CN101719732A - five-level svpwm controller - Google Patents

Abstract

The invention relates to a five-level SVPWM controller, which mainly includes: a sector discrimination module, a sector normalization module, a new coordinate calculation module, an area discrimination module, an action time calculation module, a pulse time segment selection module, and a switch state generation module , Trigger pulse generation module. Each module described in the five-level SVPWM controller uses different neural network structures to complete its functions, and is integrated on an FPGA chip EP1C6T144C for implementation. Its advantages are: using a new coordinate system to judge the area where the reference voltage vector is located, simplifying the area judgment method, and can be easily extended to the SVPWM modulation of more level inverters; and the five-level SVPWM controller is integrated into the Realized on an FPGA chip, it provides a high-performance dedicated SVPWM controller for high-voltage large-capacity power conversion and AC motor speed regulation.

Description

Five-level SVPWM controller

technical field

The invention relates to a SVPWM controller, in particular to a five-level SVPWM controller.

Background technique

In recent years, diode-clamped multilevel inverters have been widely used in high-voltage and large-capacity power conversion and AC motor speed regulation. Space voltage vector modulation (SVPWM) is one of the control technologies for diode-clamped multilevel inverters. It has become the most researched technology due to its small output harmonic content, high voltage utilization and easy digital realization. , The most widely used control technology. Two-level and three-level space vector modulation controllers are more convenient to implement, but with the increase of the number of levels, the geometric progression of space vector increases, which brings great difficulties to the realization of space vector modulation. At present, most of the space vector implementation algorithms for multi-level inverters are based on two-level and three-level space vector modulation techniques, and general algorithms for multi-level inverters are generally given. The specific algorithm of the space vector of the flat inverter and how to use each specific switch vector, switch state and other issues are carefully studied.

Contents of the invention

The purpose of the present invention is to overcome the deficiencies in the prior art. Taking the five-level inverter as an example, a five-level SVPWM controller is provided, which uses a new coordinate system to judge the area where the reference voltage vector is located, and simplifies the area The judging method can be easily extended to the SVPWM modulation of more level inverters, and the five-level SVPWM controller can be integrated into an FPGA chip to provide a high-voltage high-capacity power conversion and AC motor speed regulation. performance of a dedicated SVPWM controller.

According to the technical solution provided by the present invention, the five-level SVPWM controller includes: a sector discrimination module, a sector normalization module, a new coordinate calculation module, an area discrimination module, an action time calculation module, a pulse time period selection module, a switch A state generation module and a trigger pulse generation module;

The sector discrimination module judges which of the 6 sectors the synthetic reference voltage space vector is located in according to the reference voltage vectors u _α ^* and u _β ^* in the two-phase Cartesian stationary coordinate system; according to the symmetry of the 6 sectors, Use the sector normalization module to perform position normalized rotation on the reference voltage vector to obtain u _α1 ^* , u _β1 ^* , so that u _α1 ^* , u _β1 ^* are located in the first sector; the new coordinate calculation module adopts X1-X2-X3 coordinate system, coordinate calculation of u _α1 ^* and u _β1 ^* , and calculate the distance of the synthetic reference voltage space vector u ^* of u _α1 ^* and u _β1 ^* relative to the three coordinate axes of X1-X2-X3 ; The area discrimination module judges which of the 16 small triangular areas the synthetic reference voltage space vector u ^* is located in according to the distance of the synthetic reference voltage space vector u ^* relative to the three coordinate axes of X1-X2-X3; the synthetic reference voltage The space vector u ^* is synthesized by the action of three basic voltage _space vectors corresponding to the three vertices of the ^triangular area where it is located, and the action ^time calculation module calculates _three The action time of a basic voltage space vector; a pulse waveform of a five-level SVPWM is composed of 25 time periods, and the switching states of the three bridge arms of the inverter corresponding to different time periods in different regions are different, and the pulse time The segment selection module then calculates which time period the current moment is in; the switch state generation module determines the switch states of the three bridge arms of the inverter according to the time period, sector number, and area number; The switching status of the three bridge arms outputs 12 trigger pulses, and the generated 12 trigger pulses can be sent to the peripheral drive circuit to generate 24 complementary pulses with dead zone delay, which are respectively connected to the diode-clamped five-level inverter The trigger terminals of the 24 switch tubes;

The X1-X2-X3 coordinate system means that the X1 axis coincides with the α axis of the two-phase stationary coordinate system, the X2 axis is the α axis rotated 60° counterclockwise, and the X3 axis is the X2 axis rotated 60° counterclockwise.

]，阈值为[0；0]；输出层与隐含层的权值为[

]，阈值为[0；0；0]，所有节点的激活函数都为线性函数；所述神经网络的输出除以

并向上取整，得到隐含区域信息的3个整数，作为所述区域判别模块的输入信号。The new coordinate calculation module is realized by a neural network with a 2-2-3 structure, and the weights of the hidden layer and the input layer of the neural network are [1, 0; 1/2,

], the threshold is [0; 0]; the weight of the output layer and the hidden layer is [

], the threshold is [0; 0; 0], the activation functions of all nodes are linear functions; the output of the neural network is divided by

And round up to obtain 3 integers of hidden area information, which are used as the input signal of the area discrimination module.

并向上取整后得到的3个整数；所述神经网络的权值和阈值由离线训练得到，训练样本为16对数据，对应16个小三角形区域。The region discrimination module is realized by a BP neural network with a 3-5-1 structure, and the input of the BP neural network is the synthetic reference voltage space vector u ^* divided by the distance of the three coordinate axes of X1-X2-X3

And the 3 integers obtained after rounding up; the weights and thresholds of the neural network are obtained by off-line training, and the training samples are 16 pairs of data, corresponding to 16 small triangular areas.

The sector discrimination module, sector normalization module, new coordinate calculation module, area discrimination module, action time calculation module, pulse time segment selection module, switch state generation module and trigger pulse generation module are all implemented on the FPGA chip EP1C6T144C, The input quantities on EP1C6T144C are two 16-bit digital quantities corresponding to the reference voltage u _α ^* and u _β ^* in the two-phase stationary coordinate system, and the output is 12 PWM pulse waveforms. The input and output signals occupy 44 of EP1C6T144C. pins.

The present invention has the advantages of: adopting a new coordinate system to judge the region where the reference voltage vector is located, simplifying the method for judging the region, and conveniently extending to SVPWM modulation of more level inverters; using different neural network structures to complete all Describe the functions of each module, make full use of the parallel structure of the neural network, greatly simplify the calculation amount during work, and improve the rapidity; the five-level SVPWM controller is integrated into an FPGA chip EP1C6T144C to achieve The processing mechanism enables the parallel processing capability of each neural network to be fully reflected, and provides a high-performance dedicated SVPWM controller for high-voltage and large-capacity power conversion and AC motor speed regulation, reducing the calculation burden of the CPU in the system, and the CPU in the system. The interface is flexible and convenient.

Description of drawings

Figure 1 is a structural block diagram of a five-level SVPWM controller;

Figure 2 is a schematic diagram of sector distribution;

Fig. 3 is the realization structure diagram of the neural network of the sector discrimination module;

Fig. 4 is the realization structural diagram of sector normalization module neural network;

Fig. 5 is a schematic diagram of new coordinate calculation;

6 is a schematic diagram of reference voltage vector projection;

Fig. 7 is a new coordinate calculation module neural network realization structure diagram;

Fig. 8 is the realization structure diagram of the neural network of the region discrimination module;

Fig. 9 is the realization structure diagram of the neural network of the action time calculation module;

Fig. 10 is the realization structure diagram of the neural network of the pulse time segment selection module;

Fig. 11 is a switch state generating module neural network realization structure diagram;

Figure 12 is a structural diagram of the neural network implementation of the trigger pulse generation module.

Detailed ways

The specific embodiments of the present invention will be further described below in conjunction with the accompanying drawings.

As shown in Figure 1: the present invention mainly includes: a sector discrimination module, a sector normalization module, a new coordinate calculation module, an area discrimination module, an action time calculation module, a pulse time segment selection module, a switch state generation module, and a trigger pulse generation module.

1；1，0]，阈值为[0；0；0]；以第一隐含层的输出作为输入，与第二隐含层及输出层构成的网络为BP网络，采用离线训练获得其权值和阈值，训练样本为6对数据对应6个扇区。网络的输出经过向上取整后输出一个整数，其范围为1～6，分别对应I-VI扇区，如图3所示。图3中：I-VI这6个扇区把平面分成6等份，各扇区的边界分别是2个基本电压空间矢量。As shown in Figure 2: Neural network NN1 is used to determine which of the six sectors the synthetic reference voltage space vector is located in according to the reference voltages u _α ^* and u _β ^* in the two-phase stationary coordinate system. The structure of the NN1 is 2-3-4-1 (2-3-4-1 represents the layer structure of the neural network, which is used below), wherein the activation function of the first hidden layer is the hard limit Amplitude function, the upper limit is 1, the lower limit is 0, the weight of the first hidden layer and the input layer is [ 1;

1; 1, 0], the threshold value is [0; 0; 0]; the output of the first hidden layer is used as input, and the network composed of the second hidden layer and the output layer is a BP network, and its weight is obtained by offline training. value and threshold, the training samples are 6 pairs of data corresponding to 6 sectors. The output of the network outputs an integer after being rounded up, and its range is 1 to 6, corresponding to I-VI sectors respectively, as shown in FIG. 3 . In Fig. 3: the six sectors of I-VI divide the plane into six equal parts, and the boundaries of each sector are two basic voltage space vectors.

The sector normalization module mainly realizes the rotation of the vectors in sectors II-VI to sector I, so as to judge the area and calculate the action time of the basic voltage space vector, etc. The sector normalization module is realized by the neural network NN2 with the 1-3-4 structure and the neural network NN3 with the 2-2 structure as shown in FIG. 4 . In Figure 4: NN2 is a BP network, its input is the sector number, and its output is the rotation transformation coefficient from different sectors to sector I. The weight and threshold of NN2 are obtained by offline training, and the training samples are composed of sector numbers and transformation coefficients 6 pairs of data corresponding to 6 sectors; the output of NN2 is used as the weight of NN3, the input u _α ^* and u _β ^* of NN3 are the original reference voltage vector, and the output u _α1 ^* and u _β1 ^* are the normalized reference Voltage vector.

As shown in Figure 5: Construct a new stationary coordinate system, X1-X2-X3 coordinate system, where the X1 axis coincides with the α-axis of the two-phase rectangular stationary coordinate system, the X2 axis is the α-axis rotated 60° counterclockwise, and the X3 The axis is the X2 axis rotated 60° counterclockwise. Connect the three fixed points of the two voltage space vectors on the boundary of sector I to form a large triangle, divide the large triangle of sector I into 16 equal parts, corresponding to 1-16 triangular areas respectively; V1～V14 are inverse The basic voltage space vector corresponding to the different states of the transformer switch tube; the amplitude of the synthetic reference voltage vector u ^* is U ^* , and the angle between it and the α axis is θ; from the end point of the synthetic reference voltage vector u ^* to X1, X2, X3 axis as a vertical line, calculate the distance UX1, UX2, UX3 as follows:

In order to avoid the calculation of trigonometric functions when calculating the distance, on the one hand, the synthetic reference voltage vector is projected to the X1 and X2 axes to obtain n1 and n2, as shown in Figure 6. In Figure 6:

Therefore, according to formula (2), we can get:

$\left\{\begin{array}{c}\mathrm{<span\; class="notranslate">\; sin</span>}\mathrm{<span\; class="notranslate">\; \&theta;</span>}<span\; class="notranslate">\; =</span>\mathrm{<span\; class="notranslate">\; 1</span>}<span\; class="notranslate">\; /</span><span\; class="notranslate">\; (</span>\sqrt{\mathrm{<span\; class="notranslate">\; 3</span>}}{\mathrm{<span\; class="notranslate">\; u</span>}}^{<span\; class="notranslate">\; *</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; 2</span>}{\mathrm{<span\; class="notranslate">\; no</span>}}_{\mathrm{<span\; class="notranslate">\; 2</span>}}<span\; class="notranslate">\; -</span>{\mathrm{<span\; class="notranslate">\; no</span>}}_{\mathrm{<span\; class="notranslate">\; 1</span>}}<span\; class="notranslate">\; )</span>\\ \mathrm{<span\; class="notranslate">\; sin</span>}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; \&theta;</span>}<span\; class="notranslate">\; +</span>\mathrm{<span\; class="notranslate">\; \&pi;</span>}<span\; class="notranslate">\; /</span>\mathrm{<span\; class="notranslate">\; 3</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; =</span>\mathrm{<span\; class="notranslate">\; 1</span>}<span\; class="notranslate">\; /</span><span\; class="notranslate">\; (</span>\sqrt{\mathrm{<span\; class="notranslate">\; 3</span>}}{\mathrm{<span\; class="notranslate">\; u</span>}}^{<span\; class="notranslate">\; *</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; (</span>{\mathrm{<span\; class="notranslate">\; <figure-callout\; id="1"\; label="no"\; filenames="H2009102323393E00012.png"\; state="\{\{state\}\}">no</figure-callout></span>}}_{\mathrm{<span\; class="notranslate">\; 1</span>}}<span\; class="notranslate">\; +</span>{\mathrm{<span\; class="notranslate">\; no</span>}}_{\mathrm{<span\; class="notranslate">\; 2</span>}}<span\; class="notranslate">\; )</span>\\ \mathrm{<span\; class="notranslate">\; sin</span>}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; \&theta;</span>}<span\; class="notranslate">\; -</span>\mathrm{<span\; class="notranslate">\; \&pi;</span>}<span\; class="notranslate">\; /</span>\mathrm{<span\; class="notranslate">\; 3</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; =</span>\mathrm{<span\; class="notranslate">\; 1</span>}<span\; class="notranslate">\; /</span><span\; class="notranslate">\; (</span>\sqrt{\mathrm{<span\; class="notranslate">\; 3</span>}}{\mathrm{<span\; class="notranslate">\; u</span>}}^{<span\; class="notranslate">\; *</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; 2</span>}{\mathrm{<span\; class="notranslate">\; no</span>}}_{\mathrm{<span\; class="notranslate">\; 1</span>}}<span\; class="notranslate">\; -</span>{\mathrm{<span\; class="notranslate">\; no</span>}}_{\mathrm{<span\; class="notranslate">\; 2</span>}}<span\; class="notranslate">\; )</span>\end{array}\right.<span\; class="notranslate">\; -</span><span\; class="notranslate">\; -</span><span\; class="notranslate">\; -</span><span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; 3</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; .</span>$

On the other hand, the reference voltages u _α1 ^* and u _β1 ^* expressed in the two-phase stationary coordinate system are respectively projected onto the X1 and X2 axes, and the projected values are obviously the same as n1 and n2, namely:

$\left\{\begin{array}{c}\mathrm{<span\; class="notranslate">\; <figure-callout\; id="1"\; label="no"\; filenames="H2009102323393E00012.png"\; state="\{\{state\}\}">no</figure-callout></span>}\mathrm{<span\; class="notranslate">\; 1</span>}<span\; class="notranslate">\; =</span>{\mathrm{<span\; class="notranslate">\; u</span>}}_{\mathrm{<span\; class="notranslate">\; \&alpha;</span>}\mathrm{<span\; class="notranslate">\; 1</span>}}^{<span\; class="notranslate">\; *</span>}\\ \mathrm{<span\; class="notranslate">\; no</span>}\mathrm{<span\; class="notranslate">\; 2</span>}<span\; class="notranslate">\; =</span>{\mathrm{<span\; class="notranslate">\; u</span>}}_{\mathrm{<span\; class="notranslate">\; \&alpha;</span>}\mathrm{<span\; class="notranslate">\; 1</span>}}^{<span\; class="notranslate">\; *</span>}<span\; class="notranslate">\; /</span>\mathrm{<span\; class="notranslate">\; 2</span>}<span\; class="notranslate">\; +</span>\sqrt{\mathrm{<span\; class="notranslate">\; 3</span>}}{\mathrm{<span\; class="notranslate">\; u</span>}}_{\mathrm{<span\; class="notranslate">\; \&beta;</span>}\mathrm{<span\; class="notranslate">\; 1</span>}}^{<span\; class="notranslate">\; *</span>}<span\; class="notranslate">\; /</span>\mathrm{<span\; class="notranslate">\; 2</span>}\end{array}\right.<span\; class="notranslate">\; -</span><span\; class="notranslate">\; -</span><span\; class="notranslate">\; -</span><span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; 4</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; .</span>$

并向上取整，得到3个整数，分别代表参考电压矢量与X1-X2-X3三个轴的相对位置信息，输出的整数取值范围为1～4，V_dc为逆变器直流侧电压。新坐标计算模块用如图7所示的神经网络结构实现，图7中：NN4为2-2-3结构的神经网络，隐含层的激活函数为线性函数，隐含层与输入层的权值为[1，0；1/2，

]，阈值为[0；0]；输出层的激活函数为线性函数，输出层与隐含层的权值为[

]，阈值为[0；0；0]。Substituting the obtained trigonometric function into formula (1) can get the distance value, divide UX1, UX2, UX3 by

And round up to get 3 integers, which respectively represent the relative position information of the reference voltage vector and the three axes X1-X2-X3, the output integer ranges from 1 to 4, and V _dc is the DC side voltage of the inverter. The new coordinate calculation module is implemented with the neural network structure shown in Figure 7. In Figure 7: NN4 is a neural network with a 2-2-3 structure, the activation function of the hidden layer is a linear function, and the weight of the hidden layer and the input layer Values are [1, 0; 1/2,

], the threshold is [0; 0]; the activation function of the output layer is a linear function, and the weights of the output layer and the hidden layer are [

], the threshold is [0; 0; 0].

并向上取整得到，所述神经网络NN6的输出经过向上取整后得到代表区域号的整数；所述神经网络NN6的权值和阈值由离线训练得到，训练样本为16对数据，对应16个小三角形区域。The region discrimination module adopts the BP neural network NN6 with a structure of 3-5-1 as shown in Fig. 8 to realize, among Fig. 8: UX_1, UX_2, UX_3, as the input signal of described neural network NN6, by UX1, UX2, UX3 divide by

And round up to obtain, the output of the neural network NN6 is rounded up to obtain an integer representing the area number; the weight and threshold of the neural network NN6 are obtained by off-line training, and the training samples are 16 pairs of data, corresponding to 16 small triangular area.

The present invention is mainly aimed at the undermodulation mode. The synthesized reference voltage space vector u ^* is synthesized by the action of three basic voltage space vectors corresponding to the three vertices of the triangular area where it is located. According to the area where the reference voltage vector is located, combined with volts According to the principle of second balance, the action times T _a , T _b , and T _c of the three basic voltage space vectors are calculated, and the values of trigonometric functions involved in the calculation process are replaced by formula (2), so there is no calculation of trigonometric functions. The action time generation module is realized by using the neural network structure shown in Figure 9. In Figure 9: the neural network NN7 is a BP network with a structure of 1-5-9, and the weights and thresholds of NN7 are obtained by offline training; the training samples are 16 In each area, T _a , T _b , and T _c calculate 16 pairs of data composed of n1, n2, and Ts coefficients contained in the expression, and Ts is the time of a pulse waveform, that is, the modulation cycle; the neural network NN8 is 3- 3-structure BP network, its weight is the output of NN7, and the thresholds are all 0.

A pulse waveform of five-level SVPWM is composed of 25 time periods. The switching states of the three bridge arms corresponding to different time periods in different regions are different. The pulse time period selection module calculates which time period the current moment is in, using The neural network structure shown in Figure 10 is realized. In Fig. 10: the structure of the neural network NN9 is 4-24-1, T is the output value of a timer whose period is Ts, the activation function of the hidden layer is a hard clipping function, the upper limit is 1, and the lower limit is 0; The weight value between the layer and the input layer can be calculated according to the action time coefficient of the segment composed of the pulse waveform, and the threshold value is all 0; the activation function of the output layer is a linear function, the weight value is all 1, and the threshold value is 25; the output range of NN9 is An integer ranging from 1 to 25, representing 25 time segments.

The switch state generation module determines the switch state of the three bridge arms according to the time period, sector number, and area number. Different switch states output different trigger pulses. The neural network structure shown in Figure 11 is used to realize the switch state generation module. In Figure 11: the neural network NN10 is a BP network with a 3-20-20-3 structure, the input is the sector number, area number and time period, and the output is the switch state S _a , S _b , S _c of the three bridge arms, The values of S _a , S _b , and S _c are all 4, 3, 2, 1, and 0 to represent five states, respectively. The weight and threshold of NN10 are obtained from offline training, and the training samples are 6 (number of sectors)×16 (number of regions)×25 (number of time periods)=2400 pairs of data.

The trigger pulse generation module outputs 12 trigger pulses according to the switch states of the three bridge arms, and the 12 trigger pulses can be sent to the peripheral drive circuit to generate 24 complementary pulses with dead zones, which are respectively connected to the trigger terminals of the 24 switch tubes. The trigger pulse generation module is realized by three identical neural networks, taking the A-phase bridge arm as an example, as shown in Figure 12. In Figure 12: the neural network NN12 is a BP network with a 1-4-4 structure. The weights and thresholds of NN12 are obtained by offline training. The training samples are different switch states and corresponding trigger signals to form 5 pairs of data, corresponding to the A-phase bridge. 5 states of the arm switch state S _a .

The sector discrimination module, sector normalization module, new coordinate calculation module, area discrimination module, action time calculation module, pulse time segment selection module, switch state generation module, and trigger pulse generation module are all integrated on one FPGA chip EP1C6T144C Realization, the input is two 16-bit digital quantities corresponding to the reference voltage u _α ^* and u _β ^* in the two-phase static coordinate system, and the output is 12 PWM pulse waveforms. The input and output signals occupy 44 pins of EP1C6T144C. foot.

The present invention has the following characteristics: a new coordinate system is used to judge the region where the reference voltage vector is located, the method for judging the region is simplified, and it can be easily extended to SVPWM modulation of more level inverters; different neural network structures are used to complete all Describe the functions of each module, make full use of the parallel structure of the neural network, greatly simplify the calculation amount when the SVPWM controller is working, and improve the rapidity; the five-level SVPWM controller is integrated into an FPGA chip EP1C6T144C for realization, using The parallel processing mechanism of FPGA enables the parallel processing capability of each neural network to be fully reflected, and provides a high-performance dedicated controller for high-voltage and large-capacity power conversion and AC motor speed control systems, reducing the calculation burden of the CPU in the system, and the system The interface of the CPU is flexible and convenient.

Claims (4)

1. five-level SVPWM controller is characterized in that comprising: sector discrimination module, sector normalizing module, new coordinate Calculation module, area judging module, action time, computing module, burst length section were selected module, on off state generation module, trigger impulse generation module;

Described sector discrimination module is according to the reference voltage vector u under the two-phase right angle rest frame _α ^*, u _β ^*Judge which of 6 sectors the synthesized reference space vector of voltage be arranged in; According to the symmetry of 6 sectors, adopt described sector normalizing module that reference voltage vector is carried out the place normalization rotation and obtain u _{α 1} ^*, u _{β 1} ^*, make u _{α 1} ^*, u _{β 1} ^*Be positioned at first sector; Described new coordinate Calculation module adopts the X1-X2-X3 coordinate system, to u _{α 1} ^*, u _{β 1} ^*Carry out coordinate Calculation, calculate u _{α 1} ^*, u _{β 1} ^*Synthesized reference space vector of voltage u ^*Distance with respect to three reference axis of X1-X2-X3; Described area judging module is according to synthesized reference space vector of voltage u ^*With respect to the distance of three reference axis of X1-X2-X3, judge synthesized reference space vector of voltage u ^*Be arranged in which of 16 little delta-shaped regions; Synthesized reference space vector of voltage u ^*Synthetic by three summit pairing three fundamental voltage space vectors effects of its residing delta-shaped region, described action time, computing module was then according to regional value and reference voltage vector u _{α 1} ^*, u _{β 1} ^*Calculate the action time of three fundamental voltage space vectors; Impulse waveform of five-level SVPWM was made of 25 time periods, and the on off state of three brachium pontis of the pairing inverter of different time sections in the zones of different is different, and described burst length section selection module is then calculated current time and is in which time period; Described on off state generation module is according to the on off state of time period, sector number, three brachium pontis of regional number decision inverter; Described trigger impulse generation module is exported 12 road trigger impulses according to the on off state of three brachium pontis of inverter, 12 road trigger impulses that generate can be given the pulse that peripheral drive circuit generates 24 tunnel complementary band dead band time-delays, receive the trigger end of 24 switching tubes of diode clamp formula five-electrical level inverter respectively;

Described X1-X2-X3 coordinate system is meant that the X1 axle overlaps with the α axle of two-phase rest frame, and the X2 axle is 60 ° of α axle inverse clock rotations, and the X3 axle is 60 ° of X2 axle inverse clock rotations.

2. five-level SVPWM controller according to claim 1 is characterized in that described new coordinate Calculation module adopts the neural net of 2-2-3 structure to realize that the weights of neural net hidden layer and input layer are [1,0; 1/2, ], threshold value is [0; 0]; The weights of output layer and hidden layer be [

], threshold value is [0; 0; 0], the activation primitive of all nodes all is a linear function; The output of described neural net divided by And round up, obtain 3 integers of inclusion region information, as the input signal of described area judging module.

3. five-level SVPWM controller according to claim 1 is characterized in that described area judging module adopts the BP neural net of 3-5-1 structure to realize, described BP neural net be input as synthesized reference space vector of voltage u ^*With respect to the distance of three reference axis of described X1-X2-X3 divided by And 3 integers that obtain after rounding up; The weights and the threshold value of described neural net are obtained by off-line training, and training sample is 16 pairs of data, corresponding 16 little delta-shaped regions.

4. five-level SVPWM controller according to claim 1, it is characterized in that described sector discrimination module, sector normalizing module, new coordinate Calculation module, area judging module, action time computing module, burst length section select module, on off state generation module and trigger impulse generation module to realize on fpga chip EP1C6T144C that all the input variable on the EP1C6T144C is the reference voltage u under the two-phase rest frame _α ^*, u _β ^*Pairing two 16 digital quantity is output as 12 road pwm pulse waveforms, and input, output signal take 44 pins of EP1C6T144C altogether.

Images