TWI363546B - Real-time control system of dynamic petri recurrent-fuzzy-neural-network - Google Patents

Description

1363546 _ 101 January 101 revised replacement page. 214 Weight between network output layer and network rule layer. 215 Network output signal 8. If there is a chemical formula in this case, please reveal the chemical formula that best shows the characteristics of the invention: IX. Description of the invention: [Technical field of the invention] The technical field of the "dynamic dispatching recursive fuzzy neural network real-time control system" of the present invention mainly includes fuzzy control, neural network, and nonlinear control. And intelligent control; according to the above technology, the development of an instant control system, the real-time control system with the fuzzy neural network as the core, join the network sentiment layer and recursive structure, forming a dynamic sentimentary fuzzy neural network The road, and through the online parameter adjustment module and the learning rate update module, the controlled body achieves the purpose of the desired control. [Prior Art] In the field of traditional control, the accuracy of the dynamic mode of the control system is the most important factor affecting the performance of the control. The more detailed the information of the system dynamics, the more precise control can be achieved; however, for nonlinear systems, It is difficult to correctly describe the dynamics of the system, so researchers can easily use various methods to simplify the system dynamics to achieve control purposes, but not ideal. In other words, traditional control theory has strong and powerful control ability for clear systems, but it is powerless for systems that are too complicated or difficult to describe accurately. Therefore, many researchers have tried to deal with such control problems with fuzzy mathematics. 1363546 ιοί年01月13 Revision of the replacement page Since the development of fuzzy mathematics by Zadeh scholars, it has greatly contributed to the control of ambiguous systems. Since the Seventies, there have been some practical fuzzy controllers to complete, fuzzy control (FUZZy c〇ntr〇i) has the advantages of robustness, no need for precise mathematical models, strong approximation ability, and the use of human experience to establish fuzzy rules--[...] Reference [2] proposes to use the fuzzy map and the new map measurement to operate the self-propelled vehicle. Reference [3] developed an embedded fuzzy controller to control the self-propelled vehicle device' and proved its convergence through the Lyapun〇v stability theory. Reference [4] designed an instant fuzzy control architecture and used an infrared sensor to enable the self-propelled vehicle to achieve the desired function. Reference [5] uses fuzzy logic Month b is enough to follow the behavior of human thinking, enabling the self-propelled car to follow a specific path. Although these techniques can model the control system by emulating human behavior, it is quite difficult to achieve appropriate fuzzy rules to achieve good control performance. The Neural Network (NN) is a model that mimics the development of human brain activity. In terms of network architecture, it is made up of many simple and interconnected processing elements, ie, neurons (Neur〇ns); in terms of network functions, it is generated by biological models. New state information processing and calculation methods. In recent years, many people have studied the use of neural networks in the listening system or dynamic system control [6]_[8]. Reference [7] uses a neural network-like approach to study how to chase after a self-propelled vehicle (4) process towel. Reference [8] develops a simple ugly - neural network to manipulate the self-propelled car without the need to use the speed information of the self-propelled car. Although the * and the sacred network have powerful function approximation ability, the parameter value of the network usually must be 1363546. On January 13th, the replacement page must be subjected to long-term offline (Off-Line) training to achieve good. The initial network parameter values have not been trained to complete, resulting in network parameters that do not have corresponding values, so the initial control performance is generally poor. In recent years, reference [9] [11] proposed that the Recurrent Neural Network (RNN) achieves a fast correspondence capability, and its performance is significantly better than that of a neural network, but because the neural network is It consists of neurons, so in a more complex system, it is easy to cause the network to be too large.

Nowadays, combining fuzzy control and neural network has become a very popular research topic [12]-[14]. Fuzzy Neural Network (FNN) can combine the advantages of fuzzy control and neural network to achieve good performance. Its network architecture is simpler than that of neural networks. Corresponding fuzzy rules can use neural networks. Road learning theory is obtained. In addition, references [15], [16] propose a recurrent Fuzzy Neural Network (RFNN) architecture, because the recursive structure can speed up the network's corresponding ability, in general, recursive The fuzzy neural network control performance is superior to the fuzzy neural network. On the other hand, after the introduction of Petri Net (PN), many people began to study in different fields [π], [ι8], and reference [19] proposed the Paifu fuzzy neural network (petri). The architecture of Fuzzy Neural Network (PFNN) is applied to linear induction motors. The main concept is to add the mechanism of Pai Cui operation to the fuzzy neural network to reduce the computational complexity. Back to the structure, so the control performance is slightly inferior to the hand-like fuzzy neural network. 1363546 __ January 13, 101 revised replacement page Remarks: References [1] L. X. Wang, A Course in Fuzzy Systems and Control. New Jersey: Prentice-Hall, 1997.

[2] H. Seraji and A. Howard, “Behavior-based robot navigation on challenging terrain: a fuzzy logic approach, 5, IEEE Trans. Robot. Automat., vol. 18, no. 3, pp. 308-321, 2002.

[3] SX Yang, H. Li, MQH Meng, and PX Liu, uAn embedded fuzzy controller for a behavior-based mobile robot with guaranteed performance, 5, IEEE Trans. Fuzzy Syst., vol. 12, no. 4, pp 436-446, 2004.

[4] T. H. S. Li, S. J. Chang, and W. Tong, uFuzzy target tracking control of Lu autonomous mobile robots by using infrared sensors/5 IEEE Trans. Fuzzy

Syst., vol. 12, no. 4, pp. 491-501, 2004.

[5] G. Antonelli, S. Chiaverini, and G. Fusco, UA fuzzy-logic-based approach for mobile robot path tracking, 5, IEEE Trans. Fuzzy Syst., vol. 15, no. 2, pp. 211- 221, 2007.

[6] O. Omidvar, and D. L. Elliott, Neural Systems for Control. Academic Press, 1997.

[7] SX Yang and MQH Meng, “Real-time collision-free motion planning of a mobile robot using a neural dynamics-based approach,” /E五五7>vmy. Neural Netw., vol. 14, no. 6 , pp. 1541-1552, 2003.

[8] T. Das, IN Kar, and S. Chaudhury, “Simple neuron-based adaptive controller for a nonholonomic mobile robot including actuator dynamics,” Neurocomputing, vol. 69, no. 16-18, pp. 2140-2151, 2006.

[9] FJ Lin, RJ Wai, WD Chou and SP Hsu, “Adaptive backstepping control using recurrent neural network for linear induction motor drive, 5, IEEE Trans, lnd. Electron., vol. 49, no. 1, pp. 134 -146, 2002.

[10] RJ Wai, CM Lin and YF Peng, “Adaptive hybrid control for linear piezoelectric ceramic motor drive using diagonal recurrent CMAC network, IEEE Trans. Neural Netw., vol. 15, no. 6, pp. 1491-1506, 2004 8 1363546 Modified on January 13, 2011 [11] SJ Yoo, YH Choi, and JB Park, ''Generalized predictive control based on self-recurrent wavelet neural network for stable path tracking of mobile robots: adaptive learning rates approach , IEEE Trans. Circuit Syst., vol. 53, no. 6, pp. 1381-1394, 2006.

[12] C. T. Lin, and C. S. George Lee, Neural Fuzzy Systems, New Jersey: Prentice-Hall, 1996.

[13] FJ Lin, RJ Wai, and CC Lee, “Fuzzy neural network position controller for ultrasonic motor drive using push-pull DC-DC converter, IEE Proc. Control Theory Appl., vol. 146, no. 1, pp. 99- 107, 1999.

[14] C. Ye, NHC Yung, and D. Wang, fuzzy controller with supervised learning assisted reinforcement learning algorithm for obstacle avoidance,, 5 IEEE Trans. Syst., Man, Cybern. B, vol. 33, no. Pp. 17-27, 2003.

[15] FJ Lin, RJ Wai, KK Shyu and TM Liu, ^Recurrent fuzzy neural network control for piezoelectric ceramic linear ultrasonic motor drive,” IEEE Trans. Ultrason., Ferroelect., Freq. Cont., vol. 48, no. 4, pp. 900-914, 2001.

[16] R. J. Wai, “Total sliding-mode controller for PM synchronous servo motor drive using recurrent fuzzy neural network, 55 IEEE Trans. Ind. Electron., vol. 48, no. 5, pp. 926-944, 2001.

[17] R. David and H. Alla, “Petri nets for modeling of dynamic systems-A

Automatica, vol. 30, no. 2, pp. 175-202, 1994.

[18] V. R. L. Shen, ''Reinforcement learning for high-level fuzzy petri nets, 55 IEEE Trans. Syst., Man, Cybern. B, vol. 33, no. 2, pp. 351-362, 2003.

[19] RJ Wai and CC Chu, "Motion control of analog induction motor via petri fuzzy-neural-network, IEEE Trans. Ind. Electron., vol. 54, no. 1, pp. 281-295, 2007. SUMMARY OF THE INVENTION The block diagram of the dynamic dispatching fuzzy neural network real-time control system is shown in the first 1363546 _ 101 January 13, revised replacement page, which includes a dynamic paradigm fuzzy class. The neural network, the 101 controlled body, the 102-line parameter adjustment module, and the 103 learning rate update module, the input of the 100 dynamic dispatched fuzzy-type neural network is the error signal and its differential, and is obtained through network operation. Control signal output, 1) The number of inputs and outputs of the dynamic Pai Cui recursive fuzzy neural network is determined by the state of the controlled body and the control signal. Taking the first figure as an example, the input is 4 = ~-1 and the gas = meaning - less and the sum of the differentials, wherein x and less are the controlled body state and the corresponding control command, the output control signal is 〇, and 〇 2; The 101 controlled body is any device to be controlled, usually input as a control signal, and the output is measured by the sensor to measure the state of the controlled body, and the controlled body state is corresponding to the corresponding control command. The error signal is reduced to the error signal, and the differential signal is transmitted to the 100 dynamic dispatched fuzzy neural network to form an instant control system; the 102 online parameter adjustment module is based on the learning rate (/^, ~, ^, , ^^, error signal (1): ^^^ and its differential (Κ) to adjust the variation of the dynamic dynamic recursive fuzzy neural network parameters (△ Μ, Δ / γ, Δί /, Δα /); The 103 learning rate update module is based on a 100-dynamic dynamic recursive fuzzy neural network parameter («, «), error signal (α, \) and its differential ® (Κ), with discrete Rialpno ( Discrete-typeLyapunov) Stability theory to obtain the updated value of the learning rate. I. Dynamic Pai Cui recursive fuzzy neural network 100 dynamic The architecture of the Cui reversal fuzzy neural network is divided into five layers as shown in the "second picture", including 201 network input layer, 202 network attribution function layer, 208 network dispatch layer, 210 network. Rule layer and 213 network output layer, in addition to 202 network is 1363546 101 years in January ^ is a function layer added 204 network recursive structure, under each layer signal: means as ~ --- layer - 201 network The input layer transmits the network input signal to the 202 network attribution function layer. '~'···, ",) The second layer is the 202 network attribution function layer, and each 203 network belongs to the function element. The input is the upper-second 2〇6 network attribution function layer output multiplied by 2〇5 network construction weight value' and the current 200 network input signal can be expressed as (1), -

Where "represents discrete time, α丨 represents 2. 5 network recursive structure weight value.) —1) represents the last 206 network attribution function layer output, and factory 1 is 207 time delay unit.—Generally speaking, '(4) There are many kinds of attribution functions, including triangular functions and trapezoidal functions: inverted clock function: Gaussian function, etc., and the attribution function used in the present invention is a Shangss function, which can be expressed as (2) (10) people', —)_,"/[called(〇] = eXp[^(d (7) , , exp[] stands for exponential function (Exp〇nentiai Functi〇n), w/ and 彳0 = 1'", ", ;_/ = 1,~,~) represent the center point and width of the Gaussian function, respectively, 'the number of statement variables for each 200 network input signal. The second layer is the 208 network sendi layer, here In the first layer, according to the formula (3), it is judged whether the network attribution function layer round is transmitted to the fourth layer tj\^^tj{ri)\>dth l0j MHnetj(ri)]<dih (3) eight middle <; represents the preset threshold, (; / is 2〇9 transfer valve. The fourth layer is 210 network rule layer, the 211 network rule layer output is corresponding to 1363546 _ 101 January 13 Web product 208 to send output Chui layers replacement sheets, can be expressed as (4).%

A (4) Υ[^ΊμΙ[ηβί^)] , tj=\ =1 Ο , t/ = 0 where (A: = 1,.··, ~) represents 211 network rule layer output; < The weight value between the 212 network layer and the network rule layer is set to 1 to the total number of neurons in the 210 network rule layer. The fifth layer is the 213 network output layer. The sum of the 215 network output signals in this layer is the product of the 211 network rule layer output and the weight value between the 212 network layer and the network rule layer. (5) Formula y〇= Ew〇^ (5) k=\ where < represents the weight value between the 214 network output layer and the network rule layer; where (ο = 1,···, "σ" represents 215 Network output signal, ". The total number of neurons in the 213 network output layer. Second, the online parameter adjustment module 102 online parameter adjustment module, mainly by the Supervised Gradient Descent Method (Supervised Gradient Descent Method) to obtain the variation of the parameter value in the 100 dynamic dispatched fuzzy neural network, the 102 The amount of change in all parameter values of the online parameter adjustment module is generated by partial differentiation of the desired parameter by a specific energy function. The partial differential calculation uses the linkage rate. Since the order of partial differential is transmitted back to the internal nodes by the output of the network, it is also called the inverse transfer algorithm. In order to derive the amount of change in the value of the dynamic dynamic recursive fuzzy neural network, the energy function is defined as follows:

12 丄2 ~ X^2 + - yf + (xd ~ x)2 + (yd - y)2 ]

2 1 2 ^2+ey +K2+ey2) (6) In y, ~ and & are control commands, and nin are differential; - for controlled body state, Ί micro-knife' represents error signal and differential. Too 4 & ', 100 dynamic Pai Cui recursive fuzzy neural network input is less & and ie %=4); 100 dynamic Pai Cui recursive fuzzy neural network rounds for 01 and 02 (ie ~=2). The inverse transfer algorithm of the 102-line parameter adjustment module is described as follows: In the 213 network round-trip layer, the energy function can be expressed as δ for the partial differential of the 215 network output signal.

dE dE dex dx | dE dey dy [ dE dex dx dE dey dy dex dx dy0 dey dy dya ~dex dx dy0 dey dy^T dx dy . dx . dy ^T + ey^ + e^ + ey^r- (7) The amount of change in the weight value between the network output layer and the network rule layer can be expressed as Λ 〇 from a 7, long time 0 = % dE~ (5) 1 ^ 〇 \ l^rj nJA (8) (9) where I The learning rate of the weight value between the 214 network output layer and the network rule layer. 214 The weight value update between the network output layer and the network rule layer can be expressed as w°k (« +1) = w°k (η) + Aw°k («) The energy function is 211 network rule layer output bias Differential

13 1363546 On January 13, 101, the revised replacement page 208, the energy function for the eiy〇c, _) is slightly divided into ρΜί)=~ dE "dEdyl dnetj^rj) -H ^{netjiri) ) dnetjiri) ^ k , now = 1 (11) . Ο , ί / = 〇 w/, 彳 and 〇 change can be expressed as △ w /. dE dmj.

Vn, -Is

dE

Aaf ~Va dE dal

Vs % '9E dy0 f dnet.{rj.)^ dy^etjirj) A dmi J dE dya Ύ dnetj^r】), dy0 dnetjirl)^ dsf J ~ dE dyn ' f dnet^r^ = ^mPj· VsPj dy0 Dnety.) da{ 2(r/ - mj ) - from a ^^-1) (s!f (12) 2(r/ - mff ~W~ (13)

(Η) The discipline that represents the center point and width of the Gaussian function and the value of the network recursive structure weight. The center point and width of the Gaussian function and the update value of the 2〇5 network recursive structure weight value can be expressed as mj (« +1) = m/ (η) + Amj (η) 5/ {n + \) = sf ( η) + Asj(n)

(15) (16) +1) = α/(8) + △«/(«) Since the system contains non-green quantitative factors, the system sensitivity is caused by bees. (4). ,do. And (10) β can't be easily decided, although the smart discriminator gamempent Iden (4) can be used to estimate the sensitivity of the system, which requires a large amount of calculations (17). In order to overcome this problem, and increase the odds of online learning, the system of the present invention 14 1363546 January 13, 101 revised replacement page sensitivity is similar to the following symbolic function dx i ssgn, △ where J = sgn V dy \ = sgn .=sgn dx f 二sgn = sgn V dy ^ V = sgn —= sgn lAy〇, χ{ή) - χ{η -1) I (8)-凡〇-i) J y{ri) - y{n -1). (()_凡("_1) x{n) - x{n -1) , y〇{n)-y〇{n-\) jy{n)-y{n-\) Y〇{n)-y0{n-\)^ Λ (18) where sgn(·) represents a symbol function. Third, the learning rate update module selects different parameter learning rate has a significant impact on the performance of the network. In order to effectively train the parameter values of the 100 dynamic recursive fuzzy neural network, the present invention utilizes the discrete type. The Discrete-type Lyapunov stability theory finds the updated value of the learning rate so that the error signal converges. According to the formula (6), the discrete variation of the Discrete-type Lyapunov function is defined as AE(n)^E(n + \)-E(n) (19) so the energy function can be obtained by (8) And (12)-(14) are expressed as E{n +1) = Ε{ή) + AE(n) 15 1363546 0=1 k-\ ^f: Modified replacement page on January 13, 101

V,V't^(w)aj , dE(n) dE(n) A ;1Μ[^'+^'+-ύτΑα!] V +5(8) <E{n) + E(n) E{ n) 1 Vw 4 m 1 4 E(n) 1 4 E(n) 1 ”a 4 5(8) ··〇'y ΣΣ JH "ί n〇ΣΣΣ Hi <dE(n)dxi dyn dsf dE{n ) dy0 .^y〇dE(n) dxt dy0 dx, dy0dmi , 2 private dyQ dsj \2 (20) where Δ<, △<, M and respectively represent 214 between the network output layer and the network rule layer The amount of change in the weight value, the change in the center point and width of the Gaussian function, and the change in the weight value of the 205 network recursive structure; |.丨 indicates the absolute value. 100 Dynamic Pai Cui recursive fuzzy neural network learning The rate is designed as Ε{ή) % 4 ΘΕ(η) dyn ^ dy0 dw°k 〇=\ , 2 + ε (21) E(n) 4 "I __J ··〇ΣΣΣ i=lj=\ o=l

Vs dE{n) dxs dyn dXj dy0 dnijE{n) \2 + ε (22) 4 yyy dE(n) dxt dyn (23) + ε 1363546

Va E{n)

4 ΣΣΣ /=1 y=l o=l

dE(n) dxi dyo Y + ε as 丨dy〇 daf where 5 is a positive constant. According to the (21)-(24) learning rate design, let 1 ff rdE(n) dy0 Y 4 E(n)tif^ l ^y〇^1) 4 Ε{ή) ,=1 j=\ 〇=ι dx . dya dmj

V (24) (25)

4 five (8) combined with L \2 <1 1_-^ yyyi^(^)5y, dyn 4 [d^^Td^l thus can get five (" + 1) < five (8), then according to (6) Knowing five (8) > 〇 and 仏 (8) < 〇, so the error signal will gradually converge. [Embodiment] An embodiment of the dynamic dispatching fuzzy-type neural network real-time control system of the present invention is a tracking automatic control system used in a self-propelled vehicle, and a schematic diagram of a self-propelled vehicle device is used as a third As shown in the figure, the two steering wheels and the _ branch wheel are disposed on the vehicle body, and the steering wheels are respectively controlled by two independent DC motors and are horizontally mounted on the axle. The branch wheel is a passive freewheel that can be controlled at any degree with the steering wheel. In the figure, 26 is the distance between the two steering wheels, and the diameter of the steering wheel is dry. In the figure, point c is the centroid position of the self-propelled vehicle; the point p is the axle and the material: the vertical line of the punching point After the intersection, the P point indicates the position of the self-propelled vehicle in the coordinate system; in the figure, 17 1363546 ___ January 13, 101 revised replacement page {0, U, V} is the global coordinate system, the self-propelled vehicle in the global coordinate system The position can be expressed as a corpse =[w vf, where ί/ and V represent the horizontal and vertical axes in the global coordinates, respectively; {P, X, Y} in the figure is the local coordinate system, that is, the point P is the origin The coordinate system; in the figure, the relative angle between the global coordinates and the local coordinates, and the starting angle is measured by the U axis. Assuming that the wheel of the self-propelled vehicle only rotates and does not produce a side shift, that is, the direction in which the self-propelled vehicle moves is perpendicular to the axle, the action constraint of the self-propelled vehicle can be expressed as vcos^-«sin^ = 0 (26 The schematic diagram of the self-propelled vehicle movement is shown in the “fourth figure”. In the figure, Q and Q· represent the Lu position of the self-propelled vehicle from the sampling time % to % +1; △« and Δν represent the U-axis and V in the global coordinates. The amount of displacement of the shaft; ΔΘ represents the amount of rotation of the self-propelled vehicle angle; r, r. And & represents the radius of rotation of the Q point to the Q' point, which is the distance between the origin 0 and the left wheel; C is the distance between the origin point Q to the Q point; (between the origin 0 to the right wheel) The distance; and $, especially the element is the arc length of the corresponding radius of rotation; ΑΘ and r can be expressed as △0= taste (Vr_v/) (27) V, + V, L r〇= b (28) Δί represents the interval of sampling time; V/ and v/ represent the left wheel speed and the right wheel speed respectively, and the maximum limit is set to vmax; in the global coordinate system, the discrete dynamic equation of the self-propelled vehicle can be expressed as u{ns +1 ) = u(ns) + ^{sinf^^) + ^θ(η5)] - sin^^)} +1) = ν(«5) + ro{cos0(ns) - cosf^C^) + Δ ^(«5)]} (29) e(ns+\) = e(ns) + A0(ns) where % represents the sampling time. If the self-propelled car is in the global coordinate control position

18 1363546 ------- vdf Q01 January 13 曰修 ° again. When the controller is ten, the global coordinate Ta[c〇^ sin^l _ audition matrix-sin0 — COS0 V c〇s6>sin6>' ( -3⁄4. _sin6> ~cos0 (30) must be able to achieve the instant spur疋 Find the appropriate control signal 'to make the self-propelled vehicle longitudinal control, in order to achieve this control target = vertical error one..., its ... coordinates, then get the appropriate revolver speed (1) and right wheel speed through the control law Manipulating the self-propelled car to achieve the purpose of miscellaneous tracking control. 丨能,二=? Self-propelled car in different reference paths (control commands) to achieve tracking effect and self-propelled car dynamic sective recursive fuzzy neural network, the combination 20 The network sentiment layer and the 204 network recursive structure are in the traditional fuzzy neural network. The real-time control paper block diagram is shown in the “fifth figure”, which contains the coordinate converter self-propelled car dynamics. Pai Cui recursive fuzzy neural network, 5〇2 self-propelled car, 503 self-propelled car line parameter adjustment module and 5 () 4 self-propelled car learning rate update model. The 500 coordinate converter will be global The coordinates are converted to local coordinates; According to the invention of the invention by the network (10), the dynamic dispatching type of the two-paste type of nerves is turned, and "the chasing _ 静静印 and its good (u) find the suitable revolving speed (V,) and the right wheel speed (vj Manipulating the 5〇2 self-propelled vehicle device; the $ self-propelled vehicle device moves the self-propelled vehicle according to the left and right wheel speeds, and represents the posture of the self-propelled vehicle with the global coordinates and the vehicle orientation; the 5〇3 self-propelled vehicle line parameters The adjustment module is used to adjust the replacement page of the 1st and 2nd line parameter adjustment module of the month of January 1st, which is adjusted according to the learning rate, the tracking error signal (e〆), and its differential. 5〇1 Self-propelled car dynamics, the number of changes in the parameters of the neural network parameters (△<, _, the resistance to the class self-propelled car, the rate of the update module using the invention < 1〇3 learning The rate update module is based on 5(1) self-propelled vehicle movement I, Pai Cui's welcoming plate-like neural network parameters («f, X), tracking error signal y > and its differentiation (required update rate of learning rate) In order to verify the instant control performance of the present invention, a conventional network architecture control system is also proposed to compare the performance of the self-propelled vehicle. Can represent & φ -〇.〇 925m; 6 = 〇.l67m; Vmax=〇.4m/s (31) In order to show the dynamics of the 501 self-propelled car, the Cui recursive fuzzy neural network has superior effect. Thus, compare three other different network structures, including fuzzy neural network (FNN), recursive fuzzy neural network (rfnn), and Paifu fuzzy neural network (PFNN)' and use the same 5 〇3 self-propelled car line parameter adjustment module, 5〇4 self-propelled car S rate update module, input signal and output signal. The input signal is tracking error signal and its differential (<,€); output signal is revolver Speed (V,) and right wheel speed ® degrees (〇. The initial value of the network parameter is the previously trained value. The previously trained parameter value adopts the previously introduced 503 self-propelled vehicle line parameter adjustment module and the 5〇4 self-propelled vehicle learning rate update module, when it reaches satisfactory. Control the value of the performance, and then set the parameter value of this time to the initial value of the next execution. The software used in this experiment is Visual C++, written on the personal computer of the Pentium IV. The model of the self-propelled car is Pioneer, which is manufactured by MobileRobots. 1363546 _ 101 January 101 revised replacement page • Development board for Hitachi H8S; frequency 44.2368MHz; 32bitRISC; 32k RAM; . 128k FLASH; self-propelled car and computer connection using wireless network transmission module; wheel by 12 The volt DC motor control uses PWM signals; each motor is equipped with a 128count/mm sensor for position feedback. This experiment selects two reference paths (control commands) to test the performance of the control system. One is the 8-shaped trajectory, which is expressed as follows:

Xd = , 2πί π, cos( π Τ 2' • π 2πί. cost . 2 Τ , 2πί π' cos( τ Τ 2. .π 2πί. cos(--- 2 Τ ,0<,<40 ,40<;ί<80,80<ί<120,120</<160 yd 1 · /2^ 1 + sin(- τ --),0</<40 2 3 + sin(--2 Ύττί ) ,40<ί<80 Τ -.2πί 3 + sin(- τ --),80</<120 2 1 + sin(--2 Οττΐ ),120<ί<160 Τ (32) the other is square The manifestation is as follows: xd ΟΛί ,0<?<10 l + cos[2^-1〇>-Τ 昏], 10<ί <30 2 ,30<ί <50 l + Cos[2;r(i-3〇)-Τ camp] ,50<ί <70 1-0.1(ί-70) ,70 < ί < 90 . Γ2π(ί-50) -l + cos[ ———-- Τ -f] ,90</<110 -2 ,110<ί<130 . Γ2π{ί + \0) -1 + cos "----- Τ -f] ,130<ί< 150 -1 + 0.1(ί-110) ,150<ί<160 21 1363546 Modified on January 13, 101, replace page 1 + sin[ 2π(ί-10) ~~Τ 1 + 0.1(/ - 30) 3 + Sin[-神-3〇) 4 Τ 3 + sin[ 2π(ί - 50) ~~Τ 3-0.1(ί-110) 1+ 2^+10) Τ ,0</<10 -],10&lt ;ί<30 2 ,30 < ί < 50 -],50<ί <70 2 ,70<ί <90 -1 ,90</<110 2 ,110</<130 Fen],130<,<150 (33) Ο ,150<ί<160 Self-propelled car The initial position and angle are preset to zero, and the control parameters of the system are expressed as follows: 4=0.1 ; £= 0.05; Δί = 0.05 s (34) The control parameters in (34) are considered in the possible operating environment. A set of parameter values for better performance. In order to be able to compare the performance of each network structure control system, here

The mean error value (MSE) is

MSE = 1^V*)+^ 1 n=\ where: r represents the sum of the sampling times. This experiment firstly produces different instantaneous control performances for fuzzy neural networks (FNN) with different numbers of neurons in the attribution function layer. The average error value (MSE) and execution time are shown in the sixth graph. According to the results of the sixth graph, it is found that the neurons perform best at seven times. In other words, this network is lower than the other when %=4, '.=7, '=7x7x7x7 = 2401 and %=2. The network size is more suitable for self-propelled vehicle path tracking, so the next experiment is based on this network, the control system adjustment parameters

22 1363546 】 〇 】 January 13 曰 revised replacement page 'The total number of total 2 ^ « / + « production ' = 4858. However, the number of parameter adjustments of the 模糊 fuzzy state neural network is based on the preset threshold value AM of (3), because the memory and execution ambiguity required by A are higher than The fuzzy neural network (FNN) control system has less to come, although it requires more rules to barely track the response, and does not cause a huge amount of computation, nor does it cause the computer or microprocessor to crash. The experimental results of all control systems are shown in Figure 7 to Figure 14, including 8-word type and square path tracking, in which Figures 7, 9, 11 and 13 are 8-shaped tracking responses; , ten, twelve and fourteen are square tracking responses. In each figure, (a) and (b) represent the tracking response of the X-axis and the γ-axis, respectively; (c) and (d) represent the tracking error of the 乂 and Y axes, respectively, and (e) and (7) represent the left wheel speed and the right, respectively. Wheel speed; (8) is the path t tracking response, (h) is the path tracking error ^ is known from the experimental results, the biggest factor affecting the tracking response and robust performance of the immediate control system is the learning ability of the control system itself; (4) execution time It is the network (four) structure ^ fifteenth map to organize the average error value (MSE) and execution time of all Lu network structure control systems, the control efficiency of the recursive fuzzy God I network (RFNN) is significantly better than the fuzzy type of neural The network (F_ is good because the recursive structure of the recursive fuzzy neural network (rfnn) increases the corresponding monthly power of the network'. Although the Peifu fuzzy neural network (the average error value of pF_ is slightly better) The back-type _-like nerve (RFNN) is large, but the execution time is much less. Because of the preset threshold value limitation in the U-fuzzy H-network (PFNN), the output of the low-level lining function layer is too low. The corresponding change amount does not perform the operation; compare the figure thirteen spear 14 To - '501 self-propelled car dynamic Pai Cui recursive fuzzy neural network 23 01 101 January 101 revised the average error value of the replacement page (MSE) than the recursive fuzzy neural network (RFNN) The small 'its. execution time is close to the Paifu fuzzy-like neural network (pFNN), and as a result, it is known that the performance of the 5〇1 self-propelled car dynamic regenerative fuzzy neural network is indeed better than other network structures. The control system is superior. This embodiment successfully combines the 208 network Pai layer and the 204 network recursive structure into the traditional fuzzy neural network, and demonstrates the self-propelled dynamic dispatching mode.

The real-time control performance of the paste god, ^ network is on the path tracking. 1(9) Dynamic Pai Cui recursive fuzzy celestial network can simplify the calculation of the number of rules and increase the corresponding ability of the network. It is known from the experimental results that the 5〇1 self-propelled car dynamics are sent back to the fuzzy neural network. The performance is 23.42 higher than the fuzzy neural network j^rpxTXT, a _ A, &, and the road (FNN). /. Its execution time was also reduced by 95.65〇/〇. The main advantages of this month's dynamic dispatching fuzzy back-type fuzzy neural network real-time control system are as follows: The monthly dynamic dispatching fuzzy back-type fuzzy neural network real-time control system proposes a new dynamic jade recursive The __ neural network structure, and the addition of the parameter adjustment module __path parameters learning, and (9) the learning rate update module allows the error signal to be narrated, the invention is quite novel. 2. The invention (4) Pai Cui recursive fuzzy _ _ road real-time control system proposed network structure into a force to transport the self-propelled vehicle device, the experimental results can prove that the performance of the network than the previous network ' The control performance is superior, so the present invention is obviously progressive. & 孜 24 1363546 _ 101 January 101 revised replacement page - 3. The instant control system of the "dynamic dispatching fuzzy back-type fuzzy neural network real-time control system" of the present invention can be applied to any 101 The control body can also add 102 online parameter adjustment module and 103 learning rate update module to achieve the required immediate control performance, so the invention is quite industrially useful. The present invention has been disclosed in the foregoing preferred embodiments, and is not intended to limit the scope of the invention, and the invention may be modified and modified without departing from the spirit and scope of the invention. The scope is subject to the definition of the scope of the patent application. BRIEF DESCRIPTION OF THE DRAWINGS The first figure shows the dynamic control system of the present invention. The second diagram shows the architecture diagram of the dynamic dispatching fuzzy-type neural network of the present invention. The three diagrams show the schematic diagram of the self-propelled vehicle. The fourth diagram shows the movement of the self-propelled vehicle. The fifth diagram shows the block diagram of the self-propelled vehicle. The sixth diagram shows the average error value and execution time of the fuzzy neural network control system. The seven graphs represent the experimental results of the 8-shaped path tracking fuzzy neural network control system. The eighth figure shows the square path tracking fuzzy neural network control system. 25 1363546 _ 101 January 101 revised replacement page Results The ninth graph shows the experimental results of the 8· font path tracking and returning fuzzy-like neural network control system. The tenth graph shows the results of the square path tracking and harvesting back-type fuzzy neural network control system. The figure shows the experimental results of the 8-shaped path tracking and the acquisition of the Cui fuzzy-like neural network control system. The twelfth figure shows the square path tracking. The experimental results of the network control system β are shown in the thirteenth figure. The experimental results of the 8-shaped path tracking mining dynamic dispatching fuzzy-type neural network control system show that the square path tracking mining dynamics The experimental results of the fuzzy neural network control system, the fifteenth figure shows the fuzzy neural network, the recursive fuzzy neural network, the pie

The average error value and execution time of the C-fuzzy-like neural network and the dynamic Pai-Cui retro-type fuzzy neural network I-channel control system [Key component symbol description] 100 Dynamic Pai Cui recursive fuzzy neural network · 101 Controlled Body 102 online parameter adjustment module 103 learning rate update module

26 1363546 101 January 101 曰Revision replacement page 200 Network input signal 201 Network input layer 202 Network attribution function layer 203 Network attribution function Neuron 204 Network recursive structure 205 Network recursive structure weight value 206 Network Road attribution function layer output

207 Time Delay Unit 208 Network Pai Layer 209 Transfer Valve 210 Network Rule Layer 211 Network Rule Layer Output 212 Weight Value Between Network Pai Layer and Network Rule Layer 213 Network Output Layer

214 Network output layer and network rule layer weight value 215 Network output signal 500 coordinate converter 501 Self-propelled vehicle dynamic Pai Cui recursive fuzzy neural network 502 Self-propelled vehicle device 503 Self-propelled car line parameter adjustment Module 504 self-propelled vehicle learning rate update module 27

Claims (1)

1363546 ___ Revised replacement page on January 13, 101. Patent application scope: 1. A dynamic dispatching fuzzy neural network real-time control system, which includes: a dynamic dispatched fuzzy-type neural network a controlled body; an on-line parameter adjustment module; and a learning rate update module; the dynamic senti-return type fuzzy neural network input is an error signal and its differential, and the output is a control signal; The control body is a device to be controlled, wherein the input of the controlled body is the control signal, and the output is the state of the controlled body measured by a sensor, and the controlled body state is corresponding to the corresponding control command. Subtracting the error signal, transmitting the error signal and its differential to the dynamic dispatching fuzzy neural network to form an instant control system; the online parameter adjustment module is based on the learning rate, the error signal and its differential To adjust the variation of the dynamic dispatching fuzzy-type neural network parameters; the learning rate update module is based on the dynamic dispatching fuzzy-like neural network parameters, the error signal and its micro To discrete Calabria Puno # (Discrete-type Lyapunov) stability theory to obtain an updated value of the learning rate, such that the error signal is converged. 2. For example, in the dynamic dispatching fuzzy-type neural network real-time control system of claim 1, the dynamic dispatching fuzzy-like neural network includes a network input layer and a network attribution function layer. a network dispatch layer, a network rule layer and a network output layer; the network input layer directly transmits a network input signal to the network attribution function layer; the network attribution function layer operates the network The network input layer sent by the input layer 28 1363.546 _ 101 January 101 revised replacement page - the road input signal, the result is transmitted to the network Pai-Layer via a specific attribution function operation; the network sends a layer to transmit The mechanism of the valve determines whether the data is transmitted to the network rule layer; the network rule layer outputs the product of the output of the corresponding network layer; the network output layer outputs a network output signal Each of the network output signals is the sum of the network rule layer output and the weight value product between the network layer and the network rule layer. 3. For example, in the dynamic dispatching fuzzy-type neural network instant control system of claim 2, the network attribution function layer comprises a network attribution function neuron and a network recursive structure; The input of the path-attributed function neuron is the last network attribution function layer output multiplied by the network recursive structure weight value, and the current network input signal is added, and the result is output through a specific attribution function; the network is returned. The structure uses the network attribution function layer output as the information of the next operation in a time delay manner to achieve the capability of quickly and dynamically corresponding to the network. 4. For example, the dynamic dispatching fuzzy-type neural network real-time control system of the second patent application scope is a mechanism for transmitting a valve, which can determine the output value of the network attribution function layer. Whether it is higher than the preset threshold, prohibiting the output of the network attribute function layer that is too low, the output of the failed network attribution function layer does not need to adjust the parameters of the attribution function layer, so as to effectively reduce the network operation amount. 5. For example, the dynamic dispatching fuzzy-type neural network real-time control system of the patent application scope item 1 is mainly obtained by the Supervised Gradient Descent Method. Recursive mode 29 1363546 _, January 13, 101 Corrected the amount of change in the parameter value in the replacement page-like neural network. The amount of change in all parameter values of the parameter adjustment module on the line is determined by a specific energy function. The parameter is differentiated and generated, and the parameter value of the dynamic sentimentary fuzzy neural network is changed to achieve the learning function, so that the instant control system has better control performance. 6. A dynamic dispatching fuzzy-like neural network real-time control system, comprising: a dynamic dispatching fuzzy-like neural network; a controlled body; and an on-line parameter adjustment module; The dynamic dispatched fuzzy-type neural network input is an error signal and its differential, and the output is a control signal; the controlled body is a device to be controlled, wherein the input of the controlled body is the control signal, and the output is Measuring the state of the controlled body for a sensor, the controlled body state is subtracted from the corresponding control command to become the error signal, and the error signal and its differential are transmitted to the dynamic dispatching fuzzy The neural network forms an instant control system; the online parameter adjustment module adjusts the amount of change of the dynamic dispatching fuzzy genre @ 网路 网路 根据 according to the fixed learning rate, the error signal and its differential. 7. For example, the dynamic dispatching fuzzy-type neural network real-time control system of the sixth patent application scope includes a network input layer and a network attribution function layer. a network dispatch layer, a network rule layer and a network output layer; the network input layer directly transmits a network input signal to the network attribution function layer; the network attribution function layer operates the network The network input layer sent by the input layer 30, 101, January, 3, revised, replaces the page input signal, and transmits the result to the network Pai layer through a specific attribution function operation; the network sends a valve mechanism to transmit the valve Determining whether data is transmitted to the network rule layer; the network rule layer output is a product of the corresponding output of the network send layer; the output layer of the network is a network output signal, each of which The network output signal is the sum of the network rule layer output and the product of the weight value between the network layer and the network rule layer. 8. For example, the dynamic dispatching fuzzy-type neural network real-time control system of claim 7 includes a network-attributed function neuron and a network recursive structure; the network The input of the belonging function neuron is the last network attribution function layer output multiplied by the network recursive structure weight value, and the current network input signal is added, and the result is output through a specific attribution function; the network recursive structure The network attribution function layer output is used as the information of the next operation in a time delay manner to achieve the capability of quickly and dynamically corresponding to the network. 9. For example, the dynamic dispatching fuzzy-type neural network real-time control system of the seventh application patent scope is a mechanism for transmitting a valve, which can determine the output value of the network attribution function layer. Whether it is higher than the preset threshold, prohibiting the output of the network attribute function layer that is too low, the output of the failed network attribution function layer does not need to adjust the parameters of the attribution function layer, so as to effectively reduce the network operation amount. 10. For the dynamic dispatch control fuzzy neural network real-time control system of the sixth application patent scope, the online parameter adjustment module mainly obtains the dynamic sentiment by the Supervised Gradient Descent Method. Recursive mode 1363546 101 January 101 曰 Corrected the change of the parameter value in the page-like neural network. The variation of all parameter values of the parameter adjustment module on the line is biased by the specific energy function. The differential control is generated by changing the parameter values in the dynamic Pai recursive fuzzy neural network to achieve the learning function, so that the instant control system has better control performance. 32