JP3818777B2 - Active filter control method - Google Patents

Description

[0001]
BACKGROUND OF THE INVENTION
The present invention relates to a method for controlling an injection circuit type active filter that compensates and reduces harmonic voltage distortion generated in a power system.
[0002]
[Prior art]
In recent years, electrical products having power conversion devices using semiconductor elements such as inverter air conditioners have become widespread, and harmonic disturbances have frequently occurred accordingly. Therefore, a harmonic countermeasure active filter (active filter) has been introduced conventionally, and an example thereof will be described below with reference to an equivalent circuit of a general system in FIG.
[0003]
First, in the figure, (1) is a system power source, (2) is a system bus, (3) is an active filter (hereinafter referred to as AF), (4) is a load, and (Zs) is a system impedance. The system power supply (1) is a voltage source including a harmonic voltage component (Vn) together with a fundamental wave component, and is connected to a load (4) via a system bus (2). The load (4) is a variable load such as an arc furnace, and the impedance viewed from the AF (3) is (ZL). The system impedance (Zs) is an impedance when the system power supply (1) side is viewed from AF (3).
[0004]
AF (3) is an injection circuit type having an injection circuit unit (5), a high-frequency inverter (6), and a control calculation unit (7). The injection circuit portion (5) is a capacitor (C) and a reactor (R) connected in series, and is connected to the system bus (2) via a connection point (M). Then, the bus voltage (Va) including the harmonic voltage component (Vn) is detected by the PT (8) connected to the AF connection point (M), and the CT (9) inserted into the injection circuit unit (5) is detected. ) Detects a compensation current (Ia) described later.
[0005]
The high frequency inverter (6) injects the inverter output current (Ii) into the connection point (N) between the capacitor (C) and the reactor (R), and from there to the system bus (2) to compensate for harmonic voltage distortion. A compensation current (Ia) to be reduced is generated. The calculation unit (7) calculates a command value (Ir ′) of the inverter output current (Ii) from the harmonic voltage component (Vn) and the output compensation current (Ia) and inputs the command value (Ir ′) to the inverter (6).
[0006]
The calculation unit (7) includes a harmonic detection unit and an amplitude phase adjustment unit (hereinafter referred to as an adjustment unit). From the detected voltage (Va) of the PT (8), an AF connection point ( The harmonic voltage component (Vn) of M) is detected. The adjustment unit has an adjustment unit configured by neural network control using the harmonic voltage component (Vn) as an input, and compensates the harmonic voltage component (Vn) by a control gain (vector) (G) to inject compensation current. A reference signal (compensation current value to be realized) (Iar) is calculated. Then, the reference signal (Iar) is multiplied by the reciprocal of the shunt rate (K) of the compensation current (Ia) with respect to the inverter output current (Ii), and the inverter output current (Ii) for realizing a desired compensation current (Ia) is obtained. A command value (Ir ′) is calculated.
[0007]
At this time, it is desirable to output the compensation current (Ia) in accordance with the phase of the entire system impedance (Zt) when the power supply and the load side are viewed from the AF connection point (M), while the system impedance (Zt) always fluctuates. is doing. Therefore, it is necessary to constantly adjust and output the compensation current (Ia) using a neural network. The neural network learns (changes) the connection load (W) between neurons every predetermined calculation cycle by the back-propagation method and reviews it. Thus, the compensation current (Ia) is constantly adjusted in a constant cycle following the fluctuation of the system impedance (Zt), and the optimum compensation current (Ia) is output-controlled. The following equation (A) is applied as a teacher signal (Id ′) of the neural network that commands learning of the interneuron connection weight (W).
[0008]
| Id ′ | = | {(0−Vn) / Zt} + Iar | − | Iar |
Zt = {Vn (t-1) -Vn (t)} / (Ia-0) (b)
{However, Vn: Harmonic voltage component of AF connection point (M), Iar: Compensation current injection reference signal, Zt: System impedance viewed from AF connection point (M), Vn (t-1): No Ia (AF Vn at disconnection) (0 at startup), Vn (t): Vn with Ia (AF connection)}
According to the above configuration, the compensation current (Ia) proportional to the harmonic voltage component (Vn) is generated from the inverter (6), and AF (3) is generated by the harmonic voltage component (Vn) and the compensation current (Ia). Apparently, an equivalent resistance (Ra = Vn / Ia) is obtained at a specific frequency. Therefore, when the harmonic voltage component (Vn) of the system power supply (1) expands due to resonance between the inductive power supply system impedance (Zs) and the load (ZL) capacitive (for example, a power factor improving capacitor). The equivalent resistance (Ra) acts as a damping resistance, resulting in a reduction of the harmonic voltage component (Vn).
[0009]
At this time, it is effective to make AF (3) a resistance element having the same phase as the system impedance (Zt) viewed from the AF connection point (M). Therefore, the compensation current injection reference signal (Iar) is the system impedance (Iar). Zt) and the harmonic voltage component (Vn) are set by Iar = Vn / Zt. In a normal system, the system impedance (Zt) constantly changes, and the damping resistance action varies accordingly. Therefore, as described above, the compensation current injection reference signal (Ira) is constantly adjusted by the neural network control so that the AF (3) acts as an optimum impedance following the fluctuation of the system impedance (Zt), and the optimum The output of the compensation current (Ia) is controlled.
[0010]
[Problems to be solved by the invention]
In the above injection circuit type AF (3), the compensation current (Ia) for the inverter output current (Ii) is given by the following formulas (C) and (D).
[0011]
Ia = K · Ii (C), K = Zr / (Zt + Zc + Zr) (D)
{However, K: shunt ratio of compensation current (Ia) to inverter output current (Ii), Zt: system impedance viewed from AF connection point (M), Zc: impedance of capacitor (C) of injection circuit section (5) , Zr: Impedance of reactor (R) of injection circuit section (5)}
Here, when Zt is capacitive and | Zt + Zc |> | Zr | (for example, when the power supply side impedance and the load side impedance are resonating), the impedances (Zt + Zc) and (Zr) are negative and Positive, (Zt + Zc + Zr) is negative, and the diversion ratio (K) is also negative. For this reason, the phase of the compensation current (Ia) is reversed with respect to the command value (Ir ′) according to the formula (C), so that a normal compensation current (Ia) cannot be obtained and a compensation effect cannot be obtained.
[0012]
For example, the AF connection point (M) is set as the delivery side of the distribution transformer, and the impedance of the distribution transformer (Zs = j17.4) and the load side impedance including the power factor improving capacitor (ZL = −j14.5) Suppose that harmonic expansion occurs due to resonance. (However, the value in parentheses indicates the impedance with respect to the seventh harmonic.)
Zt = (Zs · ZL) / (Zs + ZL) = − j87.0
Zt + Zc = −j103.5, Zr = j73.9
Thus, the system impedance (Zt) is capacitive, and | Zt + Zc |> | Zr |. Here, if Ii = 1 (A),
Ia = {j73.9 / (− j103.5 + j73.9)} × 1≈−2.5 (A)
Thus, the phase of the compensation current (Ia) is reversed with respect to the command value (Ir ′), and the compensation effect cannot be obtained.
[0013]
An object of the present invention is to provide a method for controlling an injection circuit AF that provides a good compensation effect even under the condition that the system impedance (Zt) is capacitive and | Zt + Zc |> | Zr |.
[0014]
[Means for Solving the Problems]
The present invention includes an injection circuit unit in which a capacitor and a reactor are connected in series and connected to a system bus, a high-frequency inverter that injects an inverter output current at a connection point between the capacitor and the reactor of the injection circuit unit, and detection from the bus voltage And an arithmetic unit for calculating an inverter output current command value for outputting an optimum compensation current against system fluctuations by an adjusting means configured by neural network control using the input harmonic voltage component as input, and inputting the command to the inverter. In order to control the injection circuit type AF for injecting the compensation current into the system bus and compensating / reducing the harmonic voltage distortion, the neural network control teacher signal is
| Id | = | {(0−Vn) / (K · Zt)} + Iar | − | Iar |
(However, | Id |: Teacher signal, Vn: Harmonic voltage component of AF connection point, Iar: Compensation current injection reference signal, K: Shunt ratio of compensation current with respect to inverter output current, Zt: From AF connection point at start-up The neural network is commanded in consideration of the shunt from the inverter output current to the compensation current.
[0015]
According to the above configuration, the teacher signal of the present invention gives a learning command to the neural network in consideration of the shunt from the inverter output current (Ii) to the compensation current (Ia) by taking into account the shunt ratio (K), and the compensation current Adjust the injection reference signal (Iar). As a result, the inverter output current (Ii) is controlled so that the phase inversion of the compensation current (Ia) does not occur with respect to the command value (Ir) due to the shunt from the inverter output current (Ii) to the compensation current (Ia).
[0016]
DETAILED DESCRIPTION OF THE INVENTION
An embodiment of an AF control method according to the present invention will be described below with reference to FIG. A feature of the present invention is that a teacher signal for neural network control in the calculation unit (7) is set by the following equation (e), and the program content of the calculation unit (7) is changed accordingly.
| Id | = | {(0−Vn) / (K · Zt)} + Iar | − | Iar |
[However, | Id |: Teacher signal, Vn: Harmonic voltage component at AF connection point (M), Iar: Compensation current injection reference signal, K: Shunt ratio of compensation current (Ia) to inverter output current (Ii) { K = Zr / (Zt + Zc + Zr) (D)}, Zt: System impedance viewed from AF connection point (M) at start-up, Zc: Impedance of capacitor (C), Zr: Impedance of reactor (R)]
According to the equation (e) of the teacher signal, the neural network is instructed to learn by considering the shunt from the inverter output current (Ii) to the compensation current (Ia) by the shunt rate (K). That is, the equation (e) adjusts the compensation current injection reference signal (Iar) in consideration of the shunting state from the inverter output current (Ii) to the compensation current (Ia) by taking into account the shunt rate (K). Output current command value (Ir) is output. As a result, the desired compensation current (Ia) can be directly output controlled via the inverter output current (Ii).
[0017]
Therefore, for example, when Zt is capacitive, | Zt + Zc |> | Zr |, and the shunt rate (K) is negative, the inverter output current (Ii) is shunted to the compensation current (Ia) via the AF connection point (M). In doing so, the neural network is commanded so that the phase inversion of the compensation current (Ia) does not occur with respect to the inverter output current command value (Ir), and the output control of the inverter output current (Ii) can be performed, thereby obtaining a normal compensation effect. .
[0018]
On the other hand, the equation (a) of the conventional teacher signal does not consider the shunt from the inverter output current (Ii) to the compensation current (Ia), and commands the neural network, so that the compensation current (Ia) is calculated from the inverter output current (Ii). ) If phase reversal occurs due to shunting, it cannot be avoided.
[0019]
【The invention's effect】
According to the present invention, in the injection circuit type AF control having the neural network control, the shunting from the inverter output current to the compensation current is considered by adding the shunting ratio of the compensation current to the inverter output current to the teacher signal of the neural network. Since the neural network is commanded and the inverter output current command value is adjusted, when the inverter output current is shunted to the compensation current in AF, the phase inversion of the compensation current does not occur with respect to the inverter output current command value. In this way, the output current of the inverter can be controlled, and a normal compensation effect can be obtained.
[Brief description of the drawings]
FIG. 1 is a general connection system circuit diagram of an injection circuit type AF.
[Explanation of symbols]
2 system bus 3 AF
4 Load 5 Injection circuit section 6 High-frequency inverter 7 Calculation section M AF connection point C Capacitor R Reactor N Connection point between the capacitor and reactor in the injection circuit section

Claims (1)

An injection circuit unit in which a capacitor and a reactor are connected in series and connected to a system bus, a high-frequency inverter that injects an inverter output current at a connection point between the capacitor and the reactor of the injection circuit unit, and a harmonic detected from the bus voltage A calculation bus that calculates an inverter output current command value that outputs an optimum compensation current against system fluctuations by adjusting means configured by neural network control that receives a voltage component, and inputs the inverter output current command value to the inverter; Injecting a compensation current to control the active filter with injection circuit that compensates and reduces harmonic voltage distortion
The neural network control teacher signal
| Id | = | {(0−Vn) / (K · Zt)} + Iar | − | Iar |
(However, | Id |: Teacher signal, Vn: Harmonic voltage component at active filter connection point, Iar: Compensation current injection reference signal, K: Compensation current shunt ratio with respect to inverter output current, Zt: Active filter connection at start-up A control method for an active filter, characterized in that a neural network is commanded in consideration of a shunt from an inverter output current to a compensation current.

Images