AGV path tracking and optimizing method based on neural network algorithm of DSP, computer storage medium and equipment

Abstract

The invention relates to a path optimization method, in particular to an AGV path tracking and optimization method based on a neural network algorithm of a DSP, a computer storage medium and equipment, and aims at solving the problems that an existing AGV path planning algorithm is prone to being interfered by the environment, low in convergence speed and prone to falling into the difficulty of local optimum. The method comprises the following steps: 1) establishing a motion state mathematical model of the AGV; 2) establishing a servo control system mathematical model of the AGV; 3) constructing a neural network and learning to obtain a neural network training result; and 4) according to a neural network training result, optimizing a motion state mathematical model of the AGV and a servo control system mathematical model of the AGV, and performing a simulation test.

Description

Technical Field

[0001] The invention relates to a path optimization method, and in particular to an AGV path tracking and optimization method based on a DSP-based neural network algorithm, a computer storage medium and a device. Background Art

[0002] An Automated Guided Vehicle (AGV) is an unmanned automated transport vehicle that uses batteries as a power source and is equipped with a non-contact control guide device. With the gradual development of factory automation and the intelligent manufacturing industry, AGVs are widely used for their high efficiency and flexibility. To a certain extent, they can replace manual work, reduce labor costs and improve production efficiency. However, for limited usage scenarios or environments and an increased number of AGVs or tasks to be performed, it is extremely important to plan the path for each AGV so that it can reach the target point without encountering static obstacles or dynamic conflicts. Consider the position of each AGV in time and resolve route conflicts between multiple AGVs. If the command changes, such as a temporarily added task or the AGV deviates from its expected path, these centralized solutions may need to replan all AGV tasks and routes. At this time, the AGV vehicle cannot switch to the new command execution state in a short time, which will cause a certain time delay and path conflict. Therefore, the path planning problem for AGV is a current research hotspot.

[0003] Path planning refers to the optimal path planning problem of a robot, that is, finding an optimal path from the starting state to the target state that avoids obstacles in the workspace based on one or more optimization criteria (such as minimum work cost, shortest walking path, shortest walking time, etc.). The application scenarios of robot path planning are extremely rich, the most common of which are the position movement of NPCs and controlled characters in games, navigation problems in Baidu Maps, small as the cleaning routes of household sweeping robots, the operating paths of drones, and large as the automatic driving paths of driverless cars that companies are competing to develop. For the path planning problem of AGV, the current mainstream algorithms include A* algorithm and standard intelligent optimization algorithm; A* algorithm, as an extension of Dijkstra algorithm, is widely used in path finding and graph traversal due to its high efficiency, but when there are multiple minimum values, A* algorithm cannot guarantee the optimal search path, and both algorithms are easily disturbed by the environment, have slow convergence speed and are prone to fall into the dilemma of local optimality. Summary of the invention

[0004] The purpose of the present invention is to solve the problems that the existing AGV path planning algorithm is easily disturbed by the environment, has a slow convergence speed and is prone to fall into the local optimal dilemma, and to provide an AGV path tracking and optimization method, computer storage medium and device based on a DSP-based neural network algorithm.

[0005] To achieve the above purpose, the technical solution provided by the present invention is:

[0006] A DSP-based neural network algorithm AGV path tracking and optimization method, which is special in that it includes the following steps:

[0007] Step 1, establish a mathematical model of the motion state of the AGV, obtain the motion state equation of the AGV in the two-dimensional plane at time t and the motion state equation of the AGV in the two-dimensional plane at time t+Δt, and calculate the observation equation Z of the AGV relative to the target point according to the motion state equation of the AGV in the two-dimensional plane at time t and the motion state equation of the AGV in the two-dimensional plane at time t+Δt;

[0008] Step 2: Establish a mathematical model of the servo control system of the AGV and obtain the total transfer function G(s) of the servo control system;

[0009] Step 3: According to the observation equation Z of the AGV relative to the target point and the total transfer function G(s) of the servo control system, a neural network is constructed and learned until the error reaches a preset requirement or the number of learning times reaches a preset requirement, and the neural network training result is obtained;

[0010] Step 4: Call the neural network training result obtained in step 3 in the computer software, optimize the observation equation Z of the AGV relative to the target point and the total transfer function G(s) of the servo control system, and obtain the optimized observation equation Z1 of the AGV relative to the target point and the optimized total transfer function G1(s) of the servo control system. Then call the neural network training result obtained in step 3, perform simulation test according to Z1 and G1(s), and obtain the test result. If the test result does not meet the preset requirements, use Z1 and G1(s) to update Z and G(s), return to step 3, and rebuild the neural network for training; if the test result meets the preset requirements, the path planning of the AGV is completed.

[0011] Furthermore, in step 1, the motion state equation of the AGV in the two-dimensional plane at time t is:

[0012]

[0013] Wherein, t is the time, x(t) and y(t) are the x-direction position and y-direction position of the AGV at time t, respectively, u(t) is the angle between the moving direction of the AGV at time t and the x-axis, v(t) is the speed of the AGV at time t, ω(t) is the angular velocity of the AGV at time t, and a(t) is the acceleration of the AGV at time t;

[0014] d is the differential symbol;

[0015] At time t+Δt, the motion state equation of AGV in the two-dimensional plane is:

[0016]

[0017] Among them, V max and V min They represent the maximum speed and minimum speed of the AGV between t and t+Δt respectively;

[0018] At time t, the observation equation Z of the AGV relative to the target point is:

[0019]

[0020] Where D represents the distance between the current position of the AGV and the target point. is the azimuth of the AGV’s current position relative to the target point, x d and d They represent the x-direction position coordinate and y-direction position coordinate of the target point, respectively. t and t They represent the x-direction position coordinate and y-direction position coordinate of the AGV at time t respectively.

[0021] Further, in step 2, the servo control system mathematical model includes a servo driver, a servo motor, a reduction device and a controlled object connected in sequence, as well as a servo controller, a current feedback module, a speed feedback module, a posture feedback module, a first operator and a second operator; the first input end of the first operator is connected to the input end of the AGV servo control system, the output end is connected to the input end of the servo controller, the output end of the servo controller is connected to the first input end of the second operator, the input end of the current feedback module is connected between the servo motor and the reduction device, the output end is connected to the second input end of the second operator, and the output end of the second operator is connected to the input end of the servo driver; the input end of the speed feedback module is arranged between the reduction device and the controlled object, the output end is connected to the second input end of the first operator, the output end of the controlled object is connected to the input end of the posture feedback module, and the output end of the posture feedback module is connected to the third input end of the first operator;

[0022] The calculation formula of the total transfer function G(s) of the servo control system is:

[0023]

[0024] Among them, C(s) is the output function of the AGV servo control system, R(s) is the input function of the AGV servo control system; G1(s) and G2(s) are abstract functions of the servo controller in the servo control system, G3(s) is the abstract function of the servo driver in the servo control system, G4(s) is the abstract function of the servo motor in the servo control system, and G5(s) is the abstract function of the deceleration device in the servo control system; H2(s) is the abstract function of the speed feedback module of the servo control system, and H3(s) is the abstract function of the current feedback module in the servo control system.

[0025] Furthermore, step 3 is specifically as follows:

[0026] Step 3.1, initialize the neural network, the neural network includes an input layer, a hidden layer and an output layer set in sequence, set the input layer weight set {ω it}、Hidden layer weight set {ω ij} and the output layer weight set {ω ik}, and the input layer threshold set {θ t}、Hidden layer threshold set {θ j} and 00 output layer threshold set {θ k}, define the number of learning times N = 0, preset error e, and the maximum number of learning times N1; where i, j, t, and k all represent the position information of neurons in different layers, the first letter of the ω subscript represents the number of layers of the neural network, and the second letter of the ω subscript represents the sequence number of the neuron; each layer of the neural network has one and only one threshold, which represents the connection relationship between different layers;

[0027] Step 3.2: Calculate the input and output values ​​of the hidden layer neurons; the activation function of the hidden layer is where y j is the output value of the hidden layer, x j is the input variable of the hidden layer, and m is the number of records;

[0028] Step 3.3, according to the observation equation Z of the AGV relative to the target point and the total transfer function G(s) of the servo control system, calculate the input value and output value of the output layer neuron; the activation function of the output layer is expressed as where y k is the output value of the output layer, g k is the input variable of the output layer; f(x) = tanx (x≠(π / 2)+βπ, β is an integer);

[0029] Step 3.4: Calculate the generalized error of each neuron in the hidden layer and output layer and Then let N=N+1;

[0030] and The specific calculation formula is as follows:

[0031]

[0032] Among them, is the input variable of the output layer neuron, is the input variable of the input layer neuron, is the input variable of the hidden layer neuron;

[0033]

[0034] is the expected output value, is the actual output value; is the output error value,

[0035] z is the data recording variable, ω zk 、ω zt 、ω zj and s z are all input variables; a, b, and n are all the number of recording times, and their values are automatically obtained by the system;

[0036] Step 3.5: Judge the value of N. If N≥N1, then execute Step 3.7. If N<N1, then compare and with e respectively. If and are both less than e, then execute Step 3.7. If and at least one is greater than or equal to e, then execute Step 3.6;

[0037] Step 3.6: Use and obtained in Step 3.4 to update the weights ω it 、ω ij and ω tk , as well as the thresholds θ t 、θ j and θ k , and then return to Step 3.2; the specific calculation formula is as follows:

[0038]

[0039]

[0040]

[0041]

[0042]

[0043]

[0044] Where η and α are learning rates, 0<η<1, 0<α<1; p is the number of records, and its value is automatically taken by the system; Δω is the weight change value before and after two times, and △θ is the threshold change value before and after two times; and are the input variables of neurons in the hidden layer, output layer, and input layer respectively;

[0045] Step 3.7: Obtain the neural network training results, that is, the optimized input layer weight set {ω it}、Hidden layer weight set {ω ij}、Output layer weight set {ω tk}、The input layer threshold set {θ t}、Hidden layer threshold set {θ j} and the output layer threshold set {θ k}.

[0046] At the same time, the present invention also provides a computer storage medium on which a computer program is stored, and its special feature is that when the computer program is executed by a processor, the steps of the above-mentioned AGV path tracking and optimization method based on a DSP-based neural network algorithm are implemented.

[0047] In addition, the present invention also provides a computer device, including a processor, a memory connected to the processor, and a computer program that can be run on the processor, the special feature of which is that when the processor executes the computer program, the steps of the above-mentioned AGV path tracking and optimization method based on the DSP neural network algorithm are implemented.

[0048] Compared with the prior art, the present invention has the following beneficial effects:

[0049] The AGV path tracking and optimization method based on the DSP neural network algorithm provided by the present invention utilizes the neural network to process and fuse the sensor data, integrates the observation quantities provided by different types of sensors through the data fusion method, reduces the possible contradictions between multiple sensors, reduces the environmental interference, and thus improves the accuracy of the equipment status detection system failure; at the same time, based on the advantages of the neural network such as nonlinear mapping ability, self-learning ability, generalization ability, and fault tolerance, it can enhance the stability of the system and enhance the response and processing speed of the overall system. BRIEF DESCRIPTION OF THE DRAWINGS

[0050] Figure 1 It is a flow chart of an embodiment of the AGV path tracking and optimization method based on the DSP neural network algorithm of the present invention;

[0051] Figure 2 It is a specific flow chart of step 3 in the embodiment of the AGV path tracking and optimization method based on the DSP neural network algorithm of the present invention;

[0052] Figure 3 The result of the simulation test using MATLAB in step 4 of the embodiment of the AGV path tracking and optimization method based on the DSP neural network algorithm of the present invention Figure 1 ;

[0053] Figure 4 The result of the simulation test using MATLAB in step 4 of the embodiment of the AGV path tracking and optimization method based on the DSP neural network algorithm of the present invention Figure 2 ;

[0054] Figure 5 It is a schematic diagram of the mathematical model of the servo control system of the AGV in step 2 of the embodiment of the AGV path tracking and optimization method based on the DSP neural network algorithm of the present invention;

[0055] Figure 6 A schematic diagram of a neural network structure constructed in step 3 of an embodiment of an AGV path tracking and optimization method based on a DSP neural network algorithm of the present invention;

[0056] Figure 7 It is a schematic diagram of the control circuit structure of AGV;

[0057] Description of reference numerals:

[0058] 21. Servo controller; 22. Servo driver; 23. Servo motor; 24. Speed ​​reducer; 25. Controlled object; 26. Posture feedback module; 27. Speed ​​feedback module; 28. Current feedback module; 31. Input layer; 32. Hidden layer; 33. Output layer; 61. Environmental data; 62. Sensor; 63. Amplifier circuit; 64. Switch signal; 65. Other input signals; 66. DSPTMS320F2812 module; 6601. DI module; 6602. ADC module; 6603. Driver software; 6604. eCAN module; 6605. Other input modules; 67. JTAG interface circuit; 68. Clock oscillation; 69. Power supply circuit; 610. Host computer. DETAILED DESCRIPTION

[0059] The present invention will be further described below in conjunction with the accompanying drawings and specific embodiments.

[0060] A DSP-based neural network algorithm for AGV path tracking and optimization, see Figure 1 , including the following steps:

[0061] Step 1, establish a mathematical model of the motion state of the AGV, obtain the motion state equation of the AGV in the two-dimensional plane at time t and the motion state equation of the AGV in the two-dimensional plane at time t+Δt, and calculate the observation equation Z of the AGV relative to the target point according to the motion state equation of the AGV in the two-dimensional plane at time t and the motion state equation of the AGV in the two-dimensional plane at time t+Δt;

[0062] The motion state equation of AGV in the two-dimensional plane at time t is:

[0063]

[0064] Wherein, t is the time, xt and yt are the x-direction position and y-direction position of the AGV at time t, respectively, ut is the angle between the moving direction of the AGV at time t and the x-axis, vt is the speed of the AGV at time t, ωt is the angular velocity of the AGV at time t, and at is the acceleration of the AGV at time t;

[0065] d is the differential symbol;

[0066] At time t+Δt, the motion state equation of AGV in the two-dimensional plane is:

[0067]

[0068] Among them, V max and V min They represent the maximum speed and minimum speed of the AGV between t and t+Δt respectively;

[0069] At time t, the observation equation Z of the AGV relative to the target point is:

[0070]

[0071] Where D represents the distance between the current position of the AGV and the target point. is the azimuth of the AGV’s current position relative to the target point, x d and d They represent the x-direction position coordinate and y-direction position coordinate of the target point, respectively. t and t They represent the x-direction position coordinate and y-direction position coordinate of the AGV at time t respectively.

[0072] Step 2: Establish the mathematical model of the servo control system of the AGV and obtain the total transfer function Gs of the servo control system; the mathematical model of the servo control system can be found in Figure 5, including a servo driver 22, a servo motor 23, a speed reduction device 24 and a controlled object 25 connected in sequence, as well as a servo controller 21, a current feedback module 28, a speed feedback module 27, a posture feedback module 26, a first operator and a second operator; the first input end of the first operator is connected to the input end of the AGV servo control system, the output end is connected to the input end of the servo controller 21, the output end of the servo controller 21 is connected to the first input end of the second operator, the input end of the current feedback module 28 is connected between the servo motor 23 and the speed reduction device 24, the output end is connected to the second input end of the second operator, and the output end of the second operator is connected to the input end of the servo driver 22; the input end of the speed feedback module 27 is arranged between the speed reduction device 24 and the controlled object 25, the output end is connected to the second input end of the first operator, the output end of the controlled object 25 is connected to the input end of the posture feedback module 26, and the output end of the posture feedback module 26 is connected to the third input end of the first operator;

[0073] The calculation formula of the total transfer function G(s) of the servo control system is:

[0074]

[0075] Among them, C(s) is the output function of the AGV servo control system, R(s) is the input function of the AGV servo control system; G1(s) and G2(s) are abstract functions of the servo controller 21 in the servo control system, G3(s) is the abstract function of the servo driver 22 in the servo control system, G4(s) is the abstract function of the servo motor 23 in the servo control system, and G5(s) is the abstract function of the deceleration device 24 in the servo control system; H2(s) is the abstract function of the speed feedback module 27 of the servo control system, and H3(s) is the abstract function of the current feedback module 28 in the servo control system.

[0076] Step 3: See Figure 2 , according to the observation equation Z of the AGV relative to the target point and the total transfer function G(s) of the servo control system, a neural network is constructed and learned until the error reaches the preset requirement or the number of learning times reaches the preset requirement, and the neural network training result is obtained;

[0077] Step 3.1, initialize the neural network, see Figure 6 The neural network includes an input layer 31, a hidden layer 32 and an output layer 33, and sets the input layer weight set {ω it}、Hidden layer weight set {ω ij} and the output layer weight set {ω ik}, and the input layer threshold set {θ t}、Hidden layer threshold set {θ j} and 00 output layer threshold set {θk}, define the number of learning times N = 0, preset error e, and the maximum number of learning times N1; where i, j, t, and k all represent the position information of neurons in different layers, the first letter of the ω subscript represents the number of layers of the neural network, and the second letter of the ω subscript represents the sequence number of the neuron; each layer of the neural network has one and only one threshold, which represents the connection relationship between different layers;

[0078] Step 3.2, calculate the input value and output value of the neurons in the hidden layer 32; the activation function of the hidden layer 32 is where y j is the output value of hidden layer 32, x j is the input variable of the hidden layer 32, and m is the number of records;

[0079] Step 3.3, according to the observation equation Z of the AGV relative to the target point and the total transfer function Gs of the servo control system, calculate the input value and output value of the neuron in the output layer 33; the activation function of the output layer 33 is expressed as where y k is the output value of the output layer 33, g k is the input variable of the output layer 33; f(x)=tanx(x≠(π / 2)+βπ, β is an integer);

[0080] Step 3.4: Calculate the generalized error of each neuron in the hidden layer 32 and the output layer 33 and Then let N=N+1;

[0081] and The specific calculation formula is:

[0082]

[0083] in, is the input variable of the 33 neurons in the output layer, is the input variable of the 31 neurons in the input layer, is the input variable of 32 neurons in the hidden layer;

[0084]

[0085] is the expected output value, is the actual output value; is the output error value,

[0086] z is the data recording variable, ω zk ,ω zt ,ω zj and zAll are input variables; a, b, and n are all the number of records, and their values are automatically assigned by the system;

[0087] Step 3.5: Judge the value of N. If N≥N1, execute Step 3.7. If N<N1, then compare and with e respectively. If and are both less than e, execute Step 3.7. If and at least one is greater than or equal to e, execute Step 3.6;

[0088] Step 3.6: Use the and obtained in Step 3.4 to update the weight ω it , ω ij and ω tk , as well as the threshold values θ t , θ j and θ k , and then return to Step 3.2; the specific calculation formula is:

[0089]

[0090]

[0091]

[0092]

[0093]

[0094]

[0095] where η and α are both learning rates, 0<η<1, 0<α<1; p is the number of records, and its value is automatically assigned by the system; Δω is the change value of the weight between two consecutive times, and △θ is the change value of the threshold between two consecutive times; and are the input variables of the neurons in the hidden layer 32, output layer 33, and input layer 31 respectively;

[0096] Step 3.7: Obtain the neural network training results, that is, the optimized input layer weight set {ω it}, hidden layer weight set {ω ij}, output layer weight set {ω tk}, input layer threshold set {θ t}, hidden layer threshold set {θ j}, and output layer threshold set {θ k}.

[0097] Step 4: Call the neural network training result obtained in step 3 in the computer software, optimize the observation equation Z of the AGV relative to the target point and the total transfer function G(s) of the servo control system, and obtain the optimized observation equation Z1 of the AGV relative to the target point and the optimized total transfer function G1(s) of the servo control system. Then call the neural network training result obtained in step 3, perform simulation test according to Z1 and G1(s), and obtain the test result. If the test result does not meet the preset requirements, use Z1 and G1(s) to update Z and G(s), return to step 3, and rebuild the neural network for training; if the test result meets the preset requirements, the path planning of the AGV is completed.

[0098] The test results of this example are shown in Figure 3 and Figure 4 ,in Figure 3 The horizontal axis is the number of test sets, and the vertical axis is the simulation output result. Figure 3 The blue curve is the actual output curve, and the orange curve is the simulated output curve. Figure 3 It can be seen that the fitting effect is good. Figure 4 The horizontal axis is the step size and the vertical axis is the average variance. Figure 4 The blue curve is the training output error curve, the green curve is the actual output error curve, and the red curve is the test output error curve. Figure 4 It can be seen that when the test step length is 35, the set error can be achieved, which can meet the timeliness requirements of actual engineering projects.

[0099] The TMS series DSP has a very fast operation speed and a system structure and instructions suitable for digital signal processing. The DSP chip is TI's TMS320F2812, a high-performance floating-point digital signal processor, which adopts Harvard structure and has multiple adders and multipliers, 256KB on-chip memory and an operating frequency of 200MHz. Since the program implementation of the neural network involves a large number of matrix inner products and multiplication and addition links, it is difficult to implement it with ordinary single-chip microcomputers and the speed is very slow. Applying the above-mentioned AGV path planning algorithm program to the DSP hardware platform can give full play to the high-speed processing function of DSP and realize the online data processing of rotating machinery. Figure 7 The hardware circuit system is built by using the rich peripheral modules of TMS320F2812 for external signal processing and communication circuits to build a hardware platform. The hardware circuit mainly includes data acquisition module, digital input and output module, and CAN bus communication module. The specific control software can be written in CCS compiler software. See the schematic diagram of the control circuit structure of AGV Figure 7 ,In the actual application scenario, after completing the algorithm design, ,the control circuit of AGV also needs to be adaptively designed.

[0100] In actual application scenarios, when the AGV receives operation instructions and coordinate parameters, it obtains location information and surrounding environment information through its own sensors. Combined with the data sent by the dispatch center or the host computer, the AGV autonomously plans the pre-driving route according to the "shortest distance principle" and then starts to run. During the execution of the task by the AGV, it receives data from the dispatch center or the host computer in real time and obtains environmental information through sensors to obtain measurement data. After the nonlinear mapping of the neural network, the measurement data of each sensor is independent of each other. Then the data fused by each independent sensor is fused by the global neural network data at the same time. When there is noise, multiple sensors are used for identification, and then the results of the identification fusion of each sensor are fused to improve the recognition rate. At the same time, the neural network learning method is used to perceive and observe, so as to obtain a specific state feature representation. The neural network algorithm realizes the direct output control of the AGV car from the perception input of the environment to the action, helping the AGV to complete autonomous navigation and obstacle avoidance tasks. During the operation, the AGV continuously adjusts and plans the path to complete the task.

[0101] At the same time, this embodiment also provides a computer storage medium on which a computer program is stored. When the computer program is executed by a processor, the steps of the above-mentioned AGV path tracking and optimization method based on a DSP neural network algorithm are implemented.

[0102] In addition, this embodiment also provides a computer device, including a processor, a memory connected to the processor, and a computer program that can be run on the processor. When the processor executes the computer program, the steps of the above-mentioned AGV path tracking and optimization method based on the DSP neural network algorithm are implemented.

Claims

1. An AGV path tracking and optimization method based on a DSP-based neural network algorithm, characterized in that: The following steps are involved: Step 1, establish a mathematical model of the motion state of the AGV, obtain the motion state equation of the AGV in the two-dimensional plane at time t and the motion state equation of the AGV in the two-dimensional plane at time t+Δt, and calculate the observation equation Z of the AGV relative to the target point according to the motion state equation of the AGV in the two-dimensional plane at time t and the motion state equation of the AGV in the two-dimensional plane at time t+Δt; Step 2: Establish a mathematical model of the servo control system of the AGV and obtain the total transfer function G(s) of the servo control system; Step 3: According to the observation equation Z of the AGV relative to the target point and the total transfer function G(s) of the servo control system, a neural network is constructed and learned until the error reaches a preset requirement or the number of learning times reaches a preset requirement, and the neural network training result is obtained; Step 4: Call the neural network training result obtained in step 3 in the computer software, optimize the observation equation Z of the AGV relative to the target point and the total transfer function G(s) of the servo control system, and obtain the optimized observation equation Z1 of the AGV relative to the target point and the optimized total transfer function G1(s) of the servo control system. Then call the neural network training result obtained in step 3, perform simulation test according to Z1 and G1(s), and obtain the test result. If the test result does not meet the preset requirements, use Z1 and G1(s) to update Z and G(s), return to step 3, and rebuild the neural network for training; if the test result meets the preset requirements, the path planning of the AGV is completed.

2. The AGV path tracking and optimization method based on a DSP neural network algorithm according to claim 1, characterized in that: In step 1, the motion state equation of the AGV in the two-dimensional plane at time t is: Wherein, t is the time, x(t) and y(t) are the x-direction position and y-direction position of the AGV at time t, respectively, u(t) is the angle between the moving direction of the AGV at time t and the x-axis, v(t) is the speed of the AGV at time t, ω(t) is the angular velocity of the AGV at time t, and a(t) is the acceleration of the AGV at time t; d is the differential symbol; At time t+Δt, the motion state equation of AGV in the two-dimensional plane is: Among them, V max and V min They represent the maximum speed and minimum speed of the AGV between t and t+Δt respectively; At time t, the observation equation Z of the AGV relative to the target point is: Where D represents the distance between the current position of the AGV and the target point. is the azimuth of the AGV’s current position relative to the target point, x d and d They represent the x-direction position coordinate and y-direction position coordinate of the target point, respectively. t and t They represent the x-direction position coordinate and y-direction position coordinate of the AGV at time t respectively.

3. The AGV path tracking and optimization method based on a DSP neural network algorithm according to claim 2 is characterized in that: In step 2, the servo control system mathematical model includes a servo driver (22), a servo motor (23), a speed reduction device (24) and a controlled object (25) connected in sequence, as well as a servo controller (21), a current feedback module (28), a speed feedback module (27), a posture feedback module (26), a first operator and a second operator; the first input end of the first operator is connected to the input end of the AGV servo control system, and the output end is connected to the input end of the servo controller (21); the output end of the servo controller (21) is connected to the first input end of the second operator, and the current feedback module (28) is connected to the speed feedback module (27), the posture feedback module (26), the first operator and the second operator; the first input end of the first operator is connected to the input end of the AGV servo control system, and the output end of the first operator is connected to the input end of the servo controller (21); the output end of the servo controller (21) is connected to the first input end of the second operator, and the current feedback module (28) is connected to the speed feedback module (27). The input end of the flow feedback module (28) is connected between the servo motor (23) and the reduction device (24), and the output end is connected to the second input end of the second operator, and the output end of the second operator is connected to the input end of the servo driver (22); the input end of the speed feedback module (27) is arranged between the reduction device (24) and the controlled object (25), and the output end is connected to the second input end of the first operator, the output end of the controlled object (25) is connected to the input end of the posture feedback module (26), and the output end of the posture feedback module (26) is connected to the third input end of the first operator; The calculation formula of the total transfer function G(s) of the servo control system is: Among them, C(s) is the output function of the AGV servo control system, R(s) is the input function of the AGV servo control system; G1(s) and G2(s) are abstract functions of the servo controller (21) in the servo control system, G3(s) is the abstract function of the servo driver (22) in the servo control system, G4(s) is the abstract function of the servo motor (23) in the servo control system, and G5(s) is the abstract function of the deceleration device (24) in the servo control system; H2(s) is the abstract function of the speed feedback module (27) of the servo control system, and H3(s) is the abstract function of the current feedback module (28) in the servo control system.

4. The AGV path tracking and optimization method based on a DSP neural network algorithm according to claim 3 is characterized in that: Step 3 is as follows: Step 3.1, initialize the neural network, the neural network includes an input layer (31), a hidden layer (32) and an output layer (33) arranged in sequence, and set the input layer weight set {ω it }、Hidden layer weight set {ω ij } and the output layer weight set {ω ik }, and the input layer threshold set {θ t }、Hidden layer threshold set {θ j } and 00 output layer threshold set {θ k }, define the number of learning times N = 0, preset error e, and the maximum number of learning times N1; where i, j, t, and k all represent the position information of neurons in different layers, the first letter of the ω subscript represents the number of layers of the neural network, and the second letter of the ω subscript represents the sequence number of the neuron; each layer of the neural network has one and only one threshold, which represents the connection relationship between different layers; Step 3.2, calculate the input value and output value of the neurons in the hidden layer (32); the activation function of the hidden layer (32) is where y j is the output value of the hidden layer (32), x j is the input variable of the hidden layer (32), m is the number of records; Step 3.3: According to the observation equation Z of the AGV relative to the target point and the total transfer function G(s) of the servo control system, the input value and output value of the neuron in the output layer (33) are calculated; the activation function of the output layer (33) is expressed as where y k is the output value of the output layer (33), g k is the input variable of the output layer (33); f(x) = tanx (x≠(π / 2)+βπ, β is an integer); Step 3.4: Calculate the generalized error of each neuron in the hidden layer (32) and the output layer (33) and Then let N=N+1; and The specific calculation formula is: in, is the input variable of the neurons in the output layer (33), is the input variable of the neurons in the input layer (31), is the input variable of the neurons in the hidden layer (32); is the expected output value, is the actual output value; is the output error value, z is the data recording variable, ω zk ,ω zt ,ω zj and z All are input variables; a, b, and n are the number of records, and their values ​​are automatically taken by the system; Step 3.

5. Determine the value of N. If N ≥ N1, then execute Step 3.

7. If N < N1, then compare and with e respectively. If and are both less than e, then execute Step 3.

7. If and at least one is greater than or equal to e, then execute Step 3.6; Step 3.6: Using the data obtained in step 3.4 and For weight ω it ,ω ij and ω tk , and the threshold θ t ,θ j and θ k Update and then return to step 3.2; the specific calculation formula is: Where η and α are learning rates, 0<η<1, 0<α<1; p is the number of records, and its value is automatically taken by the system; Δω is the weight change value before and after two times, and △θ is the threshold change value before and after two times; and are the input variables of neurons in the hidden layer (32), output layer (33) and input layer (31) respectively; Step 3.7: Obtain the neural network training results, that is, the optimized input layer weight set {ω it }、Hidden layer weight set {ω ij }、Output layer weight set {ω tk }、The input layer threshold set {θ t }、Hidden layer threshold set {θ j } and the output layer threshold set {θ k }.

5. A computer storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the steps of the AGV path tracking and optimization method based on a DSP neural network algorithm as described in any one of claims 1 to 4 are implemented.

6. A computer device comprising a processor, a memory connected to the processor, and a computer program executable on the processor, characterized in that: When the processor executes the computer program, the steps of the AGV path tracking and optimization method based on the DSP neural network algorithm described in any one of claims 1-4 are implemented.

Images