WO2023124893A1 - Torque estimation method and apparatus based on neural network, and device and storage medium - Google Patents

Abstract

A torque estimation method and apparatus based on a neural network, and a device and a storage medium. The method comprises: acquiring an electric motor operation parameter in real time; performing feature engineering on the electric motor operation parameter, so as to obtain a feature vector, and inputting the feature vector into a trained BP neural network model, so as to obtain a first estimated torque value; inputting the electric motor operation parameter into a preset electric motor torque mathematical estimation model, so as to obtain a second estimated torque value; and performing weighted calculation according to the first estimated torque value and the second estimated torque value, so as to obtain a final estimated torque value. According to the present method, an electric motor torque is predicted by means of combining machine learning prediction with mathematical model calculation, the calculation amount is relatively small, and the prediction accuracy is improved.

Description

Torque estimation method, device, equipment and storage medium based on neural network technical field

The present application relates to the technical field of motor control, in particular to a neural network-based torque estimation method, device, equipment and storage medium.

Background technique

Permanent magnet synchronous motors have the advantages of high power density, high efficiency, and high torque density, and have been widely used in robotics, industrial automation, and electric vehicles. However, because the IPMSM is affected by factors such as the nonlinearity, volatility, and operational uncertainty of the motor parameters, it cannot accurately achieve high-precision control of the motor torque. Therefore, in the traditional motor torque control strategy, the control accuracy of the motor electromagnetic torque is relatively low, which degrades the overall performance of the system.

In order to improve the torque estimation accuracy of permanent magnet synchronous motors, scholars at home and abroad have developed many schemes, which can be divided into two types: offline and online. For offline solutions, it is usually obtained by looking up a table. The table lookup method is simulated based on off-line experiments or finite element analysis. The method based on the table lookup method is simple and robust, but it is very time-consuming to implement and requires a large number of hardware resources, occupy a large amount of storage space, and it is impractical to test on each machine, these factors greatly reduce the efficiency and application range of torque estimation. For online methods, the extended Kalman filter algorithm is usually used. The extended Kalman filter algorithm (EKF) is an algorithm derived by scholars based on the fact that the Kalman (KF) filter algorithm cannot effectively track nonlinear systems. It is an approximation to nonlinear systems. The processing method has a good estimation effect on the nonlinear characteristics of the permanent magnet synchronous motor, and has strong anti-interference performance, but the calculation amount is relatively large.

Contents of the invention

The present application provides a neural network-based torque estimation method, device, device and storage medium to solve the existing problems of large calculation and low accuracy of torque estimation.

In order to solve the above-mentioned technical problems, a technical solution adopted by this application is to provide a neural network-based torque estimation method, including: obtaining motor operating parameters in real time; performing feature engineering on the operating parameters of the motor to obtain feature vectors; Input the vector into the trained BP neural network model to obtain the first estimated torque value; input the motor operating parameters into the preset motor torque mathematical estimation model to obtain the second estimated torque value; The moment estimate and the second torque estimate are weighted to obtain a final torque estimate.

As a further improvement of the present application, the real-time acquisition of motor operating parameters includes: real-time acquisition of d-axis current, q-axis current and electrical angle parameters of the motor based on a preset acquisition device.

As a further improvement of the present application, the calculation formula of the final torque estimation value is:

表示最终转矩估测值，f(x _n)表示BP神经网络模型输出的第一转矩估测值，E(T)表示电机转矩数学估测模型输出的第二转矩估测值，α表示预设权重参数，β表示预设置参数。 in,

Represents the final torque estimation value, f(x _n ) represents the first torque estimation value output by the BP neural network model, E(T) represents the second torque estimation value output by the motor torque mathematical estimation model, α represents a preset weight parameter, and β represents a preset parameter.

As a further improvement of the present application, the step of pre-training the BP neural network model includes: constructing the BP neural network model to be trained; obtaining the input parameters of the training samples and the actual torque value corresponding to the input parameters of the training samples; inputting the parameters of the training samples Perform feature engineering to obtain sample feature vectors; input sample feature vectors to BP neural network model to obtain sample torque estimates; backpropagate updates based on pre-built loss functions, sample torque estimates, and actual torque values BP neural network model; repeating the above training process until the BP neural network model reaches a preset accuracy or the number of iterations reaches a preset number of times.

As a further improvement of the present application, after obtaining the final torque estimate based on the weighted calculation of the first torque estimate and the second torque estimate, it further includes: when the currently running control system is a torque control system, Closed loop control of motor torque based on final torque estimate.

As a further improvement of the present application, the motor torque is closed-loop controlled according to the final torque estimation value, including: obtaining the calculation time consumed by calculating the final torque estimation value and the processing time consumed by the control system operation; if the current operation If the processor of the device is a single-core processor, the motor control cycle is set as the sum of calculation time and processing time, and according to the motor control cycle, the motor torque is calculated with the final torque estimate obtained in the motor control cycle Closed-loop control; if the processor of the currently running device is a multi-core processor, set the motor control cycle to the larger value of the calculation time and processing time, and according to the motor control cycle, use the final torque obtained in the motor control cycle The estimated value performs closed-loop control of the motor torque.

As a further improvement of the present application, after the final torque estimate is obtained through weighted calculations based on the first torque estimate and the second torque estimate, it also includes: when the currently running control system is a non-torque control system , output the final torque estimate as observed data.

In order to solve the above technical problems, another technical solution adopted by the present application is to provide a neural network-based torque estimation device, including: an acquisition module for acquiring motor operating parameters in real time; a first estimation module for Perform feature engineering on the motor operating parameters to obtain the feature vector, input the feature vector to the trained BP neural network model, and obtain the first torque estimation value; the second estimation module is used to input the motor operating parameters to the preset The motor torque mathematical estimation model is used to obtain a second torque estimation value; the calculation module is used to calculate and obtain a final torque estimation value according to the weighted calculation of the first torque estimation value and the second torque estimation value.

In order to solve the above technical problems, another technical solution adopted by the present application is to provide a computer device, the computer device includes a processor, a memory coupled to the processor, and program instructions are stored in the memory, so When the program instructions are executed by the processor, the processor is made to execute the steps of the above neural network-based torque estimation method.

In order to solve the above-mentioned technical problems, another technical solution adopted by the present application is to provide a storage medium storing program instructions capable of realizing the above-mentioned neural network-based torque estimation method.

The beneficial effects of the present application are: the neural network-based torque estimation method of the present application obtains real-time motor operating parameters, and then inputs the motor operating parameters into the trained BP neural network model and motor torque mathematical estimation model to obtain the first estimated torque value and the second estimated torque value, and then calculate the final estimated torque value according to the first estimated torque value and the second estimated torque value. On the one hand, this kind The estimation method combines two prediction results of machine learning prediction and mathematical model prediction, which has better stability and higher prediction accuracy. On the other hand, this method of combining machine learning prediction and mathematical model prediction is relatively The Mann filter algorithm has a smaller amount of calculation, and can realize online prediction, does not need offline table lookup, and has lower dependence on device hardware resources.

Description of drawings

FIG. 1 is a schematic flow chart of a neural network-based torque estimation method according to a first embodiment of the present invention;

2 is a schematic diagram of functional modules of a neural network-based torque estimation device according to an embodiment of the present invention;

Fig. 3 is a schematic structural diagram of a computer device according to an embodiment of the present invention;

FIG. 4 is a schematic structural diagram of a storage medium according to an embodiment of the present invention.

Detailed ways

The following will clearly and completely describe the technical solutions in the embodiments of the present application with reference to the accompanying drawings in the embodiments of the present application. Obviously, the described embodiments are only part of the embodiments of the present application, not all of them. Based on the embodiments in this application, all other embodiments obtained by persons of ordinary skill in the art without making creative efforts belong to the scope of protection of this application.

The terms "first", "second", and "third" in this application are used for descriptive purposes only, and cannot be understood as indicating or implying relative importance or implicitly specifying the quantity of indicated technical features. Thus, features defined as "first", "second", and "third" may explicitly or implicitly include at least one of these features. In the description of the present application, "plurality" means at least two, such as two, three, etc., unless otherwise specifically defined. All directional indications (such as up, down, left, right, front, back...) in the embodiments of the present application are only used to explain the relative positional relationship between the various components in a certain posture (as shown in the drawings) , sports conditions, etc., if the specific posture changes, the directional indication also changes accordingly. Furthermore, the terms "include" and "have", as well as any variations thereof, are intended to cover a non-exclusive inclusion. For example, a process, method, system, product or device comprising a series of steps or units is not limited to the listed steps or units, but optionally also includes unlisted steps or units, or optionally further includes For other steps or units inherent in these processes, methods, products or apparatuses.

Reference herein to an "embodiment" means that a particular feature, structure, or characteristic described in connection with the embodiment can be included in at least one embodiment of the present application. The appearances of this phrase in various places in the specification are not necessarily all referring to the same embodiment, nor are independent or alternative embodiments mutually exclusive of other embodiments. It is understood explicitly and implicitly by those skilled in the art that the embodiments described herein can be combined with other embodiments.

FIG. 1 is a schematic flowchart of a neural network-based torque estimation method according to a first embodiment of the present invention. It should be noted that the method of the present invention is not limited to the flow sequence shown in FIG. 1 if substantially the same result is obtained. As shown in Figure 1, the method includes steps:

Step S101: Acquiring motor operating parameters in real time.

It should be noted that this embodiment is aimed at real-time and online control of the torque of the motor during the real-time operation of the motor.

Specifically, during the operation of the motor, the motor operating parameters of the motor are collected in real time through the device.

Wherein, the real-time acquisition of motor operating parameters includes: real-time acquisition of d-axis current, q-axis current and electrical angle parameters of the motor based on a preset acquisition device.

Step S102: Perform feature engineering on the operating parameters of the motor to obtain feature vectors, and input the feature vectors into the trained BP neural network model to obtain the first estimated torque value.

Specifically, after obtaining the operating parameters of the motor, perform feature engineering on the operating parameters of the motor to obtain the eigenvector of the operating parameters of the motor, and then input the eigenvector into the pre-trained BP neural network model, and obtain The first torque estimate.

It should be noted that the BP neural network model includes an input layer, a hidden layer, and an output layer. Among them, output layer parameters include input neuron x _n ; hidden layer parameters include hidden layer number, hidden layer neuron h _p , weight w _np and threshold b _pq from input layer to hidden layer, hidden layer to hidden layer The weight w _pq and the threshold b _pq of the output layer; the output layer parameters include output layer neurons T _q . The internal input and output relationship of the model is as follows:

ho _h (n) = f (hi _h (n)), h = 1, 2, ..., p;

To _o (n) = f (Ti _o (n)), o = 1, 2, ..., q;

where the activation function is:

f(x)=1/(1+e ^-x );

The model performance evaluation standard is based on the mean square error (RMSE) between the output value and the true value:

The neural network model can construct an accurate end-to-end mapping relationship for nonlinear models and complex models, simplifying the complexity of modeling.

Further, before using the BP neural network model to predict, the BP neural network model needs to be trained, and the steps of pre-training the BP neural network model include:

1. Construct the BP neural network model to be trained.

2. Obtain the training sample input parameters and the actual torque value corresponding to the training sample input parameters.

Specifically, the sample input parameters and the actual torque value corresponding to each sample input parameter are collected during the actual operation of the motor. In , due to the mechanical response delay, that is, the torque signal is later than the motor modulation control current signal for a certain period, it is necessary to calibrate and match the collected data.

3. Perform feature engineering on the input parameters of the training samples to obtain the sample feature vectors.

4. Input the sample feature vector into the BP neural network model to obtain the estimated value of the sample torque.

5. Update the BP neural network model based on the pre-built loss function, sample torque estimation value and actual torque value backpropagation.

Repeat the above training process until the BP neural network model reaches the preset accuracy or the number of iterations reaches the preset number.

In this embodiment, by continuously screening and adjusting the network model structure during the training process, including the number of network neurons, the number of hidden layers, the learning rate, and the number of training times, the BP neural network model reaches a preset accuracy or iteration The number of times reaches the preset number of times, and a trained BP neural network model is obtained.

Step S103: Input the motor operating parameters into a preset motor torque mathematical estimation model to obtain a second torque estimation value.

Specifically, under the two-phase coordinate system, the mathematical estimation model of the motor torque is expressed as:

Among them, U represents the voltage, L represents the inductance, i represents the current, d represents the d axis, q represents the q axis, R represents the stator coil of the motor, ψ represents the flux linkage of the motor, p represents the number of pole pairs of the motor, T _e represents the second torque Estimated value, w is the angular velocity of the motor.

Step S104: weighted calculation according to the first estimated torque value and the second estimated torque value to obtain a final estimated torque value.

It should be noted that, in the actual application of the motor torque mathematical estimation model, there is a large error between the output torque and the actual torque due to nonlinear factors such as system parameter changes, magnet saturation, and harmonic disturbances during motor operation. Therefore, in this embodiment, the prediction result of the motor torque mathematical estimation model is integrated with the prediction result of the BP neural network model, thereby improving the estimation stability and accuracy.

Further, the calculation formula of the final torque estimation value is:

表示最终转矩估测值，f(x _n)表示BP神经网络模型输出的第一转矩估测值，E(T)表示电机转矩数学估测模型输出的第二转矩估测值，α表示预设权重参数，β表示预设置参数。具体地，该α和β由用户预先根据经验设置。 in,

Represents the final torque estimation value, f(x _n ) represents the first torque estimation value output by the BP neural network model, E(T) represents the second torque estimation value output by the motor torque mathematical estimation model, α represents a preset weight parameter, and β represents a preset parameter. Specifically, the α and β are preset by the user based on experience.

Further, in some embodiments, the final torque estimation value can also be used to perform torque control on the motor, therefore, the final torque estimation value is calculated according to the weighted calculation of the first torque estimation value and the second torque estimation value to obtain the final torque After the moment estimates, also include:

When the currently running control system is a torque control system, the motor torque is closed-loop controlled according to the final torque estimation value.

Specifically, after embedding the above-mentioned BP neural network model and motor torque mathematical estimation model into the motor control program, if the current control system is a torque control system, the d Axis current, q-axis current and electrical angle parameters, and then estimate the d-axis current, q-axis current and electrical angle parameters to obtain the final torque estimation value, and then perform closed-loop control on the torque according to the final torque estimation value .

It should be noted that the difference in the number of processor cores of the current running device of the motor control program will cause differences in the execution of the motor control. Therefore, the closed-loop control of the motor torque is performed according to the final torque estimation value, including:

1. Obtain the calculation time consumed for calculating the final torque estimation value and the processing time consumed for the operation of the control system.

Among them, the calculation time refers to the time consumed by the BP neural network model and the motor torque mathematical estimation model from obtaining the motor operating parameters to obtaining the final torque estimation value according to the motor operating parameters, and the processing time refers to the time consumed by the operation of the control system time.

2. If the processor of the currently running device is a single-core processor, set the motor control cycle to the sum of the calculation time and processing time, and according to the motor control cycle, use the final torque estimate obtained in the motor control cycle Closed-loop control of motor torque.

3. If the processor of the currently running device is a multi-core processor, set the motor control cycle to the larger value of the calculation time and processing time, and according to the motor control cycle, use the final torque obtained in the motor control cycle to estimate The measured value performs closed-loop control on the motor torque.

Specifically, when the current device is a single-core processor, the BP neural network model and the motor torque mathematical estimation model are in a serial relationship with the motor control program. Therefore, it is necessary to run the BP neural network after the motor control program is completed. model and motor torque mathematical estimation model, at this time, the motor control cycle is set to the sum of calculation time and processing time, for example, the motor control program can be completed in about 25us, and this BP neural network model and motor rotation The torque mathematical estimation model takes 85us to calculate the final torque estimation value, so the motor control cycle is set to 150us.

When the current device is a multi-core processor, the BP neural network model and the motor torque mathematical estimation model are in parallel with the motor control program. Therefore, the motor control program and the BP neural network model and the motor torque mathematical estimation model can be simultaneously Execution, at this time, the motor control cycle is set to the larger value of the calculation time and processing time, for example, the motor control program can be completed in about 25us, and this time the BP neural network model and the motor torque mathematical estimation model It takes 85us to calculate the final torque estimation value, so the motor control cycle is set to 100us.

It can be seen that the single-thread processing platform needs longer calculation time, and the motor control performance is slightly reduced, while the multi-thread processing platform has better motor control performance.

Further, after the weighted calculation of the first torque estimated value and the second torque estimated value to obtain the final torque estimated value, further includes:

When the currently operating control system is a non-torque control system, the final estimated torque value is output as observation data.

In the neural network-based torque estimation method of the first embodiment of the present invention, after obtaining real-time motor operating parameters, the motor operating parameters are respectively input into the trained BP neural network model and the motor torque mathematical estimation model to obtain The first torque estimation value and the second torque estimation value, and then calculate the final torque estimation value according to the first torque estimation value and the second torque estimation value. On the one hand, the estimated The method combines machine learning prediction and mathematical model prediction with better stability and higher prediction accuracy. On the other hand, this method combines machine learning prediction and mathematical model prediction compared with the extended Kalman filter algorithm The amount of calculation is smaller, and online prediction can be realized, without the need for offline table lookup, and the dependence on device hardware resources is lower.

FIG. 2 is a schematic diagram of functional modules of a neural network-based torque estimation device according to an embodiment of the present invention. As shown in FIG. 2 , the neural network-based torque estimation device 20 includes an acquisition module 21 , a first estimation module 22 , a second estimation module 23 and a calculation module 24 .

Obtaining module 21, used for obtaining motor operating parameters in real time;

The first estimation module 22 is used to perform feature engineering on the motor operating parameters to obtain the feature vector, and input the feature vector to the trained BP neural network model to obtain the first estimated torque value;

The second estimation module 23 is configured to input the motor operating parameters into a preset motor torque mathematical estimation model to obtain a second torque estimation value;

The calculation module 24 is configured to obtain a final torque estimation value through weighted calculation according to the first torque estimation value and the second torque estimation value.

Optionally, the acquiring module 21 performs the operation of acquiring the operating parameters of the motor in real time, specifically including: acquiring d-axis current, q-axis current and electrical angle parameters of the motor in real time based on a preset acquisition device.

Optionally, the calculation formula of the final torque estimation value is:

表示最终转矩估测值，f(x _n)表示BP神经网络模型输出的第一转矩估测值，E(T)表示电机转矩数学估测模型输出的第二转矩估测值，α表示预设权重参数，β表示预设置参数。 in,

Represents the final torque estimation value, f(x _n ) represents the first torque estimation value output by the BP neural network model, E(T) represents the second torque estimation value output by the motor torque mathematical estimation model, α represents a preset weight parameter, and β represents a preset parameter.

Optionally, it also includes a training module, which is used to perform the operation of pre-training the BP neural network model, specifically including: constructing the BP neural network model to be trained; obtaining the training sample input parameters and the actual torque corresponding to the training sample input parameters value; perform feature engineering on the input parameters of the training samples to obtain the sample feature vector; input the sample feature vector to the BP neural network model to obtain the estimated value of the sample torque; based on the pre-built loss function, the estimated value of the sample torque and the actual The torque value is backpropagated to update the BP neural network model; the above training process is repeated until the BP neural network model reaches a preset accuracy or the number of iterations reaches a preset number.

Optionally, after the calculation module 24 executes the operation of weighting and calculating the final torque estimation value according to the first torque estimation value and the second torque estimation value, it is also used for: when the currently running control system is torque When controlling the system, the motor torque is closed-loop controlled based on the final torque estimate.

Optionally, the calculation module 24 executes an operation of performing closed-loop control on the motor torque according to the final torque estimation value, specifically including: obtaining the calculation time consumed by calculating the final torque estimation value and the processing time consumed by the control system operation ; If the processor of the currently running device is a single-core processor, set the motor control cycle to the sum of the calculation time and processing time, and according to the motor control cycle, use the final torque estimate obtained in the motor control cycle to compare The motor torque is closed-loop controlled; if the processor of the currently running device is a multi-core processor, the motor control cycle is set to the larger value of the calculation time and processing time, and according to the motor control cycle, the motor control cycle is obtained Closed-loop control of the motor torque is performed using the final torque estimate.

Optionally, after the calculation module 24 executes the operation of weighting and calculating the final torque estimation value according to the first torque estimation value and the second torque estimation value, it is also used for: when the currently running control system is non-rotating In the torque control system, the final torque estimated value is output as observation data.

For other details of implementing the technical solution of each module in the neural network-based torque estimation device of the above-mentioned embodiment, please refer to the description of the neural network-based torque estimation method in the above-mentioned embodiment, which will not be repeated here.

It should be noted that each embodiment in this specification is described in a progressive manner, and each embodiment focuses on the differences from other embodiments. For the same and similar parts in each embodiment, refer to each other. Can. As for the device-type embodiments, since they are basically similar to the method embodiments, the description is relatively simple, and for related parts, please refer to part of the description of the method embodiments.

Please refer to FIG. 3 . FIG. 3 is a schematic structural diagram of a computer device according to an embodiment of the present invention. As shown in FIG. 6, the computer device 60 includes a processor 61 and a memory 62 coupled to the processor 61. Program instructions are stored in the memory 62. When the program instructions are executed by the processor 61, the processor 61 performs any of the above-mentioned operations. The steps of the neural network-based torque estimation method described in the embodiment.

Wherein, the processor 61 may also be called a CPU (Central Processing Unit, central processing unit). The processor 61 may be an integrated circuit chip with signal processing capability. The processor 61 can also be a general-purpose processor, a digital signal processor (DSP), an application-specific integrated circuit (ASIC), a field programmable gate array (FPGA) or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components . A general-purpose processor may be a microprocessor, or the processor may be any conventional processor, or the like.

Referring to FIG. 4 , FIG. 4 is a schematic structural diagram of a storage medium according to an embodiment of the present invention. The storage medium in the embodiment of the present invention stores program instructions 71 capable of realizing all the above-mentioned methods, wherein the program instructions 71 can be stored in the above-mentioned storage medium in the form of software products, including several instructions to make a computer device (which can It is a personal computer, a server, or a network device, etc.) or a processor (processor) that executes all or part of the steps of the methods described in the various embodiments of the present application. The aforementioned storage media include: U disk, mobile hard disk, read-only memory (ROM, Read-Only Memory), random access memory (RAM, Random Access Memory), magnetic disk or optical disc, etc., which can store program codes. , or computer equipment such as computers, servers, mobile phones, and tablets.

In the several embodiments provided in this application, it should be understood that the disclosed computer equipment, devices and methods may be implemented in other ways. For example, the device embodiments described above are only illustrative. For example, the division of units is only a logical function division. In actual implementation, there may be other division methods. For example, multiple units or components can be combined or integrated. to another system, or some features may be ignored, or not implemented. In another point, the mutual coupling or direct coupling or communication connection shown or discussed may be through some interfaces, and the indirect coupling or communication connection of devices or units may be in electrical, mechanical or other forms.

In addition, each functional unit in each embodiment of the present invention may be integrated into one processing unit, each unit may exist separately physically, or two or more units may be integrated into one unit. The above-mentioned integrated units can be implemented in the form of hardware or in the form of software functional units. The above is only the implementation mode of this application, and does not limit the scope of patents of this application. Any equivalent structure or equivalent process transformation made by using the contents of this application specification and drawings, or directly or indirectly used in other related technical fields, All are included in the scope of patent protection of the present application in the same way.

Claims (10)

A method for estimating torque based on neural network, characterized in that it comprises: Real-time acquisition of motor operating parameters; Carrying out feature engineering on the operating parameters of the motor to obtain a feature vector, and inputting the feature vector into a trained BP neural network model to obtain a first estimated torque value; Inputting the motor operating parameters into a preset motor torque mathematical estimation model to obtain a second torque estimation value; A final torque estimate is obtained through weighted calculation based on the first torque estimate and the second torque estimate. The method for estimating torque based on a neural network according to claim 1, wherein said obtaining motor operating parameters in real time comprises: Real-time acquisition of d-axis current, q-axis current and electrical angle parameters of the motor based on a preset acquisition device. The torque estimation method based on neural network according to claim 1, wherein the calculation formula of the final torque estimation value is:

in,

Represents the final torque estimate, f(x _n ) represents the first torque estimate output by the BP neural network model, E(T) represents the second output torque of the motor torque mathematical estimate model Torque estimation value, α represents a preset weight parameter, and β represents a preset parameter. The method for estimating torque based on neural network according to claim 1, wherein the step of pre-training the BP neural network model comprises: Build the described BP neural network model to be trained; Obtain training sample input parameters and actual torque values corresponding to the training sample input parameters; performing feature engineering on the training sample input parameters to obtain sample feature vectors; Inputting the sample feature vector into the BP neural network model to obtain a sample torque estimated value; Updating the BP neural network model based on a pre-built loss function, the sample torque estimate and the actual torque value backpropagation; Repeat the above training process until the BP neural network model reaches the preset accuracy or the number of iterations reaches the preset number. The method for estimating torque based on neural network according to claim 1, characterized in that, said weighted calculation according to said first torque estimating value and said second torque estimating value obtains the final torque estimating After measuring, also include: When the currently running control system is a torque control system, the motor torque is closed-loop controlled according to the final estimated torque value. The neural network-based torque estimation method according to claim 5, wherein the closed-loop control of the motor torque according to the final torque estimation value comprises: obtaining computation time consumed in computing the final torque estimate and processing time consumed in operation of the control system; If the processor of the currently running device is a single-core processor, set the motor control cycle as the sum of the calculation time and the processing time, and according to the motor control cycle, use the motor control cycle obtained in the motor control cycle The final torque estimate performs closed-loop control of the motor torque; If the processor of the currently running device is a multi-core processor, the motor control cycle is set to the larger value of the calculation time and the processing time, and according to the motor control cycle, within the motor control cycle The acquired final torque estimate performs closed-loop control on the motor torque. The method for estimating torque based on neural network according to claim 1, characterized in that, said weighted calculation according to said first torque estimating value and said second torque estimating value obtains the final torque estimating After measuring, also include: When the currently operating control system is a non-torque control system, the final estimated torque value is output as observation data. A kind of torque estimation device based on neural network, it is characterized in that, comprises: An acquisition module is used to acquire motor operating parameters in real time; The first estimation module is used to perform feature engineering on the motor operating parameters to obtain a feature vector, and input the feature vector to a trained BP neural network model to obtain a first estimated torque value; A second estimating module, configured to input the motor operating parameters into a preset motor torque mathematical estimating model to obtain a second torque estimating value; A calculation module, configured to obtain a final torque estimation value through weighted calculation based on the first torque estimation value and the second torque estimation value. A computer device, characterized in that the computer device includes a processor and a memory coupled to the processor, and program instructions are stored in the memory, and when the program instructions are executed by the processor, the The processor executes the steps of the neural network-based torque estimation method according to any one of claims 1-7. A storage medium, characterized by storing program instructions capable of realizing the neural network-based torque estimation method according to any one of claims 1-7.

Images