efficiency points under different working conditions. The system shown in fig. 2 is used as a control system of the permanent magnet synchronous motor of the experimental platform to acquire measured data. In the experiment, the speed regulation range is 200r/min to 3000r/min, the rotating speed is set as a fixed value through a control console, the torque is adjusted on an upper computer, the actual measurement motor efficiency is obtained through AVL, and the adjustment is stopped when the power reaches the upper limit. And measuring several sets of parameters of motor efficiency, voltage, quadrature axis current, direct axis current and the like corresponding to the same rotating speed and different torques. And changing the set value of the rotating speed until the full range of speed regulation is covered, and finally using the measured data as a measured data sample for constructing a current distribution model based on the deep learning belief network regression algorithm.

Step five: the deep learning confidence network current regression model diagram is shown in FIG. 3, and the output layer outputsAnd (4) regression prediction results, and feature extraction is completed by each Restricted Boltzmann Machine (RBM) layer, which shows that the method can realize a multi-input multi-output operation mode. The optimal number of layers of the hidden layers, the optimal number of neurons in each layer and the initial learning rate of each RBM layer are obtained by continuously debugging according to training experience. The method adopts two stages of pre-training-fine tuning parameters to learn and train network parameters, and in the first stage, a greedy unsupervised learning algorithm is used for initializing the parameters of the whole model layer by layer; and in the second stage, the relevant parameters of the network space are subjected to fine adjustment from top to bottom by using a wake-sleep algorithm in a supervised manner.

The selection of the input feature set is one of important factors influencing the quality of the prediction result of the deep confidence network. To ensure i_sd、i_sqThe prediction accuracy of the motor is that different current analytic relations of the motor in a constant torque area and a constant power area are considered respectively to select characteristic values. In the constant torque region, a value (T) is selected according to the formulae (7), (8) and (10)_e、n、ψ_f、L_d、L_q、I_s) A 5-dimensional feature value; after multiple attempts, the DBN model is set to be 5 layers, the number of neurons in each layer is set to be 108, 80, 30 and 201, and the corresponding learning rates of all RBM layers are respectively 0.02, 0.1, 0.02, 0.01 and 0.01. In the constant power region, the power is selected according to the formulas (11), (13) and (14)_e、n、ψ_f、L_d、L_q、P_cu、P_fe、P_in、P_out、U_max、U_max、u_sd、u_sq) A 12-dimensional feature value; after multiple adjustments, the DBN model is set to be 7 layers, the number of neurons in each layer is set to be 202, 100, 40 and 201, and the corresponding learning rates of all RBM layers are respectively 0.03, 0.1, 0.02, 0.1, 0.03 and 0.01. And verifying the accuracy of the prediction of the deep confidence network regression modeling current by utilizing the actually measured current distribution discrete data in the constant torque area and the constant power area respectively. The invention adopts the following twoThe types of judgment standards are respectively the average absolute value error shown in the formula (15) and the root mean square error shown in the formula (16), and the smaller the numerical values of the two types of judgment standards are, the more accurate the predicted value is.

In the formula: y is_pTo predict the result, y_tAnd N is the number of the predicted samples as the actual measurement result.

Step six: and embedding the constructed deep confidence network current regression model into a vector control system, and comparing the operation results with the results shown in the figures 4, 5, 6 and 7. FIG. 4 is a comparison result between the predicted value and the measured value of the d-q axis current in the constant torque region; fig. 4 (a) and fig. 4 (b) are current regression curves at different torques at a given rotation speed of 600 rpm; fig. 4 (c) and 4 (d) are current prediction curves at different rotation speeds for a given torque of 500Nm, and illustrate that the current prediction accuracy is high in the present invention. According to the analysis of the formula (15), the overall average error between the measured value and the predicted value is less than 2%, and the error is within the actual engineering requirement range. FIG. 5 is a comparison result of the predicted current value and the measured current value of the d-q axis current in the constant power region. FIGS. 5(a) and 5(b) are current regression curves for different torques at a given speed of 1400 rpm; fig. 5 (c) and 5 (d) are predicted curves of the current at different rotational speeds for a given torque of 300 Nm. According to the analysis of the formula (15), the overall average error of the measured value and the predicted value is lower than 1.5%, and the actual engineering error requirement is met. In conclusion, the fitting degree of the current curve predicted by the DBN with the deep learning capability and the actually measured current curve is good. The errors of the predicted value and the measured value of the current distribution under different working conditions are less than 2 percent, and the requirement of the actual engineering error is met.

In order to verify the effectiveness of the current control strategy provided by the invention, the motor is experimentally verified in an AVL dynamometer. Fig. 6 and 7 are efficiency comparison diagrams of the DBN current prediction method and the conventional vector control method when the torque is constant under the conditions of constant torque and constant power, respectively, and the mechanical rotation speed range from 200r/min to 3000r/min is selected, and when the motor operation reaches a steady state, the deep confidence network current regression model of the invention effectively reduces the influence of parameter change in the actual operation of the motor by processing and analyzing the actual measurement data under different operation conditions. As can be seen from fig. 6, when the motor running speed is low, the efficiency is improved by about 5% when the rotation speed is 400r/min, and as can be seen from fig. 7, in the constant power region, the running efficiency of the motor controlled by the conventional VC decreases with the increase of the rotation speed, and the running phase ratio of the motor using the DBN current prediction method decreases relatively slowly. At a rotation speed of 2700r/min, the efficiency is improved by about 3 percent. Fig. 8 is an efficiency map of the motor adopting the DBN current prediction control method under all operating conditions, which illustrates that the present invention adopts discrete measured data for modeling, so that the deep learning belief network regression model solves the problem of discrete measured data discontinuity, realizes the current optimal matching at any operating point, and can improve the operating efficiency of the permanent magnet synchronous motor in the whole speed regulation range, as can be seen from fig. 8, the middle block represents a high-efficiency zone in which the electric vehicle runs more during operation, and the area of the high-efficiency zone is more than 80% to meet the requirements of the actual engineering.