CN104967353A

CN104967353A - An off-grid wind power inverter

Info

Publication number: CN104967353A
Application number: CN201510402092.0A
Authority: CN
Inventors: 刘卫亮; 陈文颖; 刘长良; 林永君; 马良玉
Original assignee: North China Electric Power University
Current assignee: North China Electric Power University; State Grid Qinghai Electric Power Co Ltd; Electric Power Research Institute of State Grid Qinghai Electric Power Co Ltd
Priority date: 2015-07-08
Filing date: 2015-07-08
Publication date: 2015-10-07
Anticipated expiration: 2035-07-08
Also published as: CN104967353B

Abstract

An off-grid wind power inverter, which consists of an MPU controller, a rectifier, an energy storage capacitor, a flyback converter, a power frequency inverter bridge and an LCL filter sequentially connected between the wind generator and the load, The MPU controller collects the energy storage capacitor voltage and the inverter output voltage through the first voltage sensor and the second voltage sensor respectively, collects the power frequency inverter bridge side inductor current of the LCL filter through the current sensor, and passes two drive The module controls the flyback converter and the power frequency inverter bridge respectively. The invention utilizes a composite controller composed of a neural network inverse model and a PI controller to control the output voltage of the inverter. Neural network inverse model control is mainly used near the zero-crossing point to give full play to its advantages of fast response; PI control is mainly used near the grid voltage peak to give full play to its advantages of good steady-state performance, thereby effectively improving the distance from the grid. Power quality of grid-type wind power generation system.

Description

An off-grid wind power inverter

技术领域technical field

本发明涉及一种基于复合控制的离网型风力发电逆变装置，属于发电技术领域。The invention relates to an off-grid wind power inverter based on compound control, which belongs to the technical field of power generation.

背景技术Background technique

风力发电因具有储量大、清洁可再生等特性，已成为当前新能源发电的主要方向之一。离网型风力发电系统具有成本低、安装灵活的特点，广泛应用于偏远山区、岛屿等大电网无法提供电力的地区，并多采用小型永磁直驱电机加逆变器的简单结构。与并网型风力发电系统不同的是，离网型风力发电系统必须实施有效的逆变器输出电压控制，以保证系统的稳定运行。Due to its large reserves, clean and renewable characteristics, wind power has become one of the main directions of new energy power generation. The off-grid wind power generation system has the characteristics of low cost and flexible installation. It is widely used in remote mountainous areas, islands and other areas where large power grids cannot provide power. It mostly uses a simple structure of small permanent magnet direct drive motors and inverters. Different from the grid-connected wind power generation system, the off-grid wind power generation system must implement effective inverter output voltage control to ensure the stable operation of the system.

现有的离网型风机逆变器多采用简单的PI控制器，存在输出电压谐波含量高、抗扰动能力差的缺点。通过配置多级滤波单元可以在一定程度上抑制谐波，但是该方法使得系统输出阻抗阶次过高，容易导致系统不稳定。现代社会对电能质量提出的要求越来越高，必须不断提高离网型风机逆变器的控制水平。Most of the existing off-grid wind turbine inverters use simple PI controllers, which have the disadvantages of high harmonic content in the output voltage and poor anti-disturbance ability. Harmonics can be suppressed to a certain extent by configuring multi-stage filter units, but this method makes the system output impedance order too high, which easily leads to system instability. Modern society has higher and higher requirements for power quality, and it is necessary to continuously improve the control level of off-grid wind turbine inverters.

发明内容Contents of the invention

本发明的目的在于针对现有技术之弊端，提供一种离网型风力发电逆变器，以有效提高离网型风力发电系统的电能质量。The object of the present invention is to provide an off-grid wind power inverter to effectively improve the power quality of the off-grid wind power generation system against the disadvantages of the prior art.

本发明所述问题是以下述技术方案实现的：Problem described in the present invention is realized with following technical scheme:

一种离网型风力发电逆变器，构成中包括MPU控制器以及依次连接于风力发电机与负载之间的整流器、储能电容、反激变换器、工频逆变桥和LCL滤波器，所述MPU控制器通过第一电压传感器和第二电压传感器分别采集储能电容电压和逆变器输出电压，通过电流传感器采集LCL滤波器的工频逆变桥侧电感电流，并通过两个驱动模块分别控制反激变换器和工频逆变桥，所述MPU控制器按以下方式运作：An off-grid wind power inverter, which consists of an MPU controller, a rectifier, an energy storage capacitor, a flyback converter, a power frequency inverter bridge and an LCL filter sequentially connected between the wind generator and the load, The MPU controller collects the energy storage capacitor voltage and the inverter output voltage through the first voltage sensor and the second voltage sensor respectively, collects the power frequency inverter bridge side inductor current of the LCL filter through the current sensor, and passes two drive The modules respectively control the flyback converter and the power frequency inverter bridge, and the MPU controller operates in the following manner:

①通过第一电压传感器采集储能电容电压V_d，通过第二电压传感器采集逆变器输出电压V_o，通过电流传感器采集LCL滤波器的工频逆变桥侧电感电流i_L；以逆变器作为对象，建立并训练相应的神经网络逆模型，具体如下：①Use the first voltage sensor to collect the energy storage capacitor voltage V _d , use the second voltage sensor to collect the inverter output voltage V _o , and use the current sensor to collect the power frequency inverter bridge side inductor current i _L of the LCL filter; The device is used as the object, and the corresponding neural network inverse model is established and trained, as follows:

a.选取三层BP神经网络建立系统的逆模型，其中，输入层神经元节点数为9个，隐含层神经元节点数为10个，输出层神经元节点数为1个，隐含层神经元转移函数使用双曲正切函数，输出层神经元转移函数使用S型函数；a. Select a three-layer BP neural network to establish the inverse model of the system, in which the number of neuron nodes in the input layer is 9, the number of neuron nodes in the hidden layer is 10, the number of neuron nodes in the output layer is 1, and the number of neuron nodes in the hidden layer is 1. The neuron transfer function uses the hyperbolic tangent function, and the output layer neuron transfer function uses the S-type function;

b.在逆变器上采集运行数据；b. Collect running data on the inverter;

使得风机分别处于额定风速、80％额定风速、60％额定风速、40％额定风速和20％额定风速的条件下，每种风速条件下通过改变负载使得逆变器分别工作于额定功率、80％额定功率、60％额定功率、40％额定功率和20％额定功率的工况，共计25种工况，每种工况下均利用单回路PI控制器进行逆变器输出电压控制，并连续采集N组储能电容电压V_d、逆变器输出电压V_o、反激变换器PWM信号占空比D、LCL滤波器的工频逆变桥侧电感电流i_L，共构成25N组运行数据；Make the fan under the conditions of rated wind speed, 80% of rated wind speed, 60% of rated wind speed, 40% of rated wind speed and 20% of rated wind speed, and make the inverter work at rated power, 80% of rated power and Rated power, 60% rated power, 40% rated power and 20% rated power working conditions, a total of 25 working conditions, each working condition uses a single-loop PI controller to control the inverter output voltage, and continuously collects N groups of energy storage capacitor voltage V _d , inverter output voltage V _o , flyback converter PWM signal duty ratio D, and LCL filter's power frequency inverter bridge side inductance current i _L constitute 25N groups of operating data in total;

c.利用采集的运行数据构造样本；c. Use the collected operating data to construct samples;

对于k时刻，所构造的样本为{V_o(k+1)，V_o(k)，V_o(k-1)，V_o(k-2)，i_L(k)，i_L(k-1)，D(k-1)，D(k-2)，V_d(k)，D(k)}，其中，V_o(k+1)为k+1时刻的逆变器输出电压，V_o(k)为k时刻的逆变器输出电压，V_o(k-1)为k-1时刻的逆变器输出电压，V_o(k-2)为k-2时刻的逆变器输出电压，i_L(k)为k时刻的LCL滤波器的工频逆变桥侧电感电流，i_L(k-1)为k-1时刻的LCL滤波器的工频逆变桥侧电感电流，D(k-1)为k-1时刻的反激变换器PWM信号占空比，D(k-2)为k-2时刻的反激变换器PWM信号占空比，V_d(k)为k时刻的储能电容电压，D(k)为k时刻的反激变换器PWM信号占空比，25N组运行数据共构成25N个样本；For time k, the constructed samples are {V _o (k+1), V _o (k), V _o (k-1), V _o (k-2), i _L (k), i _L (k -1), D(k-1), D(k-2), V _d (k), D(k)}, where V _o (k+1) is the inverter output voltage at time k+1 , V _o (k) is the inverter output voltage at k time, V _o (k-1) is the inverter output voltage at k-1 time, V _o (k-2) is the inverter at k-2 time i _L (k) is the inductance current of the power frequency inverter bridge side of the LCL filter at time k, i _L (k-1) is the power frequency inverter bridge side inductance of the LCL filter at time k-1 current, D(k-1) is the duty cycle of the flyback converter PWM signal at time k-1, D(k-2) is the duty cycle of the flyback converter PWM signal at time k-2, V _d (k ) is the energy storage capacitor voltage at time k, D(k) is the duty cycle of the flyback converter PWM signal at time k, and 25N groups of operating data constitute 25N samples in total;

d.随机抽取25N个样本数据中的20N个作为训练样本，其余5N个作为检验样本，训练建立的三层BP神经网络；训练过程中，对于k时刻的样本，取BP神经网络的第一个输入端为k时刻的逆变器输出电压V_o(k)，第二个输入端为k-1时刻的逆变器输出电压V_o(k-1)，第三个输入端为k-2时刻的逆变器输出电压V_o(k-2)，第四个输入端为k时刻的LCL滤波器的工频逆变桥侧电感电流i_L(k)，第五个输入端为k-1时刻的LCL滤波器的工频逆变桥侧电感电流i_L(k-1)，第六个输入端为k+1时刻的逆变器输出电压V_o(k+1)，第七个输入端为k时刻的储能电容电压V_d(k)，第八个输入端为k-1时刻的反激变换器PWM信号占空比D(k-1)，第九个输入端为k-2时刻的反激变换器PWM信号占空比D(k-2)，BP神经网络的输出为k时刻的反激变换器PWM信号占空比D(k)；d. Randomly select 20N of the 25N sample data as training samples, and the remaining 5N as test samples, and train the established three-layer BP neural network; during the training process, for the samples at time k, take the first one of the BP neural network The input terminal is the inverter output voltage V _o (k) at time k, the second input terminal is the inverter output voltage V _o (k-1) at time k-1, and the third input terminal is k-2 The inverter output voltage V _o (k-2) at time, the fourth input terminal is the power frequency inverter bridge side inductor current i _L (k) of the LCL filter at time k, and the fifth input terminal is k- The inductor current i _L (k-1) of the power frequency inverter bridge side of the LCL filter at time 1, the sixth input terminal is the inverter output voltage V _o (k+1) at time k+1, the seventh The input terminal is the energy storage capacitor voltage V _d (k) at time k, the eighth input terminal is the flyback converter PWM signal duty cycle D(k-1) at time k-1, and the ninth input terminal is k The duty cycle D(k-2) of the flyback converter PWM signal at time -2, and the output of the BP neural network is the duty cycle D(k) of the flyback converter PWM signal at time k;

②利用训练好的BP神经网络逆模型与PI控制器对逆变器进行控制，具体步骤为：② Use the trained BP neural network inverse model and PI controller to control the inverter. The specific steps are:

a.利用训练好的BP神经网络逆模型进行反激变换器PWM信号占空比的计算，对于当前k时刻，取第一个输入端为k时刻的逆变器输出电压V_o(k)，第二个输入端为k-1时刻的逆变器输出电压V_o(k-1)，第三个输入端为k-2时刻的逆变器输出电压V_o(k-2)，第四个输入端为k时刻的LCL滤波器的工频逆变桥侧电感电流i_L(k)，第五个输入端为k-1时刻的LCL滤波器的工频逆变桥侧电感电流i_L(k-1)，第六个输入端为k+1时刻的逆变器输出电压设定值V^* _o(k+1)，第七个输入端为k时刻的储能电容电压V_d(k)，第八个输入端为k-1时刻的反激变换器PWM信号占空比D(k-1)，第九个输入端为k-2时刻的反激变换器PWM信号占空比D(k-2)，得出BP神经网络的输出为d(k)；a. Use the trained BP neural network inverse model to calculate the duty cycle of the flyback converter PWM signal. For the current k moment, take the first input terminal as the inverter output voltage V _o (k) at the moment k, The second input terminal is the inverter output voltage V _o (k-1) at time k-1, the third input terminal is the inverter output voltage V _o (k-2) at time k-2, and the fourth input terminal is The first input terminal is the power frequency inverter bridge side inductor current i _L (k) of the LCL filter at time k, and the fifth input terminal is the power frequency inverter bridge side inductor current i _L of the LCL filter at time k-1 (k-1), the sixth input terminal is the set value of inverter output voltage V ^* _o (k+1) at time k+1, and the seventh input terminal is the energy storage capacitor voltage V _d ( k), the eighth input terminal is the duty cycle D(k-1) of the flyback converter PWM signal at time k-1, and the ninth input terminal is the duty cycle of the flyback converter PWM signal at time k-2 D(k-2), the output of the BP neural network is d(k);

b.将下一时刻逆变器输出电压设定值V^* _o(k+1)与当前k时刻逆变器输出电压V_o(k)的偏差送入PI控制器，得到闭环控制量d_C(k)，即b. Send the deviation between the inverter output voltage set value V ^* _o (k+1) at the next moment and the inverter output voltage V _o (k) at the current k moment to the PI controller to obtain the closed-loop control value d _C (k), namely

${d d}_{C C} ((k k)) = = {K K}_{p p} e e ((k k)) + + {K K}_{i i} {Σ Σ}_{j j = = 00}^{k k} e e ((k k))$

其中：e(k)＝V^* _o(k+1)-V_o(k)，K_p表示比例系数，K_i表示积分系数；Wherein: e (k)=V ^* _o (k+1)-V _o (k), K _p represents proportional coefficient, and K _i represents integral coefficient;

c.将神经网络逆模型与PI控制器构成复合控制器，其输出为神经网络逆模型的输出d(k)和PI控制器的输出d_C(k)按照比例K:(1-K)叠加，作为最终的反激变换器PWM信号占空比D(k)，即：c. The neural network inverse model and the PI controller constitute a composite controller, and its output is the output d(k) of the neural network inverse model and the output d _C (k) of the PI controller are superimposed according to the ratio K:(1-K) , as the final flyback converter PWM signal duty cycle D(k), namely:

D(k)＝Kd(k)+(1-K)d_C(k)D(k)＝Kd(k)+(1-K)d _C (k)

其中，K的取值为：Among them, the value of K is:

$K K = = \{\begin{matrix} 0.9 0.9 & i i f f [[((- - \frac{11}{1212} π π < < θ θ < < \frac{11}{1212} π π)) o o r r ((- - π π < < θ θ < < - - \frac{1111}{1212} π π)) o o r r ((\frac{1111}{1212} π π < < θ θ < < π π))]] \\ 0.1 0.1 & i i f f [[((- - \frac{77}{1212} π π < < θ θ < < - - \frac{55}{1212} π π)) o o r r ((\frac{55}{1212} π π < < θ θ < < \frac{77}{1212} π π))]] \\ 0.5 0.5 & e e l l s the s e e \end{matrix}$

式中θ为逆变器输出电压设定值的相位。Where θ is the phase of the set value of the inverter output voltage.

上述离网型风力发电逆变器，对BP神经网络的训练采用变学习速率的误差反传算法，具体为：首先将各层神经元中的权系数和阈值初始化为(-1，+1)之间的随机量，再将输入层9个神经元输出值作为隐含层每个神经元的输入，隐含层的10个神经元输出值作为输出层单个神经元的输入，输出层单个神经元的输出值即为网络最终输出，对于第k个训练样本，计算第t次训练后网络实际输出与期望输出的误差：The above-mentioned off-grid wind power inverter adopts the error backpropagation algorithm with variable learning rate for the training of the BP neural network, specifically: firstly, the weight coefficients and thresholds of neurons in each layer are initialized to (-1, +1) The random amount between, and then the output value of 9 neurons in the input layer is used as the input of each neuron in the hidden layer, the output value of 10 neurons in the hidden layer is used as the input of a single neuron in the output layer, and the output value of a single neuron in the output layer The output value of the element is the final output of the network. For the kth training sample, calculate the actual output of the network after the tth training with the desired output The error of:

$\begin{matrix} {δ δ}_{k k}^{t t} = = (({y the y}_{k k}^{t t} - - {c c}_{k k}^{t t})) & k k = = 11,, 22,, ... ...,, 2020 N N \end{matrix}$

及20N个训练样本的平方和误差为：And the sum of squared errors of 20N training samples is:

${E E.}_{t t} = = {Σ Σ}_{k k = = 11}^{2020 N N} {(({y the y}_{k k}^{t t} - - {c c}_{k k}^{t t}))}^{22} / / = = {Σ Σ}_{k k = = 11}^{2020 N N} {(({δ δ}_{k k}^{t t}))}^{22} / / 22$

每计算完一遍，比较E_t与期望误差E_o，若E_t<E_o，则训练终止条件满足，训练结束；否则，将E_t沿连接路径进行反向传播，并逐一调整各层的权系数和阈值，直到E_t<E_o为止；然后，利用5N个检验样本对模型进行测试，计算样本均方误差ΔMSE，若有ΔMSE<Tr(Tr为某一固定阈值)，则认为所训练的BP神经网络模型满足精度要求，此时记录各个神经元的权系数和阈值；否则，重新对BP神经网络模型进行训练，直至其满足精度要求为止。After each calculation, compare E _t with the expected error E _o , if E _t < E _o , the training termination condition is satisfied, and the training ends; otherwise, E _t is backpropagated along the connection path, and the weights of each layer are adjusted one by one Coefficients and thresholds until E _t < E _o ; then, use 5N test samples to test the model and calculate the sample mean square error ΔMSE, if ΔMSE<Tr (Tr is a fixed threshold), it is considered that the trained If the BP neural network model meets the accuracy requirements, record the weight coefficients and thresholds of each neuron; otherwise, retrain the BP neural network model until it meets the accuracy requirements.

上述离网型风力发电逆变器，所述反激变换器包括变压器、第一功率管、二极管和第二电容，所述变压器的原边线圈经第一功率管接储能电容电压，其副边线圈经二极管给工频逆变桥供电，第二电容并接于工频逆变桥的直流输入端，第一功率管的栅极接第一驱动模块的输出端。In the above-mentioned off-grid wind power inverter, the flyback converter includes a transformer, a first power tube, a diode and a second capacitor, the primary side coil of the transformer is connected to the energy storage capacitor voltage through the first power tube, and its secondary The side coil supplies power to the power frequency inverter bridge through the diode, the second capacitor is connected to the DC input end of the power frequency inverter bridge in parallel, and the grid of the first power transistor is connected to the output end of the first drive module.

上述离网型风力发电逆变器，所述工频逆变桥是由四个功率管接成的全控桥，四个功率管的栅极接第二驱动模块的输出端。In the above-mentioned off-grid wind power inverter, the power frequency inverter bridge is a full-control bridge connected by four power tubes, and the grids of the four power tubes are connected to the output end of the second drive module.

上述离网型风力发电逆变器，所述LCL滤波器包括工频逆变桥侧电感、第三电容和负载侧电感，工频逆变桥侧电感一端接工频逆变桥的第一交流输出端，另外一端接电流传感器的正极，第三电容的正极接电流传感器的负极，第三电容的负极接工频逆变桥的第二交流输出端，电流传感器的测量信号输出端与MPU控制器连接，负载侧电感一端接第三电容的正极，另外一端接负载的正极。In the above-mentioned off-grid wind power inverter, the LCL filter includes a power frequency inverter bridge side inductance, a third capacitor and a load side inductance, and one end of the power frequency inverter bridge side inductance is connected to the first AC of the power frequency inverter bridge Output terminal, the other end is connected to the positive pole of the current sensor, the positive pole of the third capacitor is connected to the negative pole of the current sensor, the negative pole of the third capacitor is connected to the second AC output terminal of the power frequency inverter bridge, the measurement signal output terminal of the current sensor is connected to the MPU control One end of the load-side inductance is connected to the positive pole of the third capacitor, and the other end is connected to the positive pole of the load.

本发明利用神经网络逆模型和PI控制器构成的复合控制器来控制逆变器的输出电压。在过零点附近，逆变器输出电压的设定值变化率最大，此时以神经网络逆模型控制为主，以充分发挥其响应速度快的优点；在电网电压峰值附近，逆变器输出电压的设定值变化率最小，此时以PI控制为主，以充分发挥其稳态性能好的优点，从而有效提高了离网型风力发电系统的电能质量。The invention utilizes a composite controller composed of a neural network inverse model and a PI controller to control the output voltage of the inverter. Near the zero-crossing point, the change rate of the set value of the inverter output voltage is the largest. At this time, the neural network inverse model control is mainly used to give full play to its advantages of fast response; near the peak value of the grid voltage, the inverter output voltage The change rate of the set value is the smallest. At this time, PI control is the main method to give full play to its advantages of good steady-state performance, thereby effectively improving the power quality of the off-grid wind power generation system.

附图说明Description of drawings

下面结合附图对本发明作进一步说明。The present invention will be further described below in conjunction with accompanying drawing.

图1是风力发电逆变器装置的结构框图；Fig. 1 is a structural block diagram of a wind power inverter device;

图2是神经网络逆模型示意图；Fig. 2 is a schematic diagram of a neural network inverse model;

图3为单个神经元的结构示意图；Fig. 3 is a structural schematic diagram of a single neuron;

图4是风力发电逆变器输出电压曲线。Figure 4 is the output voltage curve of the wind power inverter.

图中各标号清单为：GS、风机，UT1、第一电压传感器，UT2、第二电压传感器，CT、电流传感器，D1、二极管，Q1～Q5、第一功率管～第五功率管，C1、储能电容，C2、第二电容，C3、第三电容，L1、工频逆变桥侧电感，L2、负载侧电感，TX1、变压器。The list of labels in the figure is: GS, fan, UT1, first voltage sensor, UT2, second voltage sensor, CT, current sensor, D1, diode, Q1~Q5, first power tube~fifth power tube, C1, Energy storage capacitor, C2, second capacitor, C3, third capacitor, L1, power frequency inverter bridge side inductance, L2, load side inductance, TX1, transformer.

文中各符号清单为：V_d为储能电容电压，V_o为逆变器输出电压，i_L为LCL滤波器工频逆变桥侧电感电流，D为反激变换器PWM信号占空比，V_d(k)为k时刻的储能电容电压，V_o(k)为k时刻的逆变器输出电压，V_o(k-1)为k-1时刻的逆变器输出电压，V_O(k-2)为k-2时刻的逆变器输出电压，i_L(k)为k时刻的LCL滤波器工频逆变桥侧电感电流，i_L(k-1)为k-1时刻的LCL滤波器工频逆变桥侧电感电流，V_o(k+1)为k+1时刻的逆变器输出电压，V^* _o(k+1)为k+1时刻的逆变器输出电压设定值，D(k-1)为k-1时刻的反激变换器PWM信号占空比，D(k-2)为k-2时刻的反激变换器PWM信号占空比，D(k)为k时刻的反激变换器PWM信号占空比，为t次训练后网络实际输出，为t次训练后期望输出，E_o为期望误差，E_t为t次训练后20N个训练样本的平方和误差。The list of symbols in this paper is: V _d is the voltage of the energy storage capacitor, V _o is the output voltage of the inverter, i _L is the inductor current of the LCL filter power frequency inverter bridge side, D is the duty cycle of the flyback converter PWM signal, V _d (k) is the energy storage capacitor voltage at k time, V _o (k) is the inverter output voltage at k time, V _o (k-1) is the inverter output voltage at k-1 time, V _O (k-2) is the inverter output voltage at time k-2, i _L (k) is the LCL filter power frequency inverter bridge side inductor current at time k, i _L (k-1) is the time k-1 LCL filter power frequency inverter bridge side inductor current, V _o (k+1) is the inverter output voltage at k+1 time, V ^* _o (k+1) is the inverter output at k+1 time Voltage setting value, D(k-1) is the duty cycle of the flyback converter PWM signal at time k-1, D(k-2) is the duty cycle of the flyback converter PWM signal at time k-2, D (k) is the duty cycle of the flyback converter PWM signal at time k, is the actual output of the network after t times of training, is the expected output after t times of training, E _o is the expected error, and E _t is the sum of square errors of 20N training samples after t times of training.

具体实施方式Detailed ways

本发明所述风力发电逆变器包括：整流器、储能电容C1、反激变换器(由变压器TX1、第一功率管Q1、二极管D1和第二电容C2构成)、工频逆变桥(由第二功率管Q2～第五功率管Q5构成)、LCL滤波器(由第三电容C3、第一电感L1、第二电感L2构成)、第一驱动模块(即图1中的驱动模块1)、第二驱动模块(即图1中的驱动模块2)、MPU控制器、第一电压传感器UT1、第二电压传感器UT2和电流传感器CT。The wind power inverter of the present invention includes: a rectifier, an energy storage capacitor C1, a flyback converter (composed of a transformer TX1, a first power tube Q1, a diode D1 and a second capacitor C2), a power frequency inverter bridge (composed of The second power tube Q2 to the fifth power tube Q5), the LCL filter (composed of the third capacitor C3, the first inductor L1, and the second inductor L2), the first drive module (that is, the drive module 1 in Figure 1) , the second driving module (ie, the driving module 2 in FIG. 1 ), the MPU controller, the first voltage sensor UT1 , the second voltage sensor UT2 and the current sensor CT.

所述整流器的三相输入端与风机的三相输出端连接，整流器的单相输出正端与储能电容的正极连接，整流器的单相输出负端接地；储能电容负极接地；第一电压传感器的待测电压输入正端与储能电容正极连接，第一电压传感器的待测电压输入负端接地，第一电压传感器的测量信号输出端与MPU控制器连接；反激变换器的输入正端与储能电容正极连接，反激变换器的输入负端接地；反激变换器的输出正端与工频逆变桥一输入端连接，反激变换器的输出负端与工频逆变桥另一输入端连接；工频逆变桥的一输出端与LCL滤波器一输入端连接，工频逆变桥的另一输出端与LCL滤波器另一输入端连接；LCL滤波器的一输出端与第二电压传感器的待测电压输入正端连接，LCL滤波器的另一输出端与第二电压传感器的待测电压输入负端连接；电流传感器的正极与工频逆变桥侧电感另外一端连接，电流传感器的负极与第三电容的正极连接，电流传感器的测量信号输出端与MPU控制器连接；第一驱动模块的输入端与MPU控制器连接，第一驱动模块的输出端与反激变换器中的第一功率管栅极连接；第二驱动模块的输入端与MPU控制器连接，第二驱动模块的一输出端与工频逆变桥中的第二功率管、第五功率管的栅极连接，第二驱动模块的另一输出端与工频逆变桥中的第三功率管、第四功率管的栅极连接；MPU控制器与第一电压传感器、第二电压传感器、电流传感器的测量信号输出端连接，MPU控制器还与第一驱动模块、第二驱动模块的输入端连接；负载正极接负载侧电感另外一端，负载负极接第三电容的负极。The three-phase input terminal of the rectifier is connected to the three-phase output terminal of the fan, the single-phase output positive terminal of the rectifier is connected to the positive pole of the energy storage capacitor, the single-phase output negative terminal of the rectifier is grounded; the negative pole of the energy storage capacitor is grounded; the first voltage The positive terminal of the input voltage to be measured of the sensor is connected to the positive pole of the energy storage capacitor, the negative terminal of the input voltage to be measured of the first voltage sensor is grounded, and the output terminal of the measurement signal of the first voltage sensor is connected to the MPU controller; the input positive terminal of the flyback converter terminal is connected to the positive pole of the energy storage capacitor, and the negative input terminal of the flyback converter is grounded; the positive output terminal of the flyback converter is connected to the first input terminal of the power frequency inverter bridge, and the negative output terminal of the flyback converter is connected to the power frequency inverter bridge The other input end of the bridge is connected; one output end of the power frequency inverter bridge is connected to one input end of the LCL filter, and the other output end of the power frequency inverter bridge is connected to the other input end of the LCL filter; one of the LCL filter The output terminal is connected to the positive terminal of the voltage input to be measured of the second voltage sensor, and the other output terminal of the LCL filter is connected to the negative terminal of the voltage input to be measured of the second voltage sensor; the positive pole of the current sensor is connected to the power frequency inverter bridge side inductance The other end is connected, the negative pole of the current sensor is connected to the positive pole of the third capacitor, the measurement signal output end of the current sensor is connected to the MPU controller; the input end of the first driving module is connected to the MPU controller, and the output end of the first driving module is connected to the MPU controller. The first power tube grid in the flyback converter is connected; the input terminal of the second drive module is connected to the MPU controller, and one output terminal of the second drive module is connected to the second power tube and the fifth power tube in the power frequency inverter bridge. The grid of the power tube is connected, and the other output end of the second drive module is connected to the grid of the third power tube and the fourth power tube in the power frequency inverter bridge; the MPU controller is connected to the first voltage sensor, the second voltage sensor The measurement signal output terminals of the sensors and current sensors are connected, and the MPU controller is also connected with the input terminals of the first drive module and the second drive module; the positive pole of the load is connected to the other end of the inductor on the load side, and the negative pole of the load is connected to the negative pole of the third capacitor.

以下结合附图和具体实施方式进一步说明本发明。The present invention will be further described below in conjunction with the accompanying drawings and specific embodiments.

在本实施例中，风机GS为小型永磁直驱式风力发电机，风轮直径为1.2m，额定功率为300W，额定电压为24V，额定转速800r/min，启动风速1m/s，额定风速10m/s。In this embodiment, the wind turbine GS is a small permanent magnet direct-drive wind turbine with a rotor diameter of 1.2m, a rated power of 300W, a rated voltage of 24V, a rated speed of 800r/min, a starting wind speed of 1m/s, and a rated wind speed of 10m/s.

针对该风力发电机设计的逆变器结构框图如图1所示。该装置主要包括：整流器、储能电容C1、反激变换器、工频逆变桥、LCL滤波器、第一驱动模块、第二驱动模块、MPU控制器、第一电压传感器UT1、第二电压传感器UT2、电流传感器CT；其中MPU控制器选用TI公司的高性能浮点数字信号处理器TMS320F28335，该数字信号处理器集成了I/O模块、ADC模块、PWM模块、CAN模块、UART和SPI模块等功能模块；储能电容C1选用2200uF电解电容；反激变换器中变压器TX1采用NA5814-AL，第一功率管Q1选用TK50X15J1，二极管D1采用C2D05120E，第二电容C2选用0.47uF无极性电容；第二功率管Q2、第三功率管Q3、第四功率管Q4、第五功率管Q5选用IPB60R190C6，第一驱动模块与第二驱动模块为MCP14E4；LCL滤波器中的工频逆变桥侧电感L1选用0.5mH电感，负载侧电感L2选用0.1mH电感，第三电容C3选用0.47uf无极性电容，第一电压传感器UT1与第二电压传感器均采用MCP6022；电流传感器CT采用霍尔电流传感器ACS712ELCTR-058-1。The structural block diagram of the inverter designed for this wind turbine is shown in Fig. 1. The device mainly includes: rectifier, energy storage capacitor C1, flyback converter, power frequency inverter bridge, LCL filter, first drive module, second drive module, MPU controller, first voltage sensor UT1, second voltage Sensor UT2, current sensor CT; the MPU controller uses TI's high-performance floating-point digital signal processor TMS320F28335, which integrates I/O modules, ADC modules, PWM modules, CAN modules, UART and SPI modules and other functional modules; the energy storage capacitor C1 uses a 2200uF electrolytic capacitor; the transformer TX1 in the flyback converter uses NA5814-AL, the first power tube Q1 uses TK50X15J1, the diode D1 uses C2D05120E, and the second capacitor C2 uses a 0.47uF non-polar capacitor; The second power tube Q2, the third power tube Q3, the fourth power tube Q4, and the fifth power tube Q5 use IPB60R190C6, the first drive module and the second drive module are MCP14E4; the power frequency inverter bridge side inductance L1 in the LCL filter Choose 0.5mH inductance, load side inductance L2 choose 0.1mH inductance, third capacitor C3 choose 0.47uf non-polar capacitor, the first voltage sensor UT1 and the second voltage sensor both use MCP6022; current sensor CT uses Hall current sensor ACS712ELCTR-058 -1.

图2为所建立的三层BP神经网络逆模型。输入层神经元节点数为9个，隐含层神经元节点数为10个，输出层神经元节点数为1个。Figure 2 is the inverse model of the established three-layer BP neural network. The number of neuron nodes in the input layer is 9, the number of neuron nodes in the hidden layer is 10, and the number of neuron nodes in the output layer is 1.

图3为单个神经元的结构示意图，x₀,x₁,…x_i为神经元的输入信号，w_ij为神经元的权系数，θ_j为神经元的阈值，则神经元的输出为：Figure 3 is a schematic diagram of the structure of a single neuron, x ₀ , x ₁ ,... _xi is the input signal of the neuron, w _ij is the weight coefficient of the neuron, θ _j is the threshold of the neuron, and the output of the neuron is:

${y the y}_{j j} = = f f ((\underset{i i}{Σ Σ} {w w}_{i i j j} * * {x x}_{i i} - - {θ θ}_{j j}))$

隐含层神经元转移函数使用双曲正切函数，即The hidden layer neuron transfer function uses the hyperbolic tangent function, namely

$f f ((x x)) = = \frac{{e e}^{22 x x} - - {e e}^{- - 22 x x}}{{e e}^{22 x x} + + {e e}^{- - 22 x x}}$

其中：in:

x为转移函数输入；x is the transfer function input;

f(x)为转移函数输出。f(x) is the transfer function output.

输出层神经元转移函数使用S型函数，即The neuron transfer function of the output layer uses a sigmoid function, namely

$f f ((x x)) = = \frac{11}{11 + + {e e}^{- - x x}}$

其中：in:

x为转移函数输入；x is the transfer function input;

f(x)为转移函数输出。f(x) is the transfer function output.

使得风机分别处于额定风速(10m/s)、80％额定风速(8m/s)、60％额定风速(6m/s)、40％额定风速(4m/s)和20％额定风速(2m/s)的条件下，每种风速条件下通过改变负载使得逆变器分别工作于额定功率(300W)、80％额定功率(240W)、60％额定功率(180W)、40％额定功率(120W)和20％额定功率(60W)的工况，共计25种工况，每种工况下均利用单回路PI控制器进行逆变器输出电压控制，以40KHz的采样频率连续采集N＝4000组储能电容电压V_d、逆变器输出电压V_o、反激变换器PWM信号占空比D、LCL滤波器的工频逆变桥侧电感电流i_L，共构成25N＝100000个样本；随机抽取其中的80000个作为训练样本，其余20000个作为检验样本。Make the fans at rated wind speed (10m/s), 80% rated wind speed (8m/s), 60% rated wind speed (6m/s), 40% rated wind speed (4m/s) and 20% rated wind speed (2m/s) ) under the condition of each wind speed, by changing the load to make the inverter work at rated power (300W), 80% rated power (240W), 60% rated power (180W), 40% rated power (120W) and 20% of the rated power (60W) working condition, a total of 25 working conditions, each working condition uses a single-loop PI controller to control the inverter output voltage, and continuously collects N=4000 groups of energy storage at a sampling frequency of 40KHz Capacitor voltage V _d , inverter output voltage V _o , flyback converter PWM signal duty ratio D, and LCL filter's power frequency inverter bridge side inductor current i _L constitute a total of 25N = 100,000 samples; randomly select them 80,000 of them are used as training samples, and the remaining 20,000 are used as testing samples.

实施例中设置E_o＝4.0,阈值Tr＝0.02，迭代训练15422次后，Et＝3.92，满足终止条件E_t<E_o，训练结束。此时，再利用检验样本对模型进行测试，计算得出样本均方误差ΔMSE＝0.014，有ΔMSE<Tr，因此认为满足逆模型精度要求，此时记录各个神经元的权系数和阈值。In the embodiment, E _o = 4.0, threshold Tr = 0.02, after 15422 iterations of training, Et = 3.92, satisfying the termination condition E _t <E _o , and the training ends. At this time, test the model with the test sample, calculate the sample mean square error ΔMSE=0.014, and ΔMSE<Tr, so it is considered to meet the accuracy requirements of the inverse model, and record the weight coefficients and thresholds of each neuron at this time.

实施例中PI控制器的参数整定为K_p＝0.02，K_i＝1200；逆变器输出电压设定值V^* _o由数字信号处理器按照离散正弦规律变化给定。按照上述步骤编写软件程序并植入DSP芯片中，即可实现基于神经网络逆模型与PI控制器的风机逆变器输出电压复合控制。In the embodiment, the parameters of the PI controller are set as K _p =0.02, K _i =1200; the set value of the inverter output voltage V ^* _o is given by the digital signal processor according to the discrete sine law. According to the above steps, the software program is written and implanted into the DSP chip to realize the composite control of the output voltage of the fan inverter based on the neural network inverse model and the PI controller.

图4为使用本发明控制方法后风机逆变器的输出电压曲线，可见输出电压在过零点及峰值附近无明显震荡，并具有较高的正弦度。Fig. 4 is the output voltage curve of the fan inverter after using the control method of the present invention. It can be seen that the output voltage has no obvious oscillation near the zero crossing point and peak value, and has a relatively high sine degree.

为了进一步分析本发明控制方法的效果，分别在不同工况下将其与传统PI控制方法(取K_p＝0.02，K_i＝1200)进行了对比，表1为两种方法的逆变器输出电压谐波含量对比结果，可知本发明方法对于工况的变化具有较好的鲁棒性，不同工况下的逆变器输出电压谐波含量均明显小于传统PI控制方法。In order to further analyze the effect of the control method of the present invention, it was compared with the traditional PI control method (taking K _p =0.02, K _i =1200) under different working conditions. Table 1 shows the inverter output of the two methods From the comparison results of the voltage harmonic content, it can be seen that the method of the present invention has good robustness to the change of working conditions, and the harmonic content of the inverter output voltage under different working conditions is obviously smaller than that of the traditional PI control method.

表1：本发明复合控制方法与PI控制方法的效果对比Table 1: Comparison of the effects of the compound control method of the present invention and the PI control method

以上所述，仅为本发明较佳的具体实施方式，但本发明的保护范围应该并不局限于此，任何熟悉本技术领域的技术人员依据本发明揭露的核心技术、所能作出的变化或替换，都应涵盖在本发明的保护范围之内。The above is only a preferred embodiment of the present invention, but the scope of protection of the present invention should not be limited thereto. Anyone familiar with the technical field can make changes or Replacement should be covered within the protection scope of the present invention.

Claims

1. An off-grid type wind power inverter is characterized in that the composition includes an MPU controller and a rectifier, an energy storage capacitor (C1), a flyback converter, and a power converter connected successively between the wind-driven generator and the load. frequency inverter bridge and LCL filter, the MPU controller collects the energy storage capacitor voltage and the inverter output voltage through the first voltage sensor (UT1) and the second voltage sensor (UT2), respectively, and collects the output voltage through the current sensor (CT) The inductance current of the power frequency inverter bridge side of the LCL filter controls the flyback converter and the power frequency inverter bridge respectively through two drive modules. The MPU controller operates in the following manner:

① Collect the energy storage capacitor voltage V _d through the first voltage sensor (UT1), collect the inverter output voltage V _o through the second voltage sensor (UT2), and collect the power frequency inverter bridge of the LCL filter through the current sensor (CT) side inductor current i _L ; taking the inverter as the object, establish and train the corresponding neural network inverse model, as follows:

a. Select a three-layer BP neural network to establish the inverse model of the system, in which the number of neuron nodes in the input layer is 9, the number of neuron nodes in the hidden layer is 10, the number of neuron nodes in the output layer is 1, and the number of neuron nodes in the hidden layer is 1. The neuron transfer function uses the hyperbolic tangent function, and the output layer neuron transfer function uses the S-type function;

b. Collect running data on the inverter;

Make the fan under the conditions of rated wind speed, 80% of rated wind speed, 60% of rated wind speed, 40% of rated wind speed and 20% of rated wind speed, and make the inverter work at rated power, 80% of rated power and Rated power, 60% rated power, 40% rated power and 20% rated power working conditions, a total of 25 working conditions, each working condition uses a single-loop PI controller to control the inverter output voltage, and continuously collects N groups of energy storage capacitor voltage V _d , inverter output voltage V _o , flyback converter PWM signal duty ratio D, and LCL filter's power frequency inverter bridge side inductance current i _L constitute 25N groups of operating data in total;

c. Using the collected operating data to construct samples;

For time k, the constructed samples are {V _o (k+1), V _o (k), V _o (k-1), V _o (k-2), i _L (k), i _L (k -1), D(k-1), D(k-2), V _d (k), D(k)}, where V _o (k+1) is the inverter output voltage at time k+1 , V _o (k) is the inverter output voltage at k time, V _o (k-1) is the inverter output voltage at k-1 time, V _o (k-2) is the inverter at k-2 time i _L (k) is the inductance current of the power frequency inverter bridge side of the LCL filter at time k, i _L (k-1) is the power frequency inverter bridge side inductance of the LCL filter at time k-1 current, D(k-1) is the duty cycle of the flyback converter PWM signal at time k-1, D(k-2) is the duty cycle of the flyback converter PWM signal at time k-2, V _d (k ) is the energy storage capacitor voltage at time k, D(k) is the duty cycle of the flyback converter PWM signal at time k, and 25N groups of operating data constitute 25N samples in total;

d. Randomly select 20N of the 25N sample data as training samples, and the remaining 5N as test samples, and train the established three-layer BP neural network; during the training process, for the samples at time k, take the first one of the BP neural network The input terminal is the inverter output voltage V _o (k) at time k, the second input terminal is the inverter output voltage V _o (k-1) at time k-1, and the third input terminal is k-2 The inverter output voltage V _o (k-2) at time, the fourth input terminal is the power frequency inverter bridge side inductor current i _L (k) of the LCL filter at time k, and the fifth input terminal is k- The inductor current i _L (k-1) of the power frequency inverter bridge side of the LCL filter at time 1, the sixth input terminal is the inverter output voltage V _o (k+1) at time k+1, the seventh The input terminal is the energy storage capacitor voltage V _d (k) at time k, the eighth input terminal is the flyback converter PWM signal duty cycle D(k-1) at time k-1, and the ninth input terminal is k The duty cycle D(k-2) of the flyback converter PWM signal at time -2, and the output of the BP neural network is the duty cycle D(k) of the flyback converter PWM signal at time k;

② Use the trained BP neural network inverse model and PI controller to control the inverter. The specific steps are:

a. Use the trained BP neural network inverse model to calculate the duty cycle of the flyback converter PWM signal. For the current k moment, take the first input terminal as the inverter output voltage V _o (k) at the moment k, The second input terminal is the inverter output voltage V _o (k-1) at time k-1, the third input terminal is the inverter output voltage V _o (k-2) at time k-2, and the fourth input terminal is The first input terminal is the power frequency inverter bridge side inductor current i _L (k) of the LCL filter at time k, and the fifth input terminal is the power frequency inverter bridge side inductor current i _L of the LCL filter at time k-1 (k-1), the sixth input terminal is the set value of inverter output voltage V ^* _o (k+1) at time k+1, and the seventh input terminal is the energy storage capacitor voltage V _d ( k), the eighth input terminal is the duty cycle D(k-1) of the flyback converter PWM signal at time k-1, and the ninth input terminal is the duty cycle of the flyback converter PWM signal at time k-2 D(k-2), the output of the BP neural network is d(k);

b. Send the deviation between the inverter output voltage set value V ^* _o (k+1) at the next moment and the inverter output voltage V _o (k) at the current k moment to the PI controller to obtain the closed-loop control value d _C (k), namely

{d d}_{C C} ((k k)) = = {K K}_{p p} e e ((k k)) + + {K K}_{i i} {Σ Σ}_{j j = = 00}^{k k} e e ((k k))

Wherein: e (k)=V ^* _o (k+1)-V _o (k), K _p represents proportional coefficient, and K _i represents integral coefficient;

c. The neural network inverse model and the PI controller constitute a composite controller, and its output is the output d(k) of the neural network inverse model and the output d _C (k) of the PI controller are superimposed according to the ratio K:(1-K) , as the final flyback converter PWM signal duty cycle D(k), namely:

D(k)＝Kd(k)+(1-K)d _C (k)

Among them, the value of K is:

K K = = \{\begin{matrix} 0.9 0.9 & i i f f [[((- - \frac{11}{1212} π π < < θ θ < < \frac{11}{1212} π π)) o o r r ((- - π π < < θ θ < < - - \frac{1111}{1212} π π)) o o r r ((\frac{1111}{1212} π π < < θ θ < < π π))]] \\ 0.1 0.1 & i i f f [[((- - \frac{77}{1212} π π < < θ θ < < - - \frac{55}{1212} π π)) o o r r ((\frac{55}{1212} π π < < θ θ < < \frac{77}{1212} π π))]] \\ 0.5 0.5 & e e l l s the s e e \end{matrix}

Where θ is the phase of the set value of the inverter output voltage.

2. The off-grid wind power inverter according to claim 1, characterized in that, the training of the BP neural network adopts an error back-propagation algorithm with a variable learning rate, specifically as follows: at first the weights in each layer of neurons The coefficients and thresholds are initialized to a random amount between (-1, +1), and then the output value of 9 neurons in the input layer is used as the input of each neuron in the hidden layer, and the output value of 10 neurons in the hidden layer is used as The input of a single neuron in the output layer, the output value of a single neuron in the output layer is the final output of the network, for the kth training sample, calculate the actual output of the network after the tth training with the desired output The error of:

\begin{matrix} {δ δ}_{k k}^{t t} = = (({y the y}_{k k}^{r r} - - {c c}_{k k}^{t t})) & k k = = 11,, 22,, ... ...,, 2020 N N \end{matrix}

And the sum of squared errors of 20N training samples is:

{E E.}_{t t} = = {Σ Σ}_{k k = = 11}^{2020 N N} {(({y the y}_{k k}^{t t} - - {c c}_{k k}^{t t}))}^{22} / / 22 = = {Σ Σ}_{k k = = 11}^{2020 N N} {(({δ δ}_{k k}^{t t}))}^{22} / / 22

After each calculation, compare E _t with the expected error E _o , if E _t < E _o , the training termination condition is satisfied, and the training ends; otherwise, E _t is backpropagated along the connection path, and the weights of each layer are adjusted one by one Coefficients and thresholds until E _t < E _o ; then, use 5N test samples to test the model and calculate the sample mean square error ΔMSE, if ΔMSE<Tr (Tr is a fixed threshold), it is considered that the trained If the BP neural network model meets the accuracy requirements, record the weight coefficients and thresholds of each neuron; otherwise, retrain the BP neural network model until it meets the accuracy requirements.

3. The off-grid wind power inverter according to claim 1 or 2, characterized in that the flyback converter includes a transformer (TX1), a first power tube (Q1), a diode (D1) and a second Two capacitors (C2), the primary side coil of the transformer (TX1) is connected to the energy storage capacitor voltage through the first power tube (Q1), and its secondary side coil supplies power to the power frequency inverter bridge through the diode (D1), and the second capacitor (C2) is connected in parallel to the DC input end of the power frequency inverter bridge, and the gate of the first power transistor (Q1) is connected to the output end of the first drive module.

4. The off-grid wind power inverter according to claim 3, characterized in that, the power frequency inverter bridge is a full-control bridge connected by four power tubes, and the grids of the four power tubes are connected to each other. The output end of the second drive module.

5. The off-grid wind power inverter according to claim 4, characterized in that the LCL filter comprises a power frequency inverter bridge side inductance (L1), a third capacitor (C3) and a load side inductance ( L2), one end of the power frequency inverter bridge side inductance (L1) is connected to the first AC output end of the power frequency inverter bridge, the other end is connected to the positive pole of the current sensor (CT), and the positive pole of the third capacitor (C3) is connected to the current sensor ( The negative pole of CT), the negative pole of the third capacitor (C3) is connected to the second AC output terminal of the power frequency inverter bridge, the measurement signal output terminal of the current sensor (CT) is connected to the MPU controller, and one end of the load side inductor (L2) is connected to The positive pole of the third capacitor (C3) is connected to the positive pole of the load at the other end.