Three-phase inverter midpoint potential control method and device based on feedforward neural network

Abstract

The invention belongs to the technical field of inverters, and mainly provides a three-phase inverter midpoint potential control method and device based on a feed-forward neural network, in the embodiment of the invention, real-time characteristic parameters of a three-phase inverter are obtained, and the real-time characteristic parameters are input into a target feed-forward neural network model obtained through pre-training, so that the midpoint potential of the three-phase inverter is controlled. The method comprises the following steps: firstly, obtaining a midpoint potential offset predicted value output by a target feedforward neural network model, then adjusting a modulation parameter of SVPWM modulation according to the midpoint potential offset predicted value, and finally, outputting a PWM signal for controlling an inverter switching device based on the adjusted modulation parameter to drive a three-phase inverter. The active balance of the neutral-point potential of the three-phase inverter bus based on the feedforward neural network and combined with SVPWM modulation is realized, and the technical problems that the neutral-point potential of the three-phase inverter bus is unbalanced and easy to deviate are solved.

Description

Technical Field

[0001] The present application belongs to the technical field of inverters, and in particular, relates to a method and device for controlling the midpoint potential of a three-phase inverter based on a feedforward neural network. Background Art

[0002] I-type three-level three-phase four-leg inverters are widely used in the field of power conversion. However, due to load imbalance, device parameter differences and limitations of control algorithms, the bus midpoint potential of I-type three-level three-phase four-leg inverters is prone to offset, resulting in the output voltage of I-type three-level three-phase four-leg inverters being easily distorted, and problems such as increased device stress and decreased system stability occur, which reduces the dynamic performance and reliability of the inverter. Summary of the invention

[0003] The present application provides a three-phase inverter midpoint potential control method and device based on a feedforward neural network, which can solve the technical problem that the bus midpoint potential of a traditional three-phase inverter is easily offset under different load conditions.

[0004] In a first aspect, an embodiment of the present application provides a three-phase inverter midpoint potential control method based on a feedforward neural network, and the three-phase inverter midpoint potential control method includes:

[0005] Obtaining a real-time characteristic vector of the three-phase inverter;

[0006] Inputting the real-time feature vector into a pre-trained target feedforward neural network model to obtain a predicted value of a midpoint potential offset output by the target feedforward neural network model;

[0007] Adjusting the modulation parameters of the SVPWM modulation according to the predicted value of the midpoint potential offset output by the target feedforward neural network model to obtain the adjusted modulation parameters;

[0008] A PWM signal for controlling the switching device of the three-phase inverter is output based on the adjusted modulation parameter to drive the three-phase inverter.

[0009] A second aspect of the embodiments of the present application further provides a control device for a three-phase inverter, the control device for the three-phase inverter comprising:

[0010] An acquisition unit, used for acquiring a real-time feature vector of the three-phase inverter;

[0011] A prediction unit, used for inputting the real-time feature vector into a pre-trained target feedforward neural network model to obtain a predicted value of a midpoint potential offset output by the target feedforward neural network model;

[0012] An adjustment unit, used for adjusting a modulation parameter of SVPWM modulation according to a predicted value of a midpoint potential offset output by the target feedforward neural network model to obtain an adjusted modulation parameter;

[0013] A control unit is used to output a PWM signal for controlling the switching device of the three-phase inverter based on the adjusted modulation parameter to drive the three-phase inverter.

[0014] A third aspect of an embodiment of the present application provides a control device, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor. When the computer program is executed by the processor, the steps of the three-phase inverter neutral point potential control method based on a feedforward neural network described in the first aspect are implemented.

[0015] A fourth aspect of an embodiment of the present application provides a computer program product, which stores a computer program. When the computer program is executed by a processor, the steps of the three-phase inverter midpoint potential control method based on a feedforward neural network described in the first aspect are implemented.

[0016] In an embodiment of the present application, real-time characteristic parameters of a three-phase inverter are obtained, and the real-time characteristic parameters are input into a pre-trained target feedforward neural network model to obtain a predicted value of a midpoint potential offset output by the target feedforward neural network model. Then, the modulation parameters of the SVPWM modulation are adjusted according to the predicted value of the midpoint potential offset. Finally, a PWM signal for controlling the inverter switching device is output based on the adjusted modulation parameters to drive the three-phase inverter, thereby achieving active balancing of the bus midpoint potential of the three-phase inverter based on a feedforward neural network combined with SVPWM modulation, and solving the technical problem that the bus midpoint potential of the three-phase inverter is unbalanced and prone to offset.

[0017] In addition, the present application utilizes the predicted value of the midpoint potential offset output by the target feedforward neural network model in real time to quickly adjust the modulation strategy of SVPWM modulation. It has the characteristics of high real-time performance, can quickly adapt to load changes, effectively reduce output voltage distortion and device stress, and enhance the stability and reliability of the system. BRIEF DESCRIPTION OF THE DRAWINGS

[0018] Figure 1 A schematic diagram of the implementation flow of a three-phase inverter midpoint potential control method based on a feedforward neural network provided in an embodiment of the present application.

[0019] Figure 2 A schematic diagram of the structure of the main circuit of the power conversion system provided in an embodiment of the present application.

[0020] Figure 3 A schematic diagram of the training steps of the target feedforward neural network model provided in an embodiment of the present application.

[0021] Figure 4 A schematic diagram of the implementation flow of step 103 of the three-phase inverter midpoint potential control method based on a feedforward neural network provided in an embodiment of the present application.

[0022] Figure 5 A schematic diagram of the implementation flow of adjusting the zero vector action time provided in an embodiment of the present application.

[0023] Figure 6 A schematic diagram of the implementation process of online updating of parameters of a target feedforward neural network model provided in an embodiment of the present application.

[0024] Figure 7 A schematic diagram of the structure of a control device for a three-phase inverter provided in an embodiment of the present application.

[0025] Figure 8 A schematic diagram of the structure of a control device provided in an embodiment of the present application. DETAILED DESCRIPTION

[0026] In order to make the purpose, technical solution and advantages of the present application more clearly understood, the present application is further described in detail below in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present application and are not used to limit the present application.

[0027] It should be understood that in the present specification and the appended claims, the term "comprising" indicates the presence of described features, wholes, steps, operations, elements and / or components, but does not exclude the presence or addition of one or more other features, wholes, steps, operations, elements, components and / or combinations thereof.

[0028] In addition, in the description of the present application specification and the appended claims, the terms "first", "second", etc. are only used to distinguish the descriptions and cannot be understood as indicating or implying relative importance.

[0029] References to "one embodiment" or "some embodiments" etc. described in the specification of this application mean that one or more embodiments of the present application include specific features, structures or characteristics described in conjunction with the embodiment. Therefore, the phrases "in one embodiment", "in some embodiments", "in some other embodiments", "in some other embodiments", etc. appearing in different places in this specification do not necessarily refer to the same embodiment, but mean "one or more but not all embodiments", unless otherwise specifically emphasized in other ways.

[0030] I-type three-level three-phase four-leg inverters are widely used in the field of power conversion. However, due to load imbalance, device parameter differences and limitations of control algorithms, the bus midpoint potential of the I-type three-level three-phase four-leg inverter is prone to offset, that is, the bus midpoint potential is unbalanced, which causes the output voltage of the I-type three-level three-phase four-leg inverter to be easily distorted, and there are problems such as increased device stress and decreased system stability.

[0031] Traditional busbar midpoint potential balance control methods often fail to achieve ideal results when dealing with complex load changes. Space vector pulse width modulation (SVPWM), as an advanced modulation strategy, has higher DC voltage utilization and lower harmonic distortion, but there are still challenges in midpoint potential balance.

[0032] The present application provides a three-phase inverter midpoint potential control method and device based on a feedforward neural network, which can solve the technical problem that the bus midpoint potential of a traditional inverter is prone to imbalance under different load conditions.

[0033] like Figure 1 The figure shows a schematic diagram of the implementation flow of the three-phase inverter midpoint potential control method based on a feedforward neural network provided in an embodiment of the present application. The three-phase inverter midpoint potential control method based on a feedforward neural network can be executed by a control device of a three-phase inverter in a power conversion system (PCS), and the control device is integrated with a control device of the three-phase inverter, and the control device can be implemented by software or hardware or a combination of software and hardware. The three-phase inverter midpoint potential control method based on a feedforward neural network provided in an embodiment of the present application is implemented by the following steps 101 to 104.

[0034] Step 101, obtaining a real-time feature vector of a three-phase inverter.

[0035] In an embodiment of the present application, operating parameters such as the three-phase output voltage, bus voltage, three-phase output current and temperature of the three-phase inverter can be collected in real time, and pre-processing operations such as cleaning, normalization and feature extraction are performed on the operating parameters of the three-phase inverter collected in real time to obtain a real-time feature vector.

[0036] Among them, when preprocessing the real-time collected inverter operating parameters, digital filtering algorithms such as low-pass filtering and Kalman filtering can be used to remove measurement noise, set thresholds or use statistical methods to identify and remove abnormal data points to achieve outlier detection. Each operating parameter is normalized according to its maximum value or specified range, so that the characteristic value of the feature in the sample feature vector is within the effective range of the activation function of the target feedforward neural network model. The time domain characteristics of the operating parameters can also be extracted by calculating the average value, root mean square value, and peak value of the three-phase output voltage, bus voltage, and three-phase output current. By performing fast Fourier transform (FFT) on the three-phase output voltage, bus voltage, and three-phase output current, the spectral characteristics such as harmonic content of the operating parameters are extracted. The statistical characteristics such as variance, skewness, and kurtosis of the three-phase output voltage, bus voltage, three-phase output current, and temperature are calculated, and the extracted features are combined to obtain a real-time feature vector.

[0037] In one embodiment, the specific steps are as follows: set the sampling frequency f of data acquisition s , for example, 10kHz, to ensure effective capture of the system dynamic characteristics. The three-phase voltage V is collected by the voltage sensor A-sense (t), V B-sense (t), V C-sense (t). The three-phase current I is collected by the current sensor A-sense (t), I B-sense (t), I C-sense (t) and the neutral current I N-sense (t), collect the bus midpoint potential V neutral (t), and collect the temperature T(t). The collected data are stored in chronological order to form the original data set D raw .

[0038] Use statistical methods, such as the three-times standard deviation method, to identify outliers. For example, set the outlier condition: x i -μ>3σ, where x i is the sampling point, μ is the mean, and σ is the standard deviation. For detected abnormal values, interpolation or the previous effective value is used to replace them. A low-pass filter is used to filter out high-frequency noise. For example, a finite impulse response (FIR) filter is used with a cutoff frequency of f c =f signal ×1.5, where f signal The highest frequency component of the signal is filtered to obtain the filtered operating parameters, and then the filtered operating parameters are normalized. For example, scaled to the [0,1] interval: Among them, x is the original data, x min 、x max are the minimum and maximum values ​​of the data respectively. Next, feature extraction is performed.

[0039] First, calculate the mean in the time domain features Standard Deviation: Skewness: Kurtosis: Next, the frequency domain feature calculation is performed, and the signal x(t) is subjected to a fast Fourier transform (FFT) to obtain the spectrum X(f). Then, the main frequency component is extracted to identify the main frequency f in the spectrum. peak and its amplitude |X(f peak )|. The extracted features are combined into a real-time feature vector X,

[0040] Step 102, inputting the real-time feature vector into a pre-trained target feedforward neural network model to obtain a predicted value of the midpoint potential offset output by the target feedforward neural network model.

[0041] Step 103, adjusting the modulation parameters of the SVPWM modulation according to the predicted value of the midpoint potential offset output by the target feedforward neural network model to obtain the adjusted modulation parameters.

[0042] Step 104: outputting a PWM signal for controlling the inverter switching device based on the adjusted modulation parameters to drive the three-phase inverter.

[0043] In the embodiment of the present application, SVPWM (Space Vector Pulse-Width Modulation) modulation is a modulation technology used in three-phase circuits. The pulse width modulation wave (PWM signal) generated by a specific switching pattern composed of six power switching elements of a three-phase inverter can make the output current waveform or voltage waveform as close to an ideal sinusoidal waveform as possible.

[0044] In one embodiment of the present application, the three-phase inverter may be a type I three-level three-phase four-leg inverter.

[0045] like Figure 2 As shown, it is a structural schematic diagram of the main circuit of the power conversion system provided in an embodiment of the present application, and the main circuit is used to realize efficient conversion of DC power into AC power, including a DC side input 21, an I-type three-level three-phase four-bridge arm inverter 22, and an LCL filter 23. Among them, the DC bus positive electrode of the DC side input part is connected to the positive electrode of the DC battery pack, the DC bus negative electrode is connected to the negative electrode of the DC battery pack, and the DC bus capacitor is used to ensure the stability of the DC bus voltage and provide a buffer for the midpoint potential.

[0046] The I-type three-level three-phase four-bridge-arm inverter includes an A-phase bridge arm, a B-phase bridge arm, a C-phase bridge arm, and a neutral line (N) bridge arm. Each bridge arm contains four power switching devices, which are connected according to the I-type three-level topology to form a three-phase four-bridge-arm structure. A clamping diode is added to each bridge arm to limit the voltage stress of the device and protect the power device. The LCL filter is used to filter out the high-frequency harmonics output by the inverter to ensure that the quality of the output current meets the requirements of the power grid.

[0047] In one embodiment of the present application, the LCL filter includes an inverter side inductor, a filter capacitor and a grid side inductor. The value of the inverter side inductor is determined according to the switching frequency f s And the allowable current ripple ΔI is calculated, and the calculation formula is: Filter capacitor C f The capacitance value is usually 5% to 10% of the inverter rated power. The grid-side inductor L2 = L1 / 10 to L1 / 5 to ensure filtering effect and system stability.

[0048] SVPWM modulation regards the three-phase inverter as a vector source that can generate various space vectors. By switching the switching devices reasonably, a suitable non-zero vector (i.e., basic vector) and a zero vector can be selected within a switching cycle to synthesize a voltage space vector of a specific magnitude and direction. These synthesized voltage space vectors are arranged in a specific order on the complex plane to form a hexagonal or circular vector trajectory, thereby generating a smooth rotating magnetic field in the motor. SVPWM modulation generates a voltage or current that is approximately a sine wave by controlling the switching state of the switching devices. Compared with traditional pulse width modulation (PWM) technology, SVPWM can more effectively utilize the DC bus voltage, improve motor efficiency, and reduce harmonic distortion.

[0049] In an embodiment of the present application, during the actual operation of the three-phase inverter, real-time characteristic parameters of the three-phase inverter are obtained in real time, and the real-time characteristic parameters are input into a pre-trained target feedforward neural network model, and a predicted value of the midpoint potential offset is predicted in real time by the target feedforward neural network model. Then, the modulation parameters of the SVPWM modulation are adaptively adjusted according to the predicted value of the midpoint potential offset. Finally, a PWM signal for controlling the inverter switching device is output based on the adjusted modulation parameters to drive the three-phase inverter, thereby realizing active balancing of the bus midpoint potential of the three-phase inverter based on the feedforward neural network and combined with the SVPWM modulation, solving the technical problem of unbalanced bus midpoint potential of the three-phase inverter and easy offset, and improving the dynamic performance and stability of the three-phase inverter.

[0050] In addition, the present application utilizes the predicted value of the midpoint potential offset output by the target feedforward neural network model in real time to quickly adjust the modulation strategy of SVPWM modulation. It has the characteristics of high real-time performance, can quickly adapt to load changes, effectively reduce output voltage distortion and device stress, and enhance the stability and reliability of the power conversion system.

[0051] In one embodiment of the present application, Figure 3 As shown, the training steps adopted by the target feedforward neural network model in the above step 102 may include the following steps 301 to 304.

[0052] Step 301 : collecting operating parameters of a three-phase inverter under different load conditions.

[0053] In one embodiment of the present application, the operating parameters of the three-phase inverter are parameters that significantly affect the bus midpoint potential, which may include the three-phase output voltage, bus voltage, three-phase output current, midpoint potential offset (i.e., midpoint potential offset label value) and temperature of the three-phase inverter.

[0054] Specifically, a high-precision voltage transformer or isolated voltage sensor can be used to measure the three-phase output voltage and bus voltage, a Hall effect current sensor with high bandwidth and good linearity can be used to measure the three-phase output current, and a differential voltage measurement method can be used to measure the bus midpoint potential offset to improve measurement accuracy. Thermistors or digital temperature sensors can be used to monitor the temperature of power devices and the environment.

[0055] Step 302: pre-process the operating parameters to obtain a sample feature vector and a midpoint potential offset label value vector.

[0056] In the embodiment of the present application, the method for obtaining the sample feature vector can refer to the method for obtaining the real-time feature vector in the aforementioned step 101, and the sample feature vector can be obtained by performing preprocessing operations such as cleaning, normalization and feature extraction on a large number of collected three-phase output voltages, bus voltages, three-phase output currents and temperatures. I will not go into details here.

[0057] In the embodiment of the present application, by determining the target variable as the busbar midpoint potential offset V neutral (t), the sample feature vector is paired with the target variable to form a midpoint potential offset label value vector V corresponding to the sample feature vector = [V neutral (t)].

[0058] Step 303: Divide the sample feature vector and the midpoint potential offset label value vector into a training set and a validation set.

[0059] For example, the training set accounts for 70% of the total data, the validation set accounts for 15% of the total data, and the remaining 15% of the data is used as a test set to further verify the model.

[0060] Step 304, train the initial feedforward neural network model using the training set, and verify the trained initial neural network model using the verification set, and when the loss function value of the verification set changes at a rate less than a preset rate in k consecutive rounds of training, complete the training of the initial neural network model and obtain the target feedforward neural network model.

[0061] In the embodiment of the present application, the initial feedforward neural network model includes an input layer, a hidden layer and an output layer. Wherein, the number of input nodes of the input layer is equal to the dimension of the sample feature vector. The number of layers of the hidden layer is 2 or 3, and the number of nodes per layer is selected to be larger than the number of nodes of the input layer, such as 10 to 50 nodes. It can be determined experimentally, usually 2-3 times the number of input nodes, to meet the fitting requirements of complex nonlinear relationships. The number of nodes in the output layer is 1, which is used to output the sample prediction value of the midpoint potential offset. In the embodiment of the present application, the hidden layer uses a nonlinear activation function, such as ReLU (linear rectifier unit), Sigmoid, Tanh, and the output layer uses a linear activation function.

[0062] In the embodiment of the present application, the calculation from the input layer to the hidden layer is as follows: for the jth node of the lth hidden layer: Among them, x i is the i-th input feature, w ij (l) is the weight from the input layer to the hidden layer, b j (l) is the bias of the hidden layer node, and φ is the activation function.

[0063] The calculation between hidden layers is as follows: For the kth node in the lth layer (l ≥ 2) Among them, n l-1 is the number of nodes in the l-1th layer, w jk (l) is the weight from layer l-1 to layer l, b k (l) is the bias of the l-th layer node.

[0064] The output layer is calculated as follows: in, nhidden is the number of nodes in the last hidden layer, w j (o) is the weight from the hidden layer to the output layer, b (o) is the bias of the output layer.

[0065] In one embodiment of the present application, the hidden layer uses Rectified Linear Unit (ReLU) as the activation function, and the formula is f(x)=max(0,x), which has the advantages of simple calculation and fast convergence. The output layer uses a linear activation function to achieve regression prediction.

[0066] After completing the construction of the initial feedforward neural network model, the He initialization method can be used to provide initial weights for the ReLU activation function and initialize the bias to zero or a small random value.

[0067] In the process of training the initial feedforward neural network model using the training set, the mean square error loss function value of the initial feedforward neural network model can be calculated Based on the mean square error loss function value, the parameters of the initial feedforward neural network model are adjusted by back propagation and gradient descent method, or the parameters of the initial feedforward neural network model are adjusted by using Adam optimizer, and the initial feedforward neural network model is retrained, and the loss function value of the validation set is calculated until the number of training times of the initial feedforward neural network reaches the preset number threshold or the change rate of the loss function value is less than the preset change rate threshold, the training of the initial feedforward neural network model is terminated, and the target feedforward neural network model is obtained. Where N is the number of samples, y is the number of samples, i is the label value of the midpoint potential offset, is the predicted value of the midpoint potential offset.

[0068] In one embodiment of the present application, modulation parameters of SVPWM modulation may include vector selection and vector action time.

[0069] In the process of adjusting the modulation parameters of the SVPWM modulation according to the predicted value of the midpoint potential offset in the above step 103, the real-time predicted trend of the bus midpoint potential change can be obtained according to the predicted value of the midpoint potential offset output by the target feedforward neural network model, and the vector selection and vector action time of the SVPWM can be dynamically adjusted based on the real-time predicted trend of the bus midpoint potential change, so as to realize active balancing control of the bus midpoint potential.

[0070] In one embodiment of the present application, the vector action time in the modulation parameters of the SVPWM modulation may include a zero vector action time, and the bus midpoint potential is balanced by adjusting the zero vector action time.

[0071] like Figure 4 As shown, in the process of adjusting the modulation parameters of the SVPWM modulation according to the predicted value of the midpoint potential offset in the above step 103, it can be implemented based on the following steps 401 to 402.

[0072] Step 401, calculating a zero vector duty cycle correction value according to a predicted value of a midpoint potential offset and a preset adjustment coefficient.

[0073] Specifically, we can use the formula k mod =k p ΔV neutral Calculate the zero vector duty cycle correction; where k mod is the zero vector duty cycle correction, k p is the preset adjustment coefficient, ΔV neutral is the predicted value of the midpoint potential offset.

[0074] Step 402, adjusting the zero vector action time of SVPWM modulation based on the zero vector duty cycle correction amount to obtain the adjusted zero vector action time.

[0075] In one embodiment of the present application, Figure 5 As shown, in the process of adjusting the zero vector action time of SVPWM modulation based on the zero vector duty cycle correction amount, it can be implemented based on the following steps 501 to 505.

[0076] Step 501, obtain a voltage command, and calculate a reference voltage vector according to the voltage command

[0077] In the embodiment of the present application, the voltage instruction is a voltage instruction provided by the control system, and the reference voltage vector is the desired voltage vector.

[0078] Step 502: determine the sector position of the reference voltage vector based on the phase angle θ of the reference voltage vector, and select two adjacent non-zero vectors V based on the sector position of the reference voltage vector. α and V β , and the zero vector.

[0079] SVPWM modulation includes six non-zero vectors and two zero vectors. The amplitude of the two zero vectors is zero and they are located in the center. In each sector, two adjacent non-zero vectors and the zero vector are selected, and any voltage vector in each sector can be synthesized according to the principle of volt-second balance.

[0080] Step 503, calculate the non-zero vector V α and V β The action time is T1 and T2.

[0081] In the embodiment of the present application, Where Ts is the switching period, V dc is the DC bus voltage, and δθ is the angle of the reference voltage vector relative to the starting edge of the sector.

[0082] Step 504, calculate the action time of the zero vector T0=T based on T1 and T2 s -T1-T2.

[0083] Step 505, calculate the adjusted zero vector action time T 0- =T0-T 0+ , where k mod is the zero vector duty cycle correction.

[0084] In the embodiment of the present application, the adjusted zero vector action time T can be obtained based on the above steps 501 to 505. 0+ and T 0- , T 0+ and T 0- All are non-negative numbers.

[0085] In one embodiment of the present application, when a PWM signal for controlling the inverter switch device is output based on the adjusted modulation parameters, the T1, T2, T 0+ and T 0- The corresponding switching sequence is constructed to generate a three-phase PWM control signal, which controls the power switching devices of the inverter to ensure the continuity of the output voltage and the balance of the midpoint potential.

[0086] In the embodiment of the present application, the real-time characteristic parameters of the inverter are obtained in each switching cycle of the inverter, and the real-time characteristic parameters are input into the pre-trained target feedforward neural network model to obtain the real-time midpoint potential offset prediction value, and then the zero vector duty cycle correction amount is calculated according to the midpoint potential offset prediction value, and the reference voltage vector is calculated according to the received voltage command. And according to the reference voltage vector Calculate the non-zero vector V α and V β The action time T1 and T2 of the zero vector and the action time T0 of the zero vector are calculated, and the adjusted zero vector action time T is obtained based on the zero vector duty cycle correction amount. 0+ and T 0- Finally, based on T1, T2, T 0+ and T 0- The corresponding switching sequence is constructed and a three-phase PWM control signal is generated to control the switching devices of the three-phase inverter to ensure the continuity of the output voltage and the balance of the midpoint potential, thereby improving the dynamic performance and stability of the three-phase inverter.

[0087] like Figure 6As shown, in one embodiment of the present application, the three-phase inverter midpoint potential control method based on the feedforward neural network in the above-mentioned various embodiments can also establish a feedback loop after step 104 by executing the following steps 601 to 603, and update the parameters of the target feedforward neural network model online, so as to improve the accuracy and adaptability of the target feedforward neural network model, ensure that under various operating conditions, the power conversion system can maintain effective control of the bus midpoint potential, and enhance the robustness and dynamic performance of the power conversion system.

[0088] Step 601, real-time monitoring of the actual value of the bus midpoint potential of the three-phase inverter.

[0089] Step 602, calculating the prediction error of the target feedforward neural network model based on the actual value of the bus midpoint potential and the predicted value of the midpoint potential offset output by the target feedforward neural network model.

[0090] Step 603: Update the parameters of the target feedforward neural network model based on the prediction error.

[0091] In this embodiment, it is assumed that the actual value of the busbar midpoint potential measured in real time is V meas (t), the predicted value of the midpoint potential offset output by the target feedforward neural network model is V pre (t), then the prediction error e(t) = V meas (t)-V pre (t), when the incremental gradient descent algorithm is used to update the parameters of the target feedforward neural network model, for each weight w of the target feedforward neural network model i and bias b i , the gradient is calculated as follows Among them, x i (t) is the input feature, is the loss function; the weights are updated as follows: The bias is updated as follows: Among them, α(t) is the current learning rate.

[0092] In the embodiment of the present application, by updating the parameters (weights and biases) of the target feedforward neural network model, the accuracy and adaptability of the target feedforward neural network model are improved, ensuring that the power conversion system can maintain effective control of the bus midpoint potential under various operating conditions, thereby enhancing the robustness and dynamic performance of the power conversion system.

[0093] It should be noted that, for the sake of simplicity of description, the aforementioned method embodiments are all expressed as a series of action combinations, but those skilled in the art should know that the present application is not limited to the described order of actions, and in some embodiments of the present application, certain steps may be performed in other orders.

[0094] The present application also provides a control device for a three-phase inverter. Figure 7 As shown, the control device 700 of the three-phase inverter can be configured in a control device, and the control device of the three-phase inverter includes:

[0095] An acquisition unit 701 is used to acquire a real-time feature vector of the three-phase inverter;

[0096] A prediction unit 702, configured to input the real-time feature vector into a pre-trained target feedforward neural network model to obtain a predicted value of a midpoint potential offset output by the target feedforward neural network model;

[0097] An adjusting unit 703, configured to adjust a modulation parameter of SVPWM modulation according to a predicted value of a midpoint potential offset output by the target feedforward neural network model, to obtain an adjusted modulation parameter;

[0098] The control unit 704 is used to output a PWM signal for controlling the switching device of the three-phase inverter based on the adjusted modulation parameter to drive the three-phase inverter.

[0099] It should be noted that, for the convenience and simplicity of description, the specific working process of the control device 700 of the three-phase inverter described above can refer to the above Figures 1 to 6 The corresponding process of the three-phase inverter midpoint potential control method based on feedforward neural network described in will not be repeated here.

[0100] like Figure 8 As shown, the embodiment of the present application also provides a control device. Figure 8 As shown, the control device 8 may include: a processor 80, a memory 81, and a computer program 82 stored in the memory 81 and executable on the processor 80. When the processor 80 executes the computer program 82, the steps in the above-mentioned embodiments of the three-phase inverter midpoint potential control method based on the feedforward neural network are implemented, for example, Figure 1 Steps 101 to 104 are shown.

[0101] The processor 80 may be a central processing unit (CPU), other general-purpose processors, digital signal processors (DSP), application-specific integrated circuits (ASIC), field-programmable gate arrays (FPGA) or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. A general-purpose processor may be a microprocessor or any conventional processor, etc.

[0102] The memory 81 may be an internal storage unit of the control device 8, such as a hard disk or a memory. The memory 81 may also be an external storage device for the control device 8, such as a plug-in hard disk, a smart media card (SMC), a secure digital (SD) card, a flash card, etc. equipped on the control device 8. Further, the memory 81 may also include both an internal storage unit of the control device 8 and an external storage device. The memory 81 is used to store the above-mentioned computer program and other programs and data required by the control device.

[0103] The computer program may be divided into one or more units, which are stored in the memory 81 and executed by the processor 80 to complete the present application. The one or more units may be a series of computer program instruction segments capable of completing specific functions, which are used to describe the process of the computer program executing the three-phase inverter midpoint potential control method based on a feedforward neural network in a control device.

[0104] For example, the above computer program can be divided into: an acquisition unit, a prediction unit, an adjustment unit and a control unit, and the specific functions are as follows:

[0105] An acquisition unit, used for acquiring a real-time feature vector of the three-phase inverter;

[0106] A prediction unit, used for inputting the real-time feature vector into a pre-trained target feedforward neural network model to obtain a predicted value of a midpoint potential offset output by the target feedforward neural network model;

[0107] An adjustment unit, used for adjusting a modulation parameter of SVPWM modulation according to a predicted value of a midpoint potential offset output by the target feedforward neural network model to obtain an adjusted modulation parameter;

[0108] A control unit is used to output a PWM signal for controlling the switching device of the three-phase inverter based on the adjusted modulation parameter to drive the three-phase inverter.

[0109] An embodiment of the present application also provides a computer program product, which stores a computer program. When the computer program is executed by a processor, the steps of the three-phase inverter midpoint potential control method based on a feedforward neural network in any of the above embodiments are implemented.

[0110] The technicians in the relevant field can clearly understand that for the convenience and simplicity of description, only the division of the above-mentioned functional units and modules is used as an example for illustration. In practical applications, the above-mentioned function allocation can be completed by different functional units and modules as needed, that is, the internal structure of the device can be divided into different functional units or modules to complete all or part of the functions described above. The functional units and modules in the embodiment can be integrated in a processing unit, or each unit can exist physically separately, or two or more units can be integrated in one unit. The above-mentioned integrated unit can be implemented in the form of hardware or in the form of software functional units. In addition, the specific names of the functional units and modules are only for the convenience of distinguishing each other, and are not used to limit the scope of protection of this application. The specific working process of the units and modules in the above-mentioned system can refer to the corresponding process in the aforementioned method embodiment, which will not be repeated here.

[0111] In the above embodiments, the description of each embodiment has its own emphasis. For parts that are not described or recorded in detail in a certain embodiment, reference can be made to the relevant descriptions of other embodiments.

[0112] Those of ordinary skill in the art will appreciate that the units and algorithm steps of each example described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are performed in hardware or software depends on the specific application and design constraints of the technical solution. Professional and technical personnel can use different methods to implement the described functions for each specific application, but such implementation should not be considered to be beyond the scope of this application.

[0113] In the embodiments provided in the present application, it should be understood that the disclosed control devices and methods can be implemented in other ways. For example, the control device embodiments described above are merely schematic. For example, the division of modules or units is only a logical function division, and there may be other division methods in actual implementation, such as multiple units or components can be combined or integrated into another system, or some features can be ignored or not executed. Another point is that the mutual coupling or direct coupling or communication connection shown or discussed can be an indirect coupling or communication connection through some interfaces, systems or units, which can be electrical, mechanical or other forms.

[0114] The units described as separate components may or may not be physically separated, and the components shown as units may or may not be physical units, that is, they may be located in one place or distributed on multiple network units. Some or all of the units may be selected according to actual needs to achieve the purpose of the solution of this embodiment.

[0115] In addition, each functional unit in each embodiment of the present application may be integrated into one processing unit, or each unit may exist physically separately, or two or more units may be integrated into one unit. The above-mentioned integrated unit may be implemented in the form of hardware or in the form of software functional units.

[0116] If the integrated module / unit is implemented in the form of a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the present application implements all or part of the processes in the above-mentioned embodiment method, and can also be completed by instructing the relevant hardware through a computer program. The computer program can be stored in a computer-readable storage medium, and the computer program can implement the steps of the above-mentioned various method embodiments when executed by the processor. Among them, the computer program includes computer program code, and the computer program code can be in source code form, object code form, executable file or some intermediate form. The computer-readable medium may include: any entity or recording medium that can carry computer program code, U disk, mobile hard disk, disk, optical disk, computer memory, read-only memory (ROM), random access memory (RAM), electric carrier signal, telecommunication signal and software distribution medium. It should be noted that the content contained in the computer-readable medium can be appropriately increased or decreased according to the requirements of legislation and patent practice in the jurisdiction. For example, in some jurisdictions, according to legislation and patent practice, the computer-readable medium does not include electric carrier signals and telecommunication signals.

[0117] The above embodiments are only used to illustrate the technical solutions of the present application, rather than to limit them. Although the present application has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the aforementioned embodiments, or make equivalent replacements for some of the technical features therein. These modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the embodiments of the present application, and should all be included in the protection scope of the present application.

Claims

1. A three-phase inverter midpoint potential control method based on a feedforward neural network, characterized in that: The three-phase inverter midpoint potential control method comprises: Obtaining a real-time characteristic vector of the three-phase inverter; Inputting the real-time feature vector into a pre-trained target feedforward neural network model to obtain a predicted value of a midpoint potential offset output by the target feedforward neural network model; Adjusting the modulation parameters of the SVPWM modulation according to the predicted value of the midpoint potential offset output by the target feedforward neural network model to obtain the adjusted modulation parameters; A PWM signal for controlling the switching device of the three-phase inverter is output based on the adjusted modulation parameter to drive the three-phase inverter.

2. The method for controlling the midpoint potential of a three-phase inverter according to claim 1, characterized in that: The modulation parameters of the SVPWM modulation include vector selection and vector action time, and the modulation parameters of the SVPWM modulation are adjusted according to the predicted value of the midpoint potential offset output by the target feedforward neural network model, including: A real-time predicted bus midpoint potential change trend is obtained according to the predicted value of the midpoint potential offset output by the target feedforward neural network model, and the vector selection and vector action time of the SVPWM modulation are dynamically adjusted based on the real-time predicted bus midpoint potential change trend.

3. The method for controlling the midpoint potential of a three-phase inverter according to claim 1 or 2, characterized in that: The modulation parameters of the SVPWM modulation include the zero vector action time, and the modulation parameters of the SVPWM modulation are adjusted according to the predicted value of the midpoint potential offset output by the target feedforward neural network model to obtain the adjusted modulation parameters, including: Calculating a zero vector duty cycle correction amount according to the predicted value of the midpoint potential offset and a preset adjustment coefficient; The zero vector action time of the SVPWM modulation is adjusted based on the zero vector duty cycle correction amount to obtain an adjusted zero vector action time.

4. The method for controlling the midpoint potential of a three-phase inverter according to claim 3, characterized in that: The calculating of the zero vector duty cycle correction amount according to the midpoint potential offset prediction value and the preset adjustment coefficient includes: Based on the formula k mod =k p ΔV neutral Calculate the zero vector duty cycle correction; where k mod is the zero vector duty cycle correction, k p is the preset adjustment coefficient, ΔV neutral is the predicted value of the midpoint potential offset.

5. The method for controlling the midpoint potential of a three-phase inverter according to claim 4, characterized in that: The step of adjusting the zero vector action time of the SVPWM modulation based on the zero vector duty cycle correction amount to obtain the adjusted zero vector action time includes: Obtain voltage instructions and calculate reference voltage vector according to the voltage instructions The sector position of the reference voltage vector is determined based on the phase angle θ of the reference voltage vector, and the non-zero vector V is determined based on the sector position of the reference voltage vector. α 、V β and the zero vector; Calculate the non-zero vector V α and V β The action time is T1 and T2; where, Ts is the switching period, V dc is the DC bus voltage, δθ is the angle of the reference voltage vector relative to the starting edge of the sector; Based on T1 and T2, the action time of the zero vector T0=T is calculated. s -T1-T2; Calculate the adjusted zero vector action time T 0- =T0-T 0+ , where k mod is the zero vector duty cycle correction.

6. The method for controlling the midpoint potential of a three-phase inverter according to claim 1 or 2, characterized in that: The training steps adopted by the target feedforward neural network model include: Collecting operating parameters of the three-phase inverter under different load conditions; Preprocessing the operating parameters to obtain a sample feature vector and a midpoint potential offset label value vector; Dividing the sample feature vector and the midpoint potential offset label value vector into a training set and a validation set; Using the training set to train the initial feedforward neural network model, and using the verification set to verify the trained initial neural network model; When the rate of change of the loss function value of the verification set is less than a preset rate of change, the training of the initial neural network model is completed to obtain the target feedforward neural network model.

7. The method for controlling the midpoint potential of a three-phase inverter according to claim 1 or 2, characterized in that: After outputting the PWM signal for controlling the switching device of the three-phase inverter based on the adjusted modulation parameter, the three-phase inverter midpoint potential control method further includes: Real-time monitoring of the actual value of the busbar midpoint potential of the three-phase inverter; Calculating a prediction error of the target feedforward neural network model based on the actual value of the busbar midpoint potential and a predicted value of the midpoint potential offset output by the target feedforward neural network model; Parameters of the target feedforward neural network model are updated based on the prediction error.

8. A control device for a three-phase inverter, characterized in that: The control device comprises: An acquisition unit, used for acquiring a real-time feature vector of the three-phase inverter; A prediction unit, used for inputting the real-time feature vector into a pre-trained target feedforward neural network model to obtain a predicted value of a midpoint potential offset output by the target feedforward neural network model; An adjustment unit, used for adjusting the modulation parameters of the SVPWM modulation according to the predicted value of the midpoint potential offset output by the target feedforward neural network model to obtain the adjusted modulation parameters; A control unit is used to output a PWM signal for controlling the switching device of the three-phase inverter based on the adjusted modulation parameter to drive the three-phase inverter.

9. A control device, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that: When the computer program is executed by a processor, the steps of the three-phase inverter midpoint potential control method based on a feedforward neural network as described in any one of claims 1 to 7 are implemented.

10. A computer program product, characterized in that It comprises a computer program, which, when executed, implements the steps of the three-phase inverter midpoint potential control method based on a feedforward neural network as described in any one of claims 1 to 7.

Images