JPH04364304A - Controller for electric vehicle - Google Patents

Abstract

PURPOSE:To vary the torque continuously, to evaluate the status quantity of an driving wheels totally and to take full advantage of adhesion through combination of fuzzy control method and information concerning to adhesion. CONSTITUTION:A speed detector PG, an effective current detecting means 102 and a torque current detecting means 602 detect and operate status quantities of all driving wheels and then determine a correction pattern of torque current for each driving wheel based on thus operated status quantity. Thus determined correction patterns are subjected to fuzzy synthesization in fuzzy inference blocks 8A-8D and further subjected to non-fuzzy processing thus determining a definite correction pattern. Inverter control sections in the downstream of an adding point 105 evaluate the adhesion state of each driving wheel totally and controls the motor torque such that a composition of trains has maximum total adhesion at all times. When idling develops abruptly, slip frequency command is squeezed directly to perform readhesion quickly.

Description

[Detailed description of the invention]

0001

[Industrial Application Field] The present invention is an electric vehicle control device with improved adhesive performance for an electric vehicle that controls a plurality of induction motors using a variable voltage variable frequency inverter (hereinafter referred to as a VVVF inverter), a so-called VVVF inverter vehicle. It is related to.

[0002] Regarding the adhesion control method that uses fuzzy control to control the electric motor torque so as to always obtain the maximum adhesion force, the present applicant's earlier application, Patent Application No. 2-328, describes
It was proposed in issue 26. It was a control system in which the slip rate differential, adhesive force differential, and creep speed were used as antecedents to modify the current pattern proportional to the motor torque. As an extension of this, the slip rate differential for each driving wheel,
On January 7, 1991, a method was proposed in which the torque current correction pattern was determined by computing the adhesive force differential and creep velocity, performing fuzzy inference for each driving wheel, max-synthesizing them, and defuzzifying them. It was proposed in the ``Electric Vehicle Control Device'' filed in Japan. This makes it possible to make maximum use of the adhesion of an entire train.

However, as will be explained later, since there is a delay in the current control system of VVVF inverter trains, it is said that simply modifying the current pattern cannot sufficiently suppress the occurrence of slippage due to a sudden decline in adhesive properties. There were drawbacks.

In order to eliminate this drawback, the present invention configures a control system that not only corrects the current pattern but also directly controls the slip frequency that can rapidly increase or decrease the motor torque. It also has the function of suppressing the growth of idling.

[0005]

2. Description of the Related Art A method shown in FIG. 10 is known as a method for improving the adhesive performance of electric vehicles. FIG. 10 shows a control circuit of a VVVF inverter system employing a typical conventional readhesion control method.

[0006] 1 is an electric car and represents one vehicle. 101 is an inverter device. 2A, 2B, 2
C and 2D (hereinafter referred to as 2A to 2D) are induction motors, and 3A to 3D are driving wheels indicated by letters corresponding to the induction motors 2A to 2D, respectively. 4A to 4D are motor rotation speed detectors similarly indicated by corresponding alphabetical letters, and in this example, they are composed of a pulse generator (PG), which generates a pulse train with a frequency proportional to the rotation speed of the electric motor. . Further, 5A to 5D are motor rotational frequency calculation means indicated by corresponding alphabetical letters, which convert each PG signal into rotational frequencies fMA to fMD of each electric motor. Furthermore, 7
A~7D are differentiators, and each motor rotation frequency fMA~
With fMD as input, its time derivative is dfMA/
It outputs dt~dfMD/dt. 108
is a motor frequency selection means, and the same inverter 101
This is to determine the motor frequency that becomes the reference for frequency control of the motor group driven by the motor group. Algorithms for selecting the reference motor frequency include a method of fixing it to a predetermined motor frequency (a method of determining the motor frequency from the PG signal of the shaft that is least likely to idle during power running or braking), and a method of dynamically switching it. (method of setting the frequency of each motor to the minimum value during power running, and setting the frequency of each motor to the maximum value during braking).

The output of the motor frequency selection means 108 is a reference motor frequency fM, which is added to a slip frequency command fSS, which will be described later, at an addition point 109 to obtain an inverter frequency command fI. In addition, in order to supply alternating current with a constant voltage/frequency ratio to the motor group, the voltage /
The value multiplied by the frequency ratio is set as the inverter output voltage value, the value is divided by the filter capacitor voltage detection value VC by the divider 111, and the value further passed through the first-order lag filter 112 is set as the transformation ratio command α. First order lag filter 112
is provided for stabilization.

The inverter 101 determines the switching timing by finding the intersection of a carrier having a switching frequency suitable for the inverter frequency command fI and a three-phase sine wave signal having an amplitude corresponding to the transformation ratio command α. , which performs pulse width modulation and applies three-phase alternating current to each motor.

The torque current pattern generating means 104 is
A current pattern Ip proportional to the motor torque corresponding to the required acceleration is generated based on a notch command from the driver for acceleration/braking and a signal from the load adjustment device that detects the load on the vehicle. Hereinafter, the effective current proportional to the motor torque will be referred to as torque current. The deviation between the torque current pattern Ip and the detected torque current value II is calculated at the addition point 105, and the value multiplied by a certain gain G0 by the gain multiplier 116 is input to the current control means 107 via the switching means 106. The output of this current control means 107 becomes the slip frequency command fSS. The current control means 107 generally employs a complete integral system, and this example also follows this.

Reference numeral 113 denotes a slip detecting means, and the slip detecting means 113 detects a comparator means 114 when the maximum value of the motor frequency differentials dfMA/dt to dfMD/dt, which are the outputs of the differentiators 7A to 7D, exceeds a predetermined value. to drive. This comparator 114 is an on/off signal generating means with a delay of a fixed time T, and when the state of the comparator changes and is maintained for a fixed time, the slip detection signal SLIP is turned on. Also,
When readhesion control is performed and the maximum value of the motor frequency differential falls below the above-mentioned predetermined value and this state is maintained for a certain period of time T, the slip detection signal SLIP is returned to the OFF state. Here, the delay time T is provided to avoid false detection due to noise mixed into the motor frequency differential.

Reference numeral 115 denotes a coefficient generator that stores a slip frequency throttling amount represented by KS, and this slip frequency throttling amount is normally set to a negative fixed value. Switching means 1
06 is when the idle detection signal SLIP is on,
Selecting the output of the coefficient generator 115 containing the slip frequency throttling amount KS as the input of the current control means 107;
When the slip detection signal SLIP is in the OFF state, the deviation between the torque current pattern Ip and the detected torque current value II is selected. Therefore, the slip frequency command fss can be expressed by the formula: (1) When SLIP=ON, fSS=KS ∫dt+fSS0 (2) When SLIP=OFF, fSS=G0 ∫(Ip -II)dt+fSS0
becomes. Here, fSS0 represents the value of the slip frequency command fSS at the time when the slip detection signal SLIP switches from the off state to the on state or from the on state to the off state. Further, G0 is the gain of the gain multiplier 116 in FIG. 10, and KS represents the amount of reduction of the slip frequency when KS < 0 as described above.

Correspondingly, the aforementioned torque current pattern generating means 104 converts the torque current pattern Ip into the torque current detection value I when the slip detection signal SLIP is in the ON state.
It is also provided with a function of increasing by a predetermined increment KP when the idle detection signal SLIP is set to I and the slip detection signal SLIP is in the off state. That is, (1) When SLIP=ON, Ip = II (2) When SLIP=OFF, Ip = min(KP ∫dt+Ip0, Ip MAX
). Here, Ip0 represents the value of the torque current pattern Ip at the time when the slip detection signal SLIP is switched from the on state to the off state. Further, KP is a constant representing the increment of the torque current pattern, and Ip MAX represents the current limit value of the torque current.

Reference numeral 102 denotes an effective current detection means, which detects each phase current of the motor group using a current sensor, and detects the effective current I.
This is to calculate M. Further, 103 is a torque current calculating means, which calculates the torque current II from the effective current value IM which is the output of the effective current detecting means 102 and the output voltage phase θ of the inverter.

[0014]

[Problems to be Solved by the Invention] The conventional adhesion control system as described above does not work unless slipping or skidding actually occurs, exceeds a predetermined value, and elapses for a predetermined period of time. The correction amount also uses a preset fixed value, and does not depend on the degree of idling. In addition, the method of correcting the motor torque is to repeat decrease and increase, and the motor torque can be adjusted continuously.
It cannot be said that the characteristics of the VVF inverter vehicle are fully utilized. Furthermore, as a method of detecting slippage, if at least one of multiple electric motor frequencies exceeds a predetermined value for a certain period of time, it is considered that slipping has occurred, and to what extent is the slipping occurring in other driving wheels other than the one determined to have occurred? The system does not take into consideration whether or not the adhesive strength is obtained. Therefore, there are two cases in which one driving wheel is idling due to dirt on the rail surface (hereinafter referred to as uniaxial idling), and a case in which all the driving wheels are idling at the same time due to a decrease in adhesion due to rain, etc. (hereinafter referred to as single-axis idling). The amount of throttling of the electric motor torque is the same in the case of shaft idling (referred to as shaft idling), and it is easily expected that the torque will be throttled too much in the former situation.

[0015] Adhesive force is a physical phenomenon determined by the condition of the contact surface between the wheel and the rail, and strictly speaking, it must be considered that it differs for each driving wheel. Therefore, the ideal adhesion control system is one that can independently control the driving force around the driving wheels so that the maximum adhesion force can be obtained depending on the adhesion state of each driving wheel. Some electric locomotives using a VVVF inverter are provided with an inverter device for each induction motor of each driving wheel so that the driving force around the driving wheels of each driving wheel can be controlled independently. However, in VVVF inverter trains that adopt a distributed power system, due to implementation constraints or cost constraints of the inverter device, a method is generally adopted in which four to eight induction motors are controlled by one inverter device. It has become a target. Therefore, adhesion control in a VVVF inverter electric train should be regarded as a multipurpose control system that controls the adhesion force of the entire train set to be maximized as much as possible according to the adhesion state of each driving wheel.

[0016] The present invention detects and calculates the state quantities of all the driving wheels, determines a torque current correction pattern for each drive wheel based on the state quantities, fuzzy-synthesizes them, and further performs defuzzification processing. It is possible to comprehensively evaluate the adhesion state of each driving wheel and control the electric motor torque so that the total adhesion force for the entire train set is always the maximum. It is something to do. In addition, if the slipping increases rapidly, the slip frequency command can be directly narrowed down.
Allows rapid re-adhesion.

Now, FIG. 11 shows the relationship between the sliding rate and adhesive force (hereinafter referred to as adhesive property) between the driving wheels and the rail. Here, the slip rate is a value obtained by dividing the difference speed between the driving wheel circumferential speed and the ground vehicle speed by the driving wheel circumferential speed, and is defined by the following equation. λ=(VM-VO)/VM

(1) In the above formula, VO is ground vehicle speed, VM
represents the driving wheel peripheral speed. Also, the difference speed between the driving wheel circumferential speed and the ground vehicle speed is called the creep speed, and is expressed as VS VS = VM - VO

Defined in (2).

As is well known, the adhesive force is proportional to the axle load, but as shown in FIG. 11, in a range where the slip rate is small, it is also approximately proportional to the slip rate. The value of the slip rate and adhesive force at a certain point determines the position in terms of adhesive properties, which will be referred to as the operating point. Now, the adhesion force is F, and the driving force transmitted from the electric motor to the driving wheels (hereinafter referred to as the driving force around the driving wheels) is FM.
, the following equation holds true, where FR is a rotational system driving force (hereinafter simply referred to as rotational system driving force) that accelerates a rotational inertial system such as an electric motor rotor, a reduction gear, and a driving wheel. FM=F+FR

(3) The above equation shows that the sum of the adhesive force and the rotation system driving force is the driving force around the driving wheels, and the adhesive force is a component force that contributes to the acceleration of the vehicle body.

The adhesive force F has an upper limit value FO, which is called the adhesive limit. When no wheel circumferential driving force is applied (static state or coasting state), the operating point is at the origin, and as the wheel circumferential driving force is applied, it climbs the slope on the left side of the adhesion characteristic. Then, a point P is reached where the adhesive force reaches a maximum value FO at a certain slip rate λO. If the driving force around the driving wheels continues to increase, the slope on the right side of the adhesion characteristic will start to fall, and the adhesion force F will decrease.As can be seen from equation (3), the rotational driving force FR will increase, and the rotational speed of the driving wheels will decrease. will rapidly increase. This phenomenon is called idling. The area where the slip rate is smaller than the adhesive force limit point P will be called the creep area, and the area where it is larger will be called the slip area.

It is known that the adhesive properties vary depending on the weather and the condition of the rail surface, but as a general tendency, the upper limit of the adhesive force F in the dry state is as shown in FIG.
O is large, and the slip rate that gives FO is small. On the other hand, in a wet state, the upper limit of adhesive force F is low as shown in Fig. 12.
As O 2 decreases, the characteristics become flat as a whole. Normally, a torque current pattern is set that allows running with an adhesive force lower than the upper limit value FO of adhesive force in a dry state, so stable running in a creep region is possible in a dry state. However, due to rain and oil stains on the rail surface, the
In some cases, the upper limit value of adhesion force FO may be lower than the required adhesion force, and even when normal driving force is applied to the driving wheels, the operating point passes the point P that gives the upper limit value FO and enters the slipping region. It will enter.

In the conventional readhesion control method using slip detection, after entering this slip region, the driving force around the driving wheels is suddenly reduced so as to return the operating point in the direction of arrow b, and the operating point is brought into the creep region. Pull back. Then, when readhesion is determined, the driving force around the driving wheels is gradually increased, but at this time, the operating point moves in the direction of arrow a, passes through point P again, and enters the idling region. Thereafter, the operation of reciprocating left and right around point P is repeated.

Up to this point, we have described the adhesion characteristics and movement of the operating point when focusing on one driving wheel, but in reality there are multiple driving wheels, and the adhesion characteristics and operating point for each of them are as follows: As shown in FIG. 13, they are naturally different. This is because the adhesion characteristics vary depending on the physical conditions of the contact surface between the driving wheels and the rail, and the driving force around the driving wheels of each driving wheel does not necessarily match due to the characteristics of each induction motor and the difference in driving wheel diameter. This is because there is no guarantee that there will be.

[0023]

[Means for Solving the Problems] FIG. 1 is shown to facilitate understanding of the basic technical idea of the present invention. That is, in principle, the present invention can be realized by optimally utilizing the advantages of fuzzy inference. This will be explained before going into the detailed explanation of FIG. According to fuzzy control, even if a control system is complex and cannot be solved analytically, if an expert has the know-how or knowledge of how to operate for various conditions, this can be expressed in language as a control law, and i
The manipulated variable can be determined using f then -type reasoning. As mentioned above, the adhesive properties vary stochastically due to various factors (season, weather, oil stains on the rail surface, slope, passing through curves, etc.), and it can be said that it is difficult to model them using mathematical formulas. Therefore, we constructed a fuzzy control system by estimating local information on adhesive properties from observable state quantities and using the knowledge of how to control motor torque to maximize adhesive force. This becomes effective.

Fuzzy control also has the characteristic that multi-purpose control can be easily realized. In other words, when different control objectives exist, control that simultaneously satisfies them can be realized. As mentioned earlier,
Controlling multiple induction motors with one inverter
In a VVF inverter electric train, there is an independent control objective for each driving wheel to obtain the maximum adhesion force, and this objective is satisfied by the voltage frequency control of the same inverter device, so multi-purpose control is used. can be considered. Therefore, from this point of view, fuzzy control can be expected to be effective.

Next, a method of calculating antecedent variables that are input to the fuzzy inference block will be explained. As the antecedent variables of the fuzzy inference in the present invention, the time differential dλ/dt of the slip rate λ, the time differential dF/dt of the adhesive force F, and the creep velocity VS are used.
Since dt can be easily obtained by differentiating Equation (1), and the creep speed VS can be obtained from Equation (2), only the method of calculating the adhesive force differential dF/dt will be explained below.

Now, if we ignore the torsion of the joint from the electric motor to the pinion of the drive device, the backlash of the gears of the gear system, the torsion of the driving wheel shaft, etc., the driving wheel transmitted from the electric motor to the driving wheel in equation (3) above The circumferential driving force FM and the rotation system driving force FR can be approximated by the following equations. FM = (z/r)KM II

(4) FR = (J/r2)dV
M/dt
(5) Here, z is the reduction ratio from the electric motor to the driving wheels, r is the radius of the driving wheels, J is the total moment of inertia of the rotational inertia system (converted value to the driving wheels), KM is the motor torque constant, and II is per motor is the torque current. In addition, the driving wheel circumferential speed VM is the motor frequency fM
As something proportional to VM = KO ・fM

(6) can be used. However, the proportionality coefficient KO is KO = 2πr/(p・z
)
(7), where p represents the number of pole pairs of the induction motor.

From equations (4), (5) and equation (2), the adhesive force F is calculated as follows. F=(z/r)KM II −(J/r2)
dVM/dt (8
) Here, K1 = z・KM /r, K2 = J
/r2
(9), equation (8) becomes F=K1 II −K2dVM /dt

It can be written as (10). Therefore, the adhesive force differential is obtained by differentiating equation (10) as follows: dF/dt=K1dII/dt−K2d2 V
M/dt2
(10') is given by.

Next, details of the fuzzy control law will be explained. An example of the fuzzy control law is the slip rate derivative dλ/
The first rule group outputs the torque current correction pattern differential dIp'/dt according to the state of dt and the adhesive force differential dF/dt, and as a brake against the case where the slip gradually grows at a slow speed. , the torque current correction pattern is differentiated by the magnitude of the creep speed dIp'/dt
It can be written in a composite format with a second rule group that outputs.

First, the first rule group will be explained. Here, as shown in ``If the slip rate derivative dλ/dt is positive and large and the adhesive force derivative dF/dt is negative and large, reduce the torque current command greatly'', the slip rate derivative dλ/d
t, the adhesive force differential dF/dt, and the torque current correction pattern differential dIp'/dt, using the membership functions shown in FIGS. 4, 5, and 6, respectively, to express the positive/negative and magnitude relationships in language. This expression uses the following fuzzy labels. PB Positive and large PS Positive and small ZO NS close to zero NB Negative and small (absolute value)
Negative and large (in absolute value) This makes the previous example: if dλ/dt=PB and dF/dt
=NB then dIp'/dt=PB. Here, dIp'/dt=PB is a positive value because
This is because the torque current pattern in FIG. 1 is corrected by subtractive expression, similar to FIG. 10 described above.

Further, the membership function can be expressed using triangles as shown in FIGS. 4 to 6. Now, if a value of [dλ/dt=+0.32] is detected as the slip rate differential dλ/dt, as shown in FIG. 4,
The goodness of fit that dλ/dt=PS is 0.4, dλ/d
The fitness for t=PB is 0.6, and the fitness for other fuzzy labels is 0. Similarly, the adhesive force differential dF/dt
If measured values are also detected for , the goodness of fit for the two fuzzy labels is determined. Here, it is possible to leave it to the designer's discretion as to how much of each input value should be set as PB and set as NS.

Furthermore, Table 1

[Table 1] A fuzzy control law (hereinafter simply referred to as control law) can be defined. However, for simplicity, the vehicle can only move in one direction (
Consider powering operation with λ≧0). In FIG. 1, fuzzy inference blocks 8A to 8D include such a control law and a fuzzy inference section. This first set of rules, which will be explained in detail later, has the property of controlling the motor torque so that the operating point always reaches the maximum point of the adhesive property, and is the principle of maximizing the use of adhesive force. This is the control law that forms the basis of the invention.

Next, the purpose and method of the second rule group for using the creep speed VS in fuzzy inference will be explained. FIG. 7 shows the adhesion characteristics when the rail surface is extremely slippery due to early rain or dew condensation. In this case, the adhesion limit point P shown in FIG. 11 and FIG.
This is the case when the trend is not clear and shows an almost flat trend. In this case, when the operating point enters the flat region, there is almost no change in adhesive force, and dF/dt=ZO. If the creep rate grows slowly here, there is almost no change in the slip rate, so dλ/dt = ZO, and the fuzzy inference from the control law table will output almost zero, maintaining the current torque current command. It turns out. As a result, the creep rate increases indefinitely. When the creep speed increases, squeaks between the rail and the driving wheels, vibrations of the bogie, etc. occur, causing maintenance problems and at the same time impairing riding comfort.

To overcome this drawback, creep rate evaluation is incorporated as a backup to the control law described above. The control law for creep rate is as follows. if VS=PS then dIp'/dt=P
Sif VS =PB then dIp'/dt=
PB Also, membership functions for these are shown in FIG. As can be seen from Figure 8, the creep speed is 5km/h.
In the following, the goodness of fit with VS = PS is also defined as VS = PB
Both the fitness degrees become zero, and the output from this control law, that is, the torque current correction pattern differential dIp'/
dt becomes zero. Also, the creep speed increases to 5km
/h or more, the control law is such that the output is increased according to the magnitude of the creep speed.

[0034] The reason for limiting the creep speed rather than the slip rate is that it is the increase in the creep speed itself that causes the squeaking noise between the rail and the driving wheels and the vibration of the bogie. This is because it is necessary to evaluate the creep rate itself rather than the slip rate.

Up to this point, the fuzzy inference for obtaining a correction pattern of torque current when focusing on one driving wheel has been described. Next, a method of performing this inference for each driving wheel, max-synthesizing the obtained results, and further defuzzifying them to obtain a torque current correction pattern differential will be described in detail.

The above-mentioned fuzzy control law, including the first rule and the second rule, can be described in the following general form. if dλ/dt=A1 dF/dt=B1
VS=C1 then dIp'/dt=D1
if dλ/dt=A2 dF/dt=B2
VS=C2 then dIp'/dt=D2

:
: if
dλ/dt=AN dF/dt=BN VS=CN
then dIp'/dt=DN where dλ/d
t, dF/dt, VS, dIp'/dt represent the slip rate differential, adhesive force differential, creep speed, and torque current correction pattern differential in a certain driving wheel, and Aj (j=1
~N) represents one of the membership functions NB, NS, ZO, PS, and PB regarding the slip rate differential. Similarly, Bj (j=1 to N), Cj (
j=1 to N) and Dj (j=1 to N) represent membership functions of adhesive force differentiation, creep speed, and torque current correction pattern differentiation, respectively. Here, N represents the total number of control laws.

As the fuzzy inference in the present invention, the most commonly used max-min synthesis method is used. Now, the above Aj, Bj, Cj, D
The membership functions corresponding to j are μAj, μBj
, μCj, μDj. The slip rate differential, adhesive force differential, and creep speed at the driving wheel of the A-axis shown in Figure 1 are as follows:
If it is written as dλA /dt, dFA /dt, and VSA, the result of applying the fuzzy control law (hereinafter referred to as fuzzy conclusion) is given as the membership function μ10A shown in the following equation. μ10A = (ω1 ∧μD1)∪…∪(ωN
∧μDN) (
11) Here, ωj (j=1~N) is the fitness of the antecedent, and ωj = μAj (dλA /dt)∧
μBj(dFA/dt)∧μCj(VSA)
(12) can be written as In the above formula, ∧ is
min operation, and ∪ is the set sum operation (here ma
x operation).

In the first rule group described above, there is no condition regarding the creep velocity VS, while in the second rule group, there is no condition regarding the slip rate differential and the adhesive force differential. The fitness of an antecedent variable with no corresponding condition is set to 1. That is, in the first rule group, μCj (dVSA/dt) = 1, and in the second rule group, μAj (dλA /dt) = μBj (dFA/dt) =
Set to 1.

Similar to the A-axis, fuzzy inference is also performed on the B-axis, C-axis, and D-axis, and the fuzzy conclusions μO A , μ
Obtain O B , μO C , and μO D . These fuzzy conclusions about each driving wheel obtained in this way are ma
x synthesis, defuzzification is performed by gravity center calculation to obtain the final torque current correction pattern differentiation. That is, μ1 * = μ10A ∪μ10B ∪μ10
C ∪μ10D
(13), each fuzzy conclusion is m
ax synthesis, dIp'/dt={∫μ1 * (
dIp'/dt) (dIp'/dt) d(dIp'/
dt) } /{∫μ1
* (dIp'/dt) d(dIp'/dt) }
Find the center of gravity of μ1* using (14). Note that the integral in Equation (14) is an integral over the support set of membership functions, and is not a time integral.

[0040] Furthermore, when max-synthesizing the fuzzy conclusions of each driving wheel, instead of simply max-synthesizing,
Weighting can also be performed using weights γA to γD for each driving wheel. That is, the following equation is used instead of equation (13). μ1 * = γA ・μ10A ∪γB ・μ
10B ∪γC ・μ10C ∪γD・μ10D


(13') Here, . represents an algebraic product, and represents an operation of uniformly multiplying the membership value of the fuzzy conclusion by a constant. Here, for example, if the weight of the A-axis driving wheel is 1 and all other weights are 0, then Equation (13') becomes μ1 * = μ10A, and the A-axis is selected as the reference axis for fuzzy inference. The same result is obtained. That is, equation (13') is
It can be seen as a generalization of the method of setting a reference axis and performing fuzzy inference.

Next, a method of directly narrowing down the slip frequency provided to suppress rapid growth of slip will be explained. In this method, an evaluation index δ representing the degree of slipping is obtained using a method similar to the fuzzy inference using the first rule group described above, and the amount of reduction of the slip frequency is determined according to the value of the evaluation index δ. The evaluation index δ is defined as a numerical value ranging from 0 to 1. As an example of a fuzzy control law when determining the evaluation index δ by fuzzy inference, the following can be considered. if dλ/dt=PS dF/dt=NB then
δ=PBif dλ/dt=PB dF/dt=NB
then δ=PB The consequent variable of this control law, that is, the evaluation index δ, becomes a value close to 1 when the operating point enters the slipping region and a sudden decrease in adhesive force occurs, indicating that the operating point is within the creep region. In this case, or when the slip grows relatively slowly, it becomes 0. The membership function for the evaluation index δ is shown in FIG.

If the above control law is written in the same general form as the first rule group and second rule group, if dλ/
dt=A1 dF/dt=B1 then δ=E
1 if dλ/dt=A2 dF/dt=B2 t
hen δ=E2 : : if dλ/dt=AM dF/dt=BM the
n δ=EM where, dλ/dt, dF/dt
and δ represent the slip rate differential, adhesive force differential, and slip evaluation index for a certain driving wheel. Also, Aj
(j=1 to M) represents a membership function for the slip rate differential, NB, NS, ZO, PS, or PB; similarly, Bj (j=1 to M),
Ej (j=1 to M) represents the membership function of the adhesive force differential and the evaluation index of slipping, respectively. M represents the total number of control laws. (In the above example, M=2.)

Here, the fuzzy conclusion about the A axis is μ2
0A, μ20A = (ω1 ∧μE1)∪…∪(ωM
∧μEM)
(15) can be written as Here, ωj (j=
1 to M) represent the goodness of fit of the antecedent part. Then, similarly to equation (13') above, the fuzzy conclusions for each driving wheel are weighted using the weights set for each driving wheel to obtain the final fuzzy conclusion. That is, μ2 * = γA ・μ20A ∪γB ・μ
20B ∪γC ・μ20C ∪γD ・μ20D


(16) Furthermore, defuzzification is performed by calculating the center of gravity using the following equation to obtain a definite value. That is, δ=∫μ2 * (δ)δ dδ/∫μ2 *
(δ) dδ (
17). Using the slip evaluation index δ obtained in this way, the slip frequency command is calculated as fSS=G0 ∫(1-δ)・(Ip −II
)dt +KS ∫δdt (
18) Determined by. This is an amount equivalent to the weighted average of the slip frequency command when the slip detection signal is on in the conventional method and the slip frequency command when the slip frequency command is in the off state, weighted by the evaluation index δ. It is made into a fuzzy and continuous control.

As can be seen from equation (18), when the degree of slipping is low, that is, when the value of the evaluation index δ is almost zero, adhesion control is performed by modifying the torque current according to the first rule group and the second rule group. The system works. On the other hand, if the number of wheels rapidly increases at the beginning of rain, etc., the evaluation index δ
approaches 1, so the slip frequency must be narrowed down. Correspondingly, the torque current pattern Ip is modified to a value of (1-δ)・Ip +δ・II, and the torque current correction pattern Ip' is further modified to a value of (1-δ)・Ip'. do. As a result, the evaluation index δ
approaches 1, and the torque current command IIS (=Ip - Ip') during the period when the slip frequency is narrowed down becomes approximately equal to the detected torque current value II, so the slip frequency command increases due to the current deviation. There is no problem of doing so.

Next, FIG. 1 will be explained in detail. In FIG. 1, the same reference numerals as in FIG. 10 indicate parts having the same functions. The outputs of PG 4A to 4D are calculated by motor frequency calculating means 5A to 5D, respectively, to determine the motor rotation frequency f of each axis.
It is converted into MA to fMD and becomes input to the arithmetic units 6A to 6D. In addition, the phase current detection values IuA, I of each induction motor
vA to IuD, IvD, inverter device 101
The output voltage phase θ and ground vehicle speed V0 are also calculated by the calculation device 6.
It is input to A to 6D. The calculation devices 6A to 6D input the above information and calculate the slip rate differential dλA /d at each driving wheel.
t~ dλD/dt, adhesive force differential dFA/dt~
It calculates and outputs dFD/dt and creep speeds VSA to VSD.

The outputs of the arithmetic units 6A to 6D are input to fuzzy inference blocks 8A to 8D, and fuzzy inference is executed for each driving wheel based on the aforementioned fuzzy control law. There are two outputs of fuzzy inference blocks 8A-8D. One is the torque current correction pattern differential dIp'/dt, and the other is the index δ representing the degree of slippage. Each output is a membership function representing a fuzzy conclusion. The fuzzy conclusions μ10A to μ10D for each driving wheel regarding torque current correction pattern differentiation are weighted by algebraic product means 9A to 9D, and max synthesis means 1
In step 2, the max composition operation is executed, and the fuzzy conclusion μ1* shown in equation (13) is output. 13 is a defuzzification means, which converts the fuzzy inference μ1* into equation (1
The center of gravity calculation shown in 4) is performed, and a definite value can be obtained by this.

Reference numeral 11 denotes a weight determining means, which outputs weights γA to γD indicating how much the fuzzy conclusions μ10A to μ10D of each driving wheel should be respected. This may be fixed to a certain constant value for each driving wheel, or may be a quantity that dynamically changes according to each state quantity during driving. As an example, if weighting is performed according to the magnitude of adhesive force in each driving wheel, γA to γD are as follows. γA = FA / (FA + FB + FC + FD)
γB = FB / (FA + FB + FC + FD)
γC = FC / (FA + FB + FC + FD)
γD = FD / (FA + FB + FC + FD)
Here, FA to FD represent the adhesive force at each driving wheel. By performing this weighting, it is possible to achieve control in which the fuzzy conclusion of the driving wheel with a large adhesive force is given more importance, and the fuzzy conclusion of the driving wheel with a small adhesive force is not so important.

On the other hand, the fuzzy conclusions μ20A to μ20D for each driving wheel regarding the evaluation index δ, which is another output of the fuzzy inference blocks 8A to 8D, are obtained by the algebraic product means 1.
Weighted by 0A to 10D, max synthesis operation is executed by max synthesis means 14, and formula (16)
The fuzzy conclusion μ2* shown in is output. 15 is the same defuzzification means as 13, and the fuzzy conclusion μ2
* The center of gravity shown in equation (17) is calculated from *, and a definite value can be obtained thereby.

Now, the momentary torque current command correction pattern differential dIp, which is the output of the defuzzifying means 13
'/dt is integrated by the pseudo-integrator 16, and is added as the torque current correction pattern Ip' to the torque current pattern Ip generated by the torque current pattern generation means 104, which was explained in FIG. It becomes IIS. The deviation between the torque current command IIS and the detected torque current value II is determined at an addition point 105', multiplied by G0 via a gain multiplier 116, and further multiplied by (1-δ) at a multiplier 119. Then, this value and the value obtained by multiplying the slip frequency reduction amount KS (>0) stored in the coefficient generator 115 by δ in the multiplier 118 are added at the addition point 120, and the slip frequency The command becomes fSS. That is, the slip frequency command is determined according to equation (18). Below, corresponding to the slip frequency command fSS and the motor frequency fM, Fig. 1
The frequency and voltage of the inverter will be controlled in the same way as in 0. 19 is a ground vehicle speed detection means, which detects the ground vehicle speed V
Outputs 0. Although this example shows a method of calculating from the signal of the PG 18 attached to the follower wheel 17, the method is not particularly limited to this method.

FIG. 2 shows the internal structure of the arithmetic unit shown at 6A to 6D in FIG. The arithmetic units 6A to 6D have the same internal structure except for input/output variables, and in FIG. 2, the subscripts A to D that distinguish the driving wheels are omitted for simplicity. First, the motor frequency fM is determined by converting means 604 as follows:
It is converted into a driving wheel peripheral speed conversion value VM (hereinafter simply referred to as driving wheel peripheral speed for simplicity). K0 in the figure is expressed by the above formula (6
). The ground vehicle speed V0 is subtracted from the driving wheel peripheral speed VM at an addition point 605, and the result is divided by the driving wheel peripheral speed VM itself by a divider 606 to obtain the slip rate λ. That is, the slip rate is calculated according to the above equation (1). Here, it is assumed that the divider 606 includes a process of forcibly setting the slip rate to 0 without performing division when the divisor VM is a very small value, that is, during a low speed period. Further, the slip rate is differentiated by a differentiating means 607 to obtain the time differential dλ/dt of the slip rate. Here, the differentiating means 607 may perform differentiation in a low frequency range sufficient to express the rotation of the driving wheels, and is integrated with a second-order lag filter that has good attenuation against high frequency noise.

On the other hand, at the addition point 610, the driving wheel peripheral speed VM
The differential speed between the ground vehicle speed V0 and the ground vehicle speed V0 is calculated and passed through a low-pass filter 611 to become the creep speed VS. Further, the effective current calculation means 601 calculates the effective current IM from the phase current detection values Iu and Iv of the corresponding induction motor.
102 in Figures 10 and 1.
This has the same function as the effective current detection means shown in . However, phase current detection is performed for each induction motor. Torque current calculation means 602 is an effective current value IM
A current proportional to the motor torque, that is, a torque current I1, is calculated from the output voltage phase θ of the inverter. Torque current I1, which is the output of torque current calculating means 602, is added to addition point 609 via differentiating means 603. On the other hand, the driving wheel peripheral speed VM is subtracted at an addition point 609 via a second-order differentiator 608. That is, the adhesive force differential is calculated according to equation (10'). Gains K1 and K2 of the differentiating means 603 and the second-order differentiating means 608 are shown in equation (6). Also,
The differentiating means 603 and the second-order differentiating means 608, like the above-mentioned differentiating means 607, need only perform a differentiating operation on the low frequency component, and are integrated with a second-order lag filter that has good blocking characteristics against high-frequency noise. ing.

Here, filter constants ω1 to ω4, ξ
1 to ξ4 sufficiently attenuates high-frequency noise amplified by differential or second-order differential, and also reduces the circumferential speed of the driving wheels,
The phase delay should be selected so that the phase delay is as small as possible with respect to the ground vehicle speed. Furthermore, although a second-order lag filter is used as the filter, it is not particularly limited to this as long as it has similar characteristics. Through the above calculations, the slip rate differential dλ/dt, the adhesive force differential dF/dt, and the creep speed VS are determined.

FIG. 3 is a diagram showing the internal structure of the fuzzy inference blocks indicated by 8A to 8D in FIG. 1. Fuzzy inference blocks 8A to 8D differ only in input and output variables,
Since the internal structure is the same, the subscripts A to D that distinguish the driving wheels are omitted in FIG. 3 for simplicity. The slip rate differential dλ/dt, adhesion force differential dF/dt, and creep velocity VS, which are input to the fuzzy inference block, are each normalized by normalization means 801X to 801Z, and are input to the fuzzy inference means 802. Normalization refers to the range of criteria (e.g. [-
1, +1]), and this allows membership functions of the same shape to be shared. The fuzzy inference means 802 applies the fuzzy control law built in the fuzzy rule block 803 to each normalized input to execute fuzzy inference. The fuzzy rule block 803 stores the aforementioned first rule group, second rule group, and rules for inferring the evaluation index δ.

Now, in many fuzzy control systems, a complete integrator is provided after the fuzzy inference block, and is generally called a fuzzy PI control system. In the present invention, pseudo-integration as shown below is used instead of complete integration. This is to prevent the effects of noise superimposed on the antecedent variables of fuzzy inference from being accumulated and retained by complete integration and outputting an unnecessarily large torque current correction pattern. It is preferable to gradually reduce the torque current correction pattern when the output is zero. The pseudo-integrating means 16 has such characteristics, and as an example, a first-order delay means expressed by τ/(1+τ·S) can be used. Here, a relatively long time constant (on the order of several seconds) is used as the time constant τ. Furthermore, the pseudo-integration means 16 has a limiter function that sets the upper limit value of the pseudo-integral value to torque current pattern Ip and the lower limit value to zero, so that the torque current command IIS is equal to or higher than the torque current pattern Ip. It is prohibited to have a value less than or equal to zero.

[0055]

[Operation] FIG. 11 of the adhesion characteristics shown earlier for each means shown in FIGS. 1 to 3, especially the fuzzy inference blocks 8A to 8D.
, will be explained in conjunction with FIG. 12 and Table 1 of the control law. Note that in Table 1, ■ to ■ each represent an area surrounded by a dotted line.

In a normal adhesion state, the operating point is on the left side of point P in FIG. 11, and the driving wheel peripheral power is generated at a slip rate of λ0 or less. At this time, the creep rate becomes a small value, so as can be seen from the membership function in Figure 8,
Both the degree of fitness where VS = PS and the degree of fitness where VS = PB are 0, and the second rule group based on the creep rate does not work.

Now, due to rain, etc., the adhesive strength decreases, and as shown in FIG.
Suppose that the adhesive strength characteristics are as shown in 2. In this case, it is ideal to operate near point P. Now, when the adhesive force decreases, slipping occurs, the slip rate λ increases, and the operating point tends to shift in the direction of arrow d. At this time dλ/d
Since t>0 and dF/dt<0, the part indicated by ■ in Table 1 of the control law corresponds, and dIp'/dt=
Since PS or dIp'/dt=PB, it acts to reduce the motor torque and return it to the direction of arrow b. That is, if dλ/dt=PS and dF/dt in the part ■ in the control law table
=NS then dIp'/dt=PS i
f dλ/dt=PB and dF/dt=NS
then dIp'/dt=PB etc. control law is dλ/
It is used depending on the value of dt and dF/dt.

Furthermore, when shifting to the mode of arrow b, d
λ/dt<0, dF/dt>0, and the control law indicated by ■ in the control law table acts, and dIp'/dt=N
S, that is, acts to slightly increase the torque current command. This is an empirical control law that anticipates a delay until the electric motor torque actually increases and the wheel circumferential driving force starts to increase again. As a result, the operating point does not pass over the point P and enter the area indicated by the arrow c, and it is possible to promote an effect in which the operating point is balanced in the vicinity of the point P.

Next, if the operating point passes point P and returns to the creep region and moves in the direction of arrow c, then
Since dλ/dt<0, dF/dt<0, the control law for the part indicated by ■ in the control law table corresponds, and dIp'/d
t=NS or dIp'/dt=NB, that is, it acts to increase the torque current command. As a result, the motor torque increases and the operating point moves in the direction indicated by arrow a.

When the operating point is in the state indicated by arrow a, d
Since λ/dt>0, dF/dt>0, the control law for the part indicated by ■ in the control law table corresponds, and dIp'/dt
= NS, that is, it acts to slightly increase the torque current command,
The motor torque is gradually increased to reach point P. Since the adhesion force limit point P changes from time to time, the control system repeats the above-mentioned mode and operates to constantly track the point P. When the adhesive force is stabilized at a constant value, it is balanced near the center of the control law table, and the operating point can be stabilized near the point P.

Furthermore, dλ/dt>0, dF/dt<0
In the case of dF
When /dt=NB, the value of the slip evaluation index δ approaches 1, so the slip frequency is narrowed down according to equation (18), and the operating point is quickly returned to the creep region.

Furthermore, in a state exhibiting flat adhesion characteristics as shown in FIG. 7, the slip rate differential takes a small value,
The slow growth of the creep rate cannot be captured and the creep rate gradually increases, but in this case, the second set of rules based on the evaluation of the creep rate limits the creep rate to within a certain value, and the creep rate increases gradually. The torque current will be modified to obtain the maximum adhesive force within the limits.

The above is the effect of the fuzzy control law when focusing on one driving wheel. Next, the function of the multi-purpose control of the present invention, which combines fuzzy conclusions for each driving wheel to the max and defuzzifies them, will be explained.

FIGS. 13 and 14 illustrate the adhesion characteristics and operating points of each driving wheel, and the movement of the operating points when the adhesion control according to the present invention is applied. Figure 13
(a) shows a case where only the adhesive property of the A-axis is degraded due to dirt on the rail surface, etc., and only the A-axis is idling. At this time, the operating point on the A-axis enters the idling region, so a fuzzy conclusion can be obtained in the direction of reducing the motor torque, that is, increasing the torque current correction pattern, but the operating point on the other B to D axes creeps. Since it is on the region side, a fuzzy conclusion can be obtained in the direction of increasing the motor torque, that is, decreasing the torque current correction pattern. Then, these two contradictory fuzzy conclusions are combined to the max, and a nearly intermediate conclusion that slightly increases the motor torque, ie, slightly decreases the torque current modification pattern, is obtained. As a result, each operating point moves to a position as shown in FIG. 13(b), and it becomes possible to make maximum use of the entire adhesive force without reducing the motor torque more than necessary.

FIG. 14(a) shows a case where the adhesion characteristics of all the driving wheels are degraded due to rain or the like. At this time, the operating points of all driving wheels enter the idling region, so
The same fuzzy conclusion is obtained in the direction of decreasing motor torque, ie increasing the torque current modification pattern. Even if these are combined to the maximum, the conclusion is that the torque current correction pattern will still increase, so each operating point will be pulled back into the creep region, as shown in FIG. 14(b).

[0066]

DESCRIPTION OF THE PREFERRED EMBODIMENTS Specifically, the control block diagram shown in FIG. 1 can be completely digitalized. That is, 16-bit DSP (digital signal processor)
This can be achieved by adopting such methods. In this example, since a PG is used as the speed detector, digital calculation is possible for speed detection. Moreover, by digitizing the current detection value using an A/D converter inside the effective current detection means 102 and the torque current calculation means 602, all signals to the next stage can be processed by digital calculation. Furthermore, normalization means 801X to 8
01Z also serves to quantize each input into 29 steps. Therefore, fuzzy inference blocks 8A-8
The membership function included in D can be realized as a staircase waveform in which the gradient from 0 to 1 is approximated by steps. In addition, the inverter control section after addition point 105' is a digital control system using another microprocessor, but if a faster, higher bit number microprocessor is used, D
It is also possible to include a fuzzy control section composed of SP.

[0067]

As described in detail above, by combining the fuzzy control method and information regarding adhesion, it is possible to realize a control system that has the advantages of a mathematical information processing method and a designer's knowledge. As described above, according to the present invention, it is possible to eliminate the problem of repeating on-off cycles caused by the conventional readhesion method based on slip detection, and to utilize the VVVF inverter that continuously changes torque. Furthermore, it is possible to comprehensively evaluate the state quantities of all the driving wheels, to make maximum use of the adhesion force, and to maintain acceleration even when the adhesion characteristics are degraded. In addition, in response to a sudden increase in slip, the slip frequency is narrowed down.
Rapid re-adhesion is possible.

[Brief explanation of drawings]

FIG. 1 is an overall block diagram of an example of an adhesion control system shown to facilitate understanding of the basic technical idea of the present invention.

FIG. 2 is a block diagram of an arithmetic device for calculating a slip rate differential, an adhesion force differential, and a creep rate.

FIG. 3 is a block diagram showing the internal structure of a fuzzy inference block.

FIG. 4 is a graph showing membership functions of slip rate differentials.

FIG. 5 is a graph showing a membership function of adhesive force differentiation.

FIG. 6 is a graph showing membership functions of torque current modification pattern differentiation.

FIG. 7 is a graph showing flat adhesion properties in slippery conditions.

FIG. 8 is a graph showing the membership function of creep rate.

FIG. 9 is a graph showing a membership function of an evaluation index representing the degree of idling.

[Figure 10] Figure 10 shows VVVF using conventional readhesion control.
FIG. 2 is a block diagram showing a control circuit of an inverter.

FIG. 11 is a graph showing adhesive properties when dry.

FIG. 12 is a graph showing adhesive properties in a wet state.

FIG. 13 is a diagram showing the adhesion characteristics and movement of the operating point of each driving wheel when one axis is idling.

FIG. 14 is a diagram showing the adhesion characteristics and movement of the operating point in each driving wheel when all axes are idling.

[Explanation of symbols]

1 Electric vehicle 2A-2D Induction motor 3A-3D Driving wheel 4A-4D, 18 Motor speed detector (PG) 5A-
5D Motor frequency calculating means 6A to 6D Calculating devices 7A to 7D Differentiators 8A to 8D Fuzzy inference blocks 9A to 9D, 10A to 10D Algebraic product means 11
Weight determination means 12, 14 Max synthesis means 13, 15 Defuzzification means 16 Pseudo-integration means 17 Follower wheel 19 Ground vehicle speed calculation means 101 Inverter 102 Effective current detection means 103, 602 Torque current calculation means 104 Torque current pattern generation means 106 Switching Means 107 Current control means 108 Motor frequency selection means 110 V/f ratio 111, 606 Divider 112 First-order lag filter 113 Slipping detection means 114 Comparator means 115 Coefficient generator 116 Gain multiplier 118, 119 Multiplier 601 Effective current calculation means 603, 607 differentiating means 604 converting means 608 second-order differentiating means 611 low-pass filters 801X, 801Y, 801Z normalizing means 80
2 Fuzzy inference means 803 Fuzzy rule block

Claims (4)

[Claims] 1. An electric vehicle in which a plurality of induction motors are controlled by a variable voltage variable frequency inverter, comprising means for detecting and calculating the driving wheel circumferential speed of each driving wheel, means for detecting and calculating the ground vehicle speed, and the driving wheel circumferential speed. means for calculating the slip rate of each driving wheel from the ground vehicle speed, means for calculating the driving wheel circumferential adhesion force of each driving wheel, means for calculating the time derivative of the slip rate and the adhesion force, and the time derivative of the slip rate. The time derivative of the adhesive force and the differential speed are used as the antecedent variables, and the torque command value of the induction motor is used as the consequent variable. When the differential speed is small, the slip rate differential and the adhesive force derivative are used to calculate the adhesive force. A first rule group that modifies the motor torque command in a direction that can increase Execute the inference and write the obtained result m
ax synthesis and further fuzzification processing to obtain the final correction pattern of the motor torque command, and for rapidly increasing slippage, a fuzzy inference means to obtain an evaluation index representing the degree of slippage of each driving wheel. , the obtained results are synthesized to the maximum, and further subjected to defuzzification processing to obtain a final evaluation index, and a means for reducing the slip frequency of the motor according to this evaluation index is provided. Electric vehicle control device. 2. Claim 1, wherein the torque of the induction motor is command-controlled by a current corresponding to a torque component.
The electric vehicle control device described. 3. The electric vehicle control device according to claim 1, wherein the torque of the induction motor is command-controlled by a torque calculated from an electric signal. 4. The first electric motor torque command correction signal according to claim 1, wherein the result inferred from the first rule group is passed through a first-order delay means and is used as the first electric motor torque command correction signal. Electric vehicle control device.