A Tracking Turntable Deceleration Control System and Method Based on ANN Neural Network

Abstract

The present invention discloses a deceleration control system for a tracking turntable based on an ANN neural network, and also discloses a deceleration control method for a tracking turntable based on an ANN neural network. It includes a self-check and debugging module and a model training module. The self-check and debugging module is used to test the corresponding overshoot amount σ by fixing the deceleration gradient a and changing the deceleration rate b; establish a database according to the obtained data; the model training module obtains a deceleration prediction model through ANN neural network model training and verification. Set the overshoot amount σ through the upper computer, and the rotation speed sensor set on the tracking turntable feeds back the rotation speed. Substitute the obtained deceleration gradient a and overshoot amount σ into the deceleration prediction model to obtain the deceleration rate b; perform deceleration control according to the obtained deceleration gradient a, overshoot amount σ and deceleration rate b. In this way, it is possible to improve the control of the overshoot amount of the tracking turntable and improve the accuracy of the tracking turntable.

Description

Technical Field

[0001] The present invention belongs to the technical field of servo control of tracking turntables, and specifically relates to a deceleration control system and method for a tracking turntable. Background Art

[0002] The accuracy of the turntable directly affects the tracking accuracy of the entire tracking system. Taking the azimuth axis of the tracking system as an example, the existing mature method uses a three-loop control algorithm of current loop, speed loop, and position loop to achieve control. However, in actual engineering, the tracking turntable has characteristics such as heavy load, large friction, and high tracking accuracy requirements. Therefore, higher requirements are imposed on acceleration and deceleration, especially deceleration. If the deceleration is too large, the tracking turntable will be sluggish and affect the tracking accuracy. If the deceleration is too large, an overshoot phenomenon of the tracking turntable will occur, that is, it cannot brake in time, which is likely to cause safety accidents. Therefore, controlling the overshoot of the tracking turntable is an urgent problem to be solved in the turntable tracking system.

[0003] Patent (110690835A) discloses a method for motor acceleration and deceleration using an S-shaped curve, and the equation of the S-shaped curve makes the deceleration process smoother. Deceleration equation: y = a / (1 + e -bx ), where a is the deceleration gradient coefficient, which directly reflects the speed difference of deceleration; b is the deceleration rate coefficient, which directly reflects the slope of deceleration. Usually, the deceleration equation of the S-shaped curve is solidified in the deceleration system to make its deceleration process smooth. However, different deceleration gradient coefficients a and different deceleration rate coefficients b will affect the overshoot of deceleration. Therefore, the fixed deceleration gradient coefficient a and deceleration rate coefficient b are not sufficient to meet the control requirements of the overshoot of the tracking turntable, let alone meet the accuracy requirements of the tracking turntable. Summary of the Invention

[0004] The purpose of the present invention is to provide a deceleration control system for a tracking turntable based on an ANN neural network to solve the problem of overshoot in the deceleration of the tracking turntable in the prior art mentioned in the background art; in addition, a deceleration control method for a tracking turntable based on an ANN neural network is also provided.

[0005] To solve the above technical problems, the technical solution adopted by the present invention is:

[0006] A deceleration control system for a tracking turntable based on an ANN neural network includes a self-check and debugging module and a model training module; the self-check and debugging module is used to obtain the overshoot σ through the deceleration equation y = a / (1 + e -bx );

[0007] In the formula, the variable x is time, and y is the actual speed; theoretically, a is y(0) - y(x), and in fact, during the deceleration process, overshoot occurs, and y(x) + Δ(x) appears. The overshoot σ is Among them, the value of the fixed deceleration gradient a is kept unchanged, while the value of the deceleration rate b is changed to obtain the corresponding overshoot value σ; a database is established based on the obtained values of the deceleration gradient a, the deceleration rate b, and the overshoot σ.

[0008] The database is established as follows:

[0009] Step 1: Through the DSP main control chip, establish a deceleration equation, and manually set the values of the deceleration gradient a and the deceleration rate b, and record the overshoot σ of different deceleration gradients a and different deceleration rates b.

[0010] Step 2: Store the data through DDR; including the deceleration gradient a, the deceleration rate b, and the velocity overshoot σ.

[0011] Step 3: Export the data stored in DDR as a txt format and import it into MATLAB to generate a neural network database.

[0012] The model training module uses the ANN neural network model to set the learning data of the input layer and the output layer of the ANN neural network model, and uses the obtained database to train and verify the ANN neural network model; the ANN neural network model with qualified verification is used to predict the deceleration control to obtain a deceleration prediction model.

[0013] The network error E is defined as:

[0014]

[0015] where: d i is the ideal output of the network; y i is the actual output of the network;

[0016] The problem of minimizing the network error can be expressed as:

[0017] minE(W) W = [w1, w2, …, w l , ] T

[0018] w1 ∈ [a i , b i i = 1, 2, …, l

[0019] where: l is the number of elements of the optimization vector W; the network error E is the objective function value corresponding to W; the GA+BP hybrid neural network algorithm is adopted; that is, first use GA to initially optimize the weight W, and then use the BP algorithm to iteratively optimize the weight W.

[0020] The GA+BP hybrid neural network algorithm is as follows:

[0021] Let the velocity overshoot of the deceleration gradient a and the deceleration rate b at time T be x (0)(T), the fitted value of the network is is the average value of the original data, that is e(T) is the difference between the observed value and the fitted value, also known as the residual at time T, that is is the average value of the original data, that is The mean square error of the residuals is Then the posterior difference ratio C is defined as C = S2 / S1, and the small error probability is defined as A small C value indicates that the mean square error S1 is large and the mean square error of the residuals S2 is small; a large S1 means that the data dispersion of the overshoot is large and the regularity of the original data is poor, while a small S2 indicates that the prediction error dispersion is small.

[0022] Furthermore, a deceleration control module is provided. The deceleration control module is electrically connected to a controller, and the deceleration control module is used to write the deceleration prediction model into the controller.

[0023] Furthermore, a rotational speed sensor is provided on the tracking turntable to feedback the rotational speed and obtain the value of the deceleration gradient a. The deceleration gradient a = y(0) - y(s), where y(0) is the current speed measured by the rotational speed sensor at the start of deceleration, and y(s) is the specified speed to which deceleration is required. If deceleration to a standstill is required, then y(s) = 0; according to the performance requirements of the tracking turntable, the value of the overshoot σ is set by the host computer, and the obtained value of the deceleration gradient a and the value of the overshoot σ are brought into the deceleration prediction model in the controller to obtain the deceleration rate b; the controller controls the deceleration according to the deceleration prediction model.

[0024] Furthermore, the ANN uses a neural network model with an input layer width of 2, a hidden layer width of 3, and an output layer width of 1 for training.

[0025] Furthermore, a neural network training model is adopted, which has self - adaptability for the uncertainty of the overshoot σ control system in this case and can approximate any complex non - linear relationship.

[0026] Furthermore, in the operation database, it includes the influence on the overshoot σ control based on the deceleration curve y = a / (1 + e -bx ) division, including the combined influence of the deceleration gradient a and the deceleration rate b; it can make a deceleration decision according to the requirement of the overshoot σ.

[0027] A deceleration control method for a tracking turntable based on an ANN neural network includes the following steps:

[0028] S1: Power on the deceleration control system of the tracking turntable and initialize the parameters of the controller;

[0029] S2: Determine whether the system has received a new device self-check command; if yes, the device starts the self-check and debugging module; if no, proceed to the next step;

[0030] Start the self-check and debugging module to start the self-check; during the self-check, determine whether the automatic debugging module has completed the self-check. If yes, export the values of the deceleration gradient a, deceleration rate b, and overshoot σ for each group to establish a database, use the database to train and verify the ANN neural network model, use the verified ANN neural network model to predict the deceleration control to obtain a deceleration prediction model, and then download the deceleration prediction model to the controller; if no, continue the self-check;

[0031] S3: Determine whether the controller has a deceleration prediction model. If yes, proceed to the next step; if no, return to step S2;

[0032] S4: The tracking turntable device starts to start and run; the speed sensor detects the current speed; the host computer calculates the deceleration gradient a;

[0033] S5: Input the allowable value of the overshoot σ through the host computer;

[0034] S6: Determine whether the system has received the allowable value command of the overshoot σ; if yes, proceed to the next step, if no, return to step S5 and prompt to input the allowable value of the overshoot σ;

[0035] S7: The controller predicts the deceleration rate b through the deceleration prediction model, and substitutes the deceleration rate b into the deceleration equation to generate a deceleration curve;

[0036] S8: Determine whether the system has received a start deceleration command; if yes, proceed to the next step; if no, return to step S4;

[0037] S9: The controller controls the turntable device to perform a deceleration action, and the tracking turntable device starts to decelerate according to the deceleration curve.

[0038] Further, when the tracking turntable device starts to decelerate, the overshoot σ of the tracking turntable is detected by the detection device; determine whether the overshoot σ meets the requirements; if yes, the deceleration is completed; if no, issue a warning of "the device needs to be self-checked again".

[0039] Further, the specific process of the self-check of the self-check and debugging module is as follows:

[0040] A1: Set the initial deceleration gradient a = 20;

[0041] A2: Set the initial deceleration rate b = 1;

[0042] A3: Execute the loop process b = b + 0.1; Determine whether b ≥ 5 in each loop; If not, run and record the overshoot σ; If so, run the next step;

[0043] A4: Execute the loop process a = a + 5; Determine whether a ≥ 100 in each loop; If not, return to step A3; If so, end the self-check.

[0044] Compared with the prior art, the present invention has the following beneficial effects:

[0045] The present invention obtains three data, namely the deceleration gradient a, the deceleration rate b, and the overshoot σ, through self-check. Through the ANN neural network model, the learning data of the input layer and the output layer of the ANN neural network model are set, and the obtained database is used to train and verify the ANN neural network model; The verified ANN neural network model is used to predict the deceleration control to obtain a deceleration prediction model. The controller downloads the deceleration prediction model. After the tracking turntable starts, the allowable value of the overshoot σ is input through the host computer, and the predicted deceleration rate b is calculated to generate a predicted deceleration curve. After the system receives the deceleration command, the tracking turntable starts to decelerate according to the deceleration curve. In this way, it is possible to improve the control of the overshoot of the tracking turntable and improve the accuracy of the tracking turntable.

[0046] Using the self-check data of its own system as the neural network database to eliminate the influence of mechanical, electrical, environmental and other parameters of different devices, which has higher accuracy. Description of the Drawings

[0047] Figure 1 is the control flow chart of the present invention;

[0048] Figure 2 is the process of establishing the operation database of the present invention;

[0049] Figure 3 is to establish a curve for the deceleration gradient a;

[0050] Figure 4 is to establish a curve for the deceleration rate b;

[0051] Figure 5 is the neural network training model;

[0052] Figure 6 is the deceleration flow chart of the present invention. Detailed Embodiment

[0053] Next, the technical solutions in the embodiments of the present invention will be clearly and completely described in conjunction with the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are only a part of the embodiments of the present invention, rather than all the embodiments. All other embodiments obtained by those of ordinary skill in the art based on the embodiments of the present invention without creative efforts shall fall within the protection scope of the present invention.

[0054] Embodiment 1

[0055] A tracking turntable deceleration control system based on an ANN neural network includes a self-checking and debugging module and a model training module; the self-checking and debugging module is used to obtain the overshoot amount σ through the deceleration equation y = a / (1 + e -bx )

[0056] In the formula, the variable x is time, and y is the actual speed; theoretically, in the process of x changing with time, the speed decelerates from the starting speed y(0) to y(x), that is, the speed gradient a that needs to be decelerated is y(0) - y(x). In fact, during the deceleration process, overshoot occurs, and y(x) + Δ(x) appears. The overshoot amount σ is Among them, by fixing the value of the deceleration gradient a and changing the value of the deceleration rate b, the corresponding value of the overshoot amount σ is obtained; specifically: fix a to a certain gradient value, change the value of the deceleration rate b, and obtain the value of the overshoot amount σ corresponding to the current gradient a. Stepwise change a to an increasing gradient value, and obtain the value of the overshoot amount σ corresponding to the corresponding gradient a. For example,

[0057] Fix a = 20, and change b = 1, 1.2, 1.3,...

[0058] Gradient increase a = 25, and change b = 1, 1.2, 1.3,...

[0059] Gradient increase a = 30, and change b = 1, 1.2, 1.3,...

[0060] Establish a database according to the obtained values of the deceleration gradient a, the deceleration rate b, and the overshoot amount σ.

[0061] The database establishment method is as follows:

[0062] The first step: Through the DSP main control chip, build a deceleration equation, and manually set the value of the deceleration gradient a and the value of the deceleration rate b, and record the overshoot amount σ of different deceleration gradients a and different deceleration rates b;

[0063] Write the dsp code as follows:

[0064] float a = 20.0; / / Define and initialize the deceleration gradient a, that is, the speed that needs to be finally reached

[0065] float b = 1; / / Define and initialize the deceleration rate b

[0066] float time; / / Define the time to start decelerating

[0067] float Vel; / / Define the actual feedback speed during operation

[0068] float VelMax; / / Define the overshoot speed Δ

[0069] float VelRate; / / Define the speed overshoot amount σ

[0070] float VelSet; / / Define the set speed during deceleration

[0071] if (VelSet > VelTar) / / The set speed is greater than the target speed, indicating that the deceleration requirement has not been met, execute the deceleration function

[0072] {

[0073] VelSet = a / (1 + pow(e, -b * time)); / / Set the speed at the current deceleration state

[0074] time++;

[0075] VelMax = Vel - a; / / Record the overshoot speed Δ

[0076] VelRate = VelMax / a; / / Calculate the speed overshoot amount σ

[0077] }

[0078] Step 2: Store data through DDR; including deceleration gradient a, deceleration rate b, and speed overshoot amount σ;

[0079] Step 3: Export the DDR stored data as a txt format and import it into MATLAB to generate a neural network database;

[0080] The model training module uses the ANN neural network model to set the learning data for the input layer and output layer of the ANN neural network model, and trains and validates the ANN neural network model using the obtained database. Among them, 70% of the data in the database is the training set, and 30% of the data is the validation set; the ANN neural network model with qualified validation is used to predict the deceleration control to obtain a deceleration prediction model;

[0081] In order to obtain the corresponding relationship between different deceleration gradients a and different deceleration rates b and the overshoot σ, it is established through a neural network model. For a neural network, once the network structure and training samples are determined, the error of the network is completely determined by the network weights W. Therefore, the training process of the neural network is also the process of finding a set of network weights W to minimize the corresponding network error E. The network error E is defined as:

[0082]

[0083] where: d i is the ideal output of the network; y i is the actual output of the network;

[0084] The problem of minimizing the network error can be expressed as:

[0085] minE(W) W=[w1,w2,…,w l ,] T

[0086] w1∈[a i ,b i i=1,2,…,l

[0087] where: l is the number of elements of the optimization vector W; the network error E is the objective function value corresponding to W; the GA+BP hybrid neural network algorithm is adopted; that is, first use GA to initially optimize the weight W, and then use the BP algorithm to iteratively optimize the weight W;

[0088] The GA+BP hybrid neural network algorithm has 4 steps:

[0089] (1) Set k = 0, determine the population size N and the fitness value F corresponding to the objective function. The fitness value of the jth individual adopts the form of F j =E max -E ij where E max is the maximum value of the objective function in the ith generation, and E ij is the objective function value of the jth individual in the ith generation. Randomly generate an initial population of size N, where each individual corresponds to a connection relationship of the BP network, and the connection weights are between (-1, 1).

[0090] (2) Perform the crossover and mutation algorithm operations of GA to obtain N new individuals, calculate their fitness values respectively, and perform individual selection according to the optimal individual retention strategy according to the fitness value size to achieve survival of the fittest.

[0091] (3) If the preset maximum number of iterations K is reached or the given accuracy requirement ε is satisfied, the solution W * and the target value E * are obtained. If the accuracy requirement is satisfied, that is, E* ≤ ε, the algorithm ends; otherwise, go to step (4). If the maximum number of generations k is not reached, let k = k + 1 and return to step (2).

[0092] (4) Take W * as the initial weight, and use the BP algorithm to learn the weight parameters of the neural network. The iterative learning reaches the maximum number of generations Epochs or reaches the termination condition, and the algorithm ends.

[0093] The GA+BP hybrid neural network algorithm is as follows:

[0094] Since there is no obvious rule among the deceleration gradient a, deceleration rate b, and deceleration overshoot σ. To verify the effectiveness of the network model, the posterior difference test method is used to analyze the fitting results of the model. Let the velocity overshoot of the deceleration gradient a and deceleration rate b at time T be x (0) (T), and the fitting value of the network is is the average value of the original data, that is e(T) is the difference between the observed value and the fitting value, also called the residual at time T, that is is the average value of the original data, that is The mean square error of the residuals is Then the posterior difference ratio C is defined as C = S2 / S1, and the small error probability is defined as That is, the indicators for evaluating the extrapolation performance of the network are: the C value and the P value. A small C value indicates a large mean square error S1 and a small mean square error of residuals S2; a large S1 indicates a large data dispersion of the test overshoot and a poor regularity of the original data, and a small S2 indicates a small prediction error dispersion. Therefore, it is required that S2 is as small as possible under the premise of a large S1, that is, the smaller the C value, the better. It shows that although the original data has poor regularity, the prediction error has a small swing range. In addition, the extrapolation performance of the network can be evaluated through the small error probability P. The larger P is, the stronger the generalization ability of the network.

[0095] Using the self-check data of its own system as the neural network database to exclude the influence of mechanical, electrical, environmental and other parameters of different devices, which has higher accuracy.

[0096] In a preferred embodiment, a deceleration control module is further provided. The deceleration control module is electrically connected to a controller, and the deceleration control module is used to write the deceleration prediction model into the controller.

[0097] In a preferred embodiment, a rotational speed sensor is provided on the tracking turntable to feedback the rotational speed and obtain the value of the deceleration gradient a, where a = y(0) - y(s). In the formula, y(0) is the current speed measured by the rotational speed sensor at the start of deceleration, and y(s) is the specified speed to which deceleration is required. If deceleration to a standstill is required, then y(s) = 0. According to the performance requirements of the tracking turntable, the value of the overshoot σ is set by the host computer. The obtained value of the deceleration gradient a and the value of the overshoot σ are substituted into the deceleration prediction model in the controller to obtain the deceleration rate b. The controller controls the deceleration according to the deceleration prediction model.

[0098] In a preferred embodiment, the ANN is trained using a neural network model with an input layer width of 2, a hidden layer width of 3, and an output layer width of 1.

[0099] In a preferred embodiment, a neural network training model is adopted. For the uncertainty of the overshoot σ control system in this case, it has self - adaptability and can approximate any complex non - linear relationship.

[0100] In a preferred embodiment, in the running database, it includes the influence on the overshoot σ control which is based on the division of the deceleration curve y = a / (1 + e -bx ) and includes the combined influence of the deceleration gradient a and the deceleration rate b; it can make a deceleration decision according to the requirement of the overshoot σ.

[0101] A deceleration control method for a tracking turntable based on an ANN neural network includes the following steps:

[0102] S1: Power on the deceleration control system of the tracking turntable and initialize the parameters of the controller;

[0103] S2: Determine whether the system receives a new device self - check command; if so, the device starts the self - check and debugging module; if not, proceed to the next step;

[0104] Start the self - check and debugging module to start self - checking; during self - checking, determine whether the automatic debugging module has completed self - checking. If so, export the corresponding values of the deceleration gradient a, the deceleration rate b, and the overshoot σ to establish a database, use the database to train and verify the ANN neural network model, obtain a deceleration prediction model by predicting the deceleration control with the verified ANN neural network model, and then download the deceleration prediction model to the controller; if not, continue self - checking;

[0105] S3: Determine whether the controller has a deceleration prediction model; if so, proceed to the next step; if not, return to step S2;

[0106] S4: The tracking turntable device starts to run; the rotational speed sensor detects the current rotational speed; the host computer calculates the deceleration gradient a;

[0107] S5: Input the allowable value of overshoot σ through the host computer;

[0108] S6: Determine whether the system has received the command of the allowable value of overshoot σ; if yes, proceed to the next step, if no, return to step S5 and prompt to input the allowable value of overshoot σ;

[0109] S7: The controller predicts the deceleration rate b through the deceleration prediction model, and substitutes the deceleration rate b into the deceleration equation to generate a deceleration curve;

[0110] S8: Determine whether the system has received the start deceleration command; if yes, proceed to the next step; if no, return to step S4;

[0111] S9: The controller controls the turntable device to perform a deceleration action, and tracks the turntable device to start decelerating according to the deceleration curve.

[0112] In a preferred embodiment, when the tracked turntable device starts to decelerate, the overshoot σ of the tracked turntable is detected by the detection device; it is judged whether the overshoot σ meets the requirements; if yes, the deceleration is completed; if no, a warning of "the device needs to be self-checked again" is issued.

[0113] In a preferred embodiment, the specific process of self-checking of the self-checking and debugging module is as follows:

[0114] A1: Set the initial deceleration gradient a = 20;

[0115] A2: Set the initial deceleration rate b = 1;

[0116] A3: Execute the loop process b = b + 0.1; judge whether b ≥ 5 in each loop; if no, record the overshoot σ; if yes, proceed to the next step;

[0117] A4: Execute the loop process a = a + 5; judge whether a ≥ 100 in each loop; if no, return to step A3; if yes, end the self-check.

[0118] In this way, the control of the overshoot of the tracked turntable can be achieved, and the accuracy of the tracked turntable during rotation is improved.

[0119] Embodiment 2

[0120] During the deceleration control process:

[0121] First step, the tracked turntable system is powered on, and the controller (DSP chip main control) initializes the parameters of the control system;

[0122] Second step, the controller judges whether there is a new device self-check command from the host computer. If yes, execute the third step; if no, execute the eighth step;

[0123] In the third step, start the device self-check mode, input the deceleration gradient a (from 20 to 100, step size 5), input the deceleration rate b (from 1 to 5, step size 0.1), and measure the overshoot σ of the device;

[0124] In the fourth step, the controller determines whether there is a self-check completion command. If there is, execute the fifth step; if not, return to the third step;

[0125] In the fifth step, export the data from the controller, mainly including the input data (deceleration gradient a, deceleration rate b) and the output data (overshoot σ);

[0126] In the sixth step, store the data in the PC, generate an ANN model training library, and use the neural network algorithm to obtain a deceleration prediction model between the input and output;

[0127] In the seventh step, download the generated deceleration prediction model to the controller;

[0128] In the eighth step, the controller determines whether there is a deceleration prediction model. If there is, execute the ninth step; if not, return to the third step;

[0129] In the ninth step, the turntable device starts to run, the sensor measures the current rotation speed, calculates the deceleration gradient a, and inputs the allowable value of the overshoot σ through the host computer;

[0130] In the tenth step, the controller determines whether it has received the command of the allowable value of the overshoot σ. If there is, execute the eleventh step; if not, return to the ninth step;

[0131] In the eleventh step, predict the deceleration rate b through the prediction model, substitute it into the deceleration equation y = a / (1 + e -bx ), and generate a deceleration curve;

[0132] In the twelfth step, the controller determines whether it has received the start deceleration command. If there is, execute the thirteenth step; if not, return to the ninth step;

[0133] In the thirteenth step, execute the deceleration action. The turntable device starts to decelerate according to the deceleration curve and detects the overshoot σ after deceleration;

[0134] In the fourteenth step, the controller determines whether the overshoot σ meets the requirements. If it meets the requirements, the deceleration is completed; if it does not meet the requirements, a warning that the device needs to be self-checked again is issued.

[0135] Although the embodiments of the present invention have been shown and described, for those of ordinary skill in the art, it can be understood that various changes, modifications, substitutions, and variations can be made to these embodiments without departing from the principle and spirit of the present invention. The scope of the present invention is defined by the appended claims and their equivalents.

Claims

1. A tracking turntable deceleration control system based on an ANN neural network, characterized in that: It includes a self-check and debugging module and a model training module; the self-check and debugging module is used to obtain the overshoot σ through the deceleration equation y = a / (1 + e -bx ); The overshoot σ is as follows: Among them, the value of the fixed deceleration gradient a is kept unchanged, and the value of the deceleration rate b is changed to obtain the corresponding value of the overshoot σ; a database is established based on the obtained values of the deceleration gradient a, the deceleration rate b, and the overshoot σ. The database is established as follows: Step 1: Through the DSP main control chip, build a deceleration equation, and manually set the values of the deceleration gradient a and the deceleration rate b, and record the overshoot σ of different deceleration gradients a and different deceleration rates b; Step 2: Store data through DDR; including the deceleration gradient a, the deceleration rate b, and the velocity overshoot σ; Step 3: Export the data stored in DDR to a txt format and import it into MATLAB to generate a neural network database; The model training module sets the learning data of the input layer and the output layer of the ANN neural network model through the ANN neural network model, and uses the obtained database to train and verify the ANN neural network model; Use the ANN neural network model with qualified verification to predict the deceleration control to obtain a deceleration prediction model; For a neural network, after the network structure and training samples are determined, the network error E is determined by the weights W of the network; the network error E is defined as: where: d i is the ideal output of the network; y i is the actual output of the network; The problem of minimizing the network error can be expressed as: minE(W)W=[w1,w2,…,w l ,] T w1∈[a i ,b i for i = 1, 2, …, l In the formula: l is the number of elements of the optimization vector W; the network error E is the objective function value corresponding to W; the GA+BP hybrid neural network algorithm is adopted; that is, first use GA to initially optimize the weights W, and then use the BP algorithm to iteratively optimize the weights W; The GA+BP hybrid neural network algorithm is as follows: Let the velocity overshoot of the deceleration gradient \(a\) and the deceleration rate \(b\) at time \(T\) be \(x\). (0) (T), the fitted value of the network is is the average value of the original data, that is \(e(T)\) is the difference between the observed value and the fitted value, also called the residual at time \(T\), that is is the average value of the original data, that is The mean square error of the residuals is Then the posterior difference ratio \(C\) is defined as \(C = S2 / S1\), and the small error probability is defined as A small \(C\) value indicates a large mean square error \(S1\) and a small mean square error of the residuals \(S2\); a large \(S1\) means a large dispersion of the overshoot data and a poor regularity of the original data, and a small \(S2\) indicates a small dispersion of the prediction error.

2. The deceleration control system of a tracking turntable based on an ANN neural network according to claim 1, characterized in that: There is also a deceleration control module, and the deceleration control module is electrically connected to a controller. The deceleration control module is used to write the deceleration prediction model into the controller.

3. The tracking turntable deceleration control system based on an ANN neural network according to claim 2, characterized in that: A rotational speed sensor is provided on the tracking turntable to feedback the rotational speed and obtain the value of the deceleration gradient a. The deceleration gradient a = y(0) - y(s), where y(0) is the current speed measured by the rotational speed sensor at the start of deceleration, and y(s) is the specified speed to which it needs to decelerate. If it needs to decelerate to a standstill, then y(s) = 0; according to the performance requirements of the tracking turntable, set the value of the overshoot σ through the host computer, and substitute the obtained value of the deceleration gradient a and the value of the overshoot σ into the deceleration prediction model in the controller to obtain the deceleration rate b; the controller controls the deceleration according to the deceleration prediction model.

4. A tracking turntable deceleration control system based on an ANN neural network according to claim 1, characterized in that: ANN uses a neural network model with an input layer width of 2, a hidden layer width of 3, and an output layer width of 1 for training.

5. A tracking turntable deceleration control system based on an ANN neural network according to claim 1, characterized in that: Adopt a neural network training model, which has self-adaptability for the uncertainty of the overshoot σ control system in this case and can approximate any complex nonlinear relationship.

6. The tracking turntable deceleration control system based on an ANN neural network according to claim 1, characterized in that: In the running database, it contains that the influence on the overshoot σ control is based on the deceleration curve y = a / (1 + e -bx ) division, which includes the combined influence of the deceleration gradient a and the deceleration rate b; and it can make a deceleration decision according to the overshoot σ requirement.

7. A deceleration control method for a tracking turntable based on an ANN neural network, characterized in that: It includes the following steps: S1: Power on the deceleration control system of the tracking turntable and initialize the parameters of the controller; S2: Determine whether the system receives a new device self-check command; If yes, the device starts the self-check and debugging module; if no, proceed to the next step; Start the self-check and debugging module to start self-checking; during self-checking, determine whether the automatic debugging module has completed self-checking. If yes, export the corresponding values of the deceleration gradient a, the deceleration rate b, and the overshoot σ to establish a database, use the database to train and verify the ANN neural network model, use the ANN neural network model with qualified verification to predict the deceleration control to obtain a deceleration prediction model, and then download the deceleration prediction model to the controller; if no, continue self-checking; S3: Determine whether the controller has a deceleration prediction model. If yes, proceed to the next step; if no, return to step S2. S4: The tracking turntable device starts to operate. The rotational speed sensor detects the current rotational speed. The host computer calculates the deceleration gradient a. S5: Input the allowable value of overshoot σ through the host computer. S6: Determine whether the system has received the command for the allowable value of overshoot σ. If yes, proceed to the next step; if no, return to step S5 and prompt to input the allowable value of overshoot σ. S7: The controller predicts the deceleration rate b through the deceleration prediction model, and substitutes the deceleration rate b into the deceleration equation to generate a deceleration curve. S8: Determine whether the system has received the start deceleration command. If yes, proceed to the next step; if no, return to step S4. S9: The controller controls the tracking turntable device to perform a deceleration action, and the tracking turntable device starts to decelerate according to the deceleration curve.

8. A deceleration control method for a tracking turntable based on an ANN neural network according to claim 7, characterized in that: When the tracking turntable device starts to decelerate, the detection device detects the overshoot σ of the tracking turntable. Determine whether the overshoot σ meets the requirements. If yes, the deceleration is completed; if no, issue a warning of "The device needs to be self - tested again".

9. A deceleration control method for a tracking turntable based on an ANN neural network according to claim 7, characterized in that: The specific process of self - testing of the self - testing and debugging module is as follows: A1: Set the initial deceleration gradient a = 20. A2: Set the initial deceleration rate b = 1. A3: Execute the loop process b = b + 0.1; determine whether b ≥ 5 in each loop. If no, record the overshoot σ; if yes, proceed to the next step. A4: Execute the loop process a = a + 5; determine whether a ≥ 100 in each loop. If no, return to step A3; if yes, end the self - testing.

Images