Fault diagnosis method for multi-source heterogeneous new energy power system based on FFT-ResNet

Abstract

The application discloses a kind of based on FFT-ResNet's multi-source heterogeneous new energy power system fault diagnosis method, it is related to new energy power system fault diagnosis technical field, based on FFT-ResNet's multi-source heterogeneous new energy power system fault diagnosis method mainly includes: based on the model of current control follow network type converter and the model of virtual synchronous generator control network type converter construction follow / construct network coupling new type power system model, set up a variety of faults and simulate, according to simulation result, construct fault database, using lightweight residual network to the electrical signal to be diagnosed is diagnosed, obtains diagnostic result.The application provides based on FFT-ResNet's multi-source heterogeneous new energy power system fault diagnosis method, can realize accurate fault diagnosis under complex scene.

Description

Technical Field

[0001] This invention relates to the field of fault diagnosis technology for new energy power systems, and more specifically, to a fault diagnosis method for multi-source heterogeneous new energy power systems based on FFT-ResNet. Background Technology

[0002] Driven by both the tight supply of fossil fuels and the "dual-carbon" strategy, renewable energy sources such as wind power and photovoltaics have experienced explosive growth. New power systems are gradually forming a "dual-high" development pattern characterized by "high proportion of new energy access and high proportion of power electronic equipment connected to the grid." As the installed capacity of wind and photovoltaic power continues to rise significantly, the diverse power sources, such as photovoltaic inverters, wind power converters, and energy storage devices, exhibit significant differences in topology, control strategies, and dynamic response characteristics. These differences, coupled with the high impedance characteristics of weak end-point grids and the influence of distributed parameters along long lines, result in complex dynamic behaviors of the power system characterized by "low inertia, weak damping, and strong interaction." This significantly increases the probability of system failures and the difficulty of stability management, posing a severe challenge to fault diagnosis. In new power systems with a high proportion of new energy deeply integrated, coupled operation of grid-connected and grid-connected converters has become common. The fundamental difference in their response mechanisms makes fault scenarios more complex, further increasing the difficulty of fault diagnosis.

[0003] Currently, mainstream new energy equipment mainly achieves grid-connected operation through grid-connected or grid-connected control technologies. While these types of equipment have the significant advantage of flexible power regulation, their anti-interference capabilities and grid support capabilities are inferior to traditional synchronous generator sets. Especially in end-point weak grid application scenarios, high impedance characteristics can lead to increased voltage drop duration after a fault and further exacerbate harmonic interference problems. The phase-locked loop system of grid-connected converters is susceptible to instability and grid disconnection due to harmonic interference, while grid-connected converters suffer from prominent problems such as power regulation lag and multi-machine coupled oscillations. In addition, the low inertia and damping level of the system itself further amplifies the risks of voltage instability and synchronization instability, placing higher demands on the accuracy and timeliness of fault diagnosis. However, most existing research focuses on fault condition analysis under independent operation of grid-connected and grid-connected converters, lacking a holistic system perspective to assess fault scenarios when the two are coupled, making it difficult to support the accurate fault diagnosis needs in complex coupled scenarios.

[0004] How to achieve accurate fault diagnosis in complex scenarios is an urgent problem to be solved. Summary of the Invention

[0005] The purpose of this invention is to provide a fault diagnosis method for multi-source heterogeneous new energy power systems based on FFT-ResNet, which can achieve accurate fault diagnosis in complex scenarios.

[0006] This invention provides a fault diagnosis method for multi-source heterogeneous new energy power systems based on FFT-ResNet, comprising the following steps: S1: Based on the grid-connected converter model with current control and the grid-connected converter model with virtual synchronous generator control, a new power system model with grid-connected / grid-connected coupling is constructed. S2: Based on the new power system model coupled with the grid, set up various faults and perform simulations, and construct a fault database based on the simulation results; S3: Based on the fault database, a lightweight residual network is used to diagnose the electrical signals to be diagnosed, and the diagnostic results are obtained.

[0007] The present invention also provides a computer program product, including a computer program that, when executed by a processor, implements the steps of the above-described method for fault diagnosis of multi-source heterogeneous new energy power systems based on FFT-ResNet.

[0008] The fault diagnosis method for multi-source heterogeneous new energy power systems based on FFT-ResNet provided by this invention has the following beneficial effects: This invention takes the large-scale grid connection of multi-source heterogeneous new energy equipment as the background, relies on the essential differences in the fault response mechanisms of grid-connected and grid-connected converters, integrates the dynamic operating characteristics of the two, deeply explores the fault evolution mechanism of new power systems under coupled operation mode, systematically analyzes the changing laws of various electrical parameters after the occurrence of faults, provides theoretical and data support for achieving accurate fault diagnosis in complex scenarios, and then carries out complex fault diagnosis of power systems. Attached Figure Description

[0009] The present invention will be further described below with reference to the accompanying drawings and embodiments. In the accompanying drawings: Figure 1 This is a flowchart of the fault diagnosis method for multi-source heterogeneous new energy power systems based on FFT-ResNet provided by the present invention; Figure 2 This is a diagram of the grid-connected converter topology and control structure based on current control provided by the present invention; Figure 3 This is a block diagram of the phase-locked loop control provided by the present invention; Figure 4 This is the VSG control principle diagram provided by the present invention; Figure 5 This is a network converter aggregation diagram provided by the present invention; Figure 6 This is the equivalent circuit diagram of the grid converter provided by the present invention; Figure 7 This is a schematic diagram of a single-phase grounding fault provided by the present invention; Figure 8This is a block diagram of the ResNet structure provided by the present invention; Figure 9 This is a line graph showing the loss value and accuracy of the FFT-ResNet-based power grid fault diagnosis model provided by this invention. Figure 10 This is a statistical analysis of the accuracy of 17 types of fault tests using the fault diagnosis model provided by this invention. Detailed Implementation

[0010] To provide a clearer understanding of the technical features, objectives, and effects of the present invention, specific embodiments of the present invention will now be described in detail with reference to the accompanying drawings.

[0011] Figure 1 A schematic diagram of the fault diagnosis method for multi-source heterogeneous renewable energy power systems based on FFT-ResNet in this embodiment is shown. In this embodiment, the fault diagnosis method for multi-source heterogeneous renewable energy power systems based on FFT-ResNet includes the following steps: S1: Based on the grid-connected converter model with current control and the grid-connected converter model with virtual synchronous generator control, a new power system model with grid-connected / grid-connected coupling is constructed.

[0012] In one exemplary embodiment, the process of constructing a novel power system model coupled with the grid includes: Establish a grid-connected converter model based on current control; Establish a grid-type converter model based on virtual synchronous generator control; By performing dimensionality reduction on the grid-connected converter model and the grid-connected converter model, a new power system model with grid-connected / grid-connected coupling is obtained.

[0013] In an exemplary embodiment, the dimensionality reduction process of the aggregation model includes: aggregating converters with consistent control modes based on the grid-connected converter model and the grid-connected converter model. The aggregation process follows the principle of dynamic characteristic consistency. By implementing equivalent mapping on the power and impedance parameters of the converters, a centralized equivalent representation of similar converters is achieved. The grid-connected converter is equivalent to a current source, and the grid-connected converter is equivalent to a voltage source. The control strategy adopted by the converter remains unchanged. The aggregated equivalent model is constructed to obtain a new power system model coupled with the grid.

[0014] In one exemplary embodiment, the grid-connected converter model includes a phase-locked loop (PLL) and an AC current loop, and the mathematical modeling equations for the PLL are as follows:

[0015] In the formula, It is the synchronization phase angle of the phase-locked loop output; By adjusting the grid voltage Axial components The angular frequency obtained after performing proportional-integral control; It is the angular frequency of the power grid; These are the proportional parameters of the phase-locked loop; These are the integral parameters of the phase-locked loop; For grid voltage d Axial components; For grid voltage q Axial components.

[0016] In one exemplary embodiment, the grid-type converter model includes an active-frequency control loop and a reactive-voltage control loop, and the mathematical model of the active-frequency control loop is as follows:

[0017] The mathematical model for reactive power-voltage control is as follows:

[0018] In the formula, The virtual moment of inertia parameter for the virtual synchronous generator; The system's rated angular frequency parameter; This is the actual angular frequency; For virtual mechanical power input; It reflects the actual active power output of the virtual synchronous generator; This represents the damping coefficient of the virtual synchronous generator system; This is the active frequency droop factor; The power angle between the output voltage of the virtual synchronous generator and the grid voltage; This indicates the preset active power reference value; This represents the actual output reactive power of the inverter. This is a preset reactive power reference value; This is the reference amplitude of the internal potential of the virtual synchronous generator; The amplitude of the internal potential of the virtual synchronous generator; This is the integral gain.

[0019] S2: Based on the aforementioned new power system model coupled with the grid, various faults are set and simulated, and a fault database is constructed based on the simulation results.

[0020] In one exemplary embodiment, the multiple faults include single-phase ground fault, single-phase open-circuit fault, two-phase ground fault, two-phase open-circuit fault, two-phase short-circuit fault, three-phase open-circuit fault, and three-phase short-circuit fault.

[0021] In one exemplary embodiment, the fault database includes fault type labels and voltage and current time series data at common coupling points.

[0022] S3: Based on the fault database, a lightweight residual network is used to diagnose the electrical signals to be diagnosed, and the diagnostic results are obtained.

[0023] In one exemplary embodiment, step S3 specifically includes: S31: Based on the fault database, the fundamental frequency is extracted and the data is standardized and preprocessed using the Fast Fourier Transform method to obtain the fault dataset; S32: Use the fault dataset to train the lightweight residual network to obtain the trained lightweight residual network; S33: Use the trained lightweight residual network to diagnose the electrical signal to be diagnosed and obtain the diagnostic result.

[0024] In one exemplary embodiment, the lightweight residual network includes an input layer, multiple stacked residual blocks, a global average pooling layer, a fully connected output layer, and a Softmax function; the multiple stacked residual blocks are used for feature extraction and nonlinear transformation, the global average pooling layer is used to compress the feature map into a vector, the fully connected output layer includes 17 nodes corresponding to 17 fault types, and the Softmax function is used to output the probability of each fault.

[0025] In some embodiments, the above-mentioned fault diagnosis method for multi-source heterogeneous new energy power systems based on FFT-ResNet can also be implemented in the following ways.

[0026] In this embodiment, the fault diagnosis method for multi-source heterogeneous new energy power systems based on FFT-ResNet includes the following steps: Step 1: Establish a new power system model coupled with the grid. 1. Establish a grid-connected converter model based on current control. The core control components of a grid-connected converter mainly include phase-locked loops (PLLs), current control loops, and pulse width modulation (PWM) techniques. Among these control modules, the PLL plays the most crucial role, ensuring stable synchronous operation between the converter and the main grid through real-time estimation and dynamic tracking of the voltage phase at the point of common coupling. The control principle structure of a grid-connected converter based on a current control strategy is as follows: Figure 2 As shown. The DC side of this converter is supplied by a constant voltage source. The AC side provides power, while the AC side uses a filter inductor. Connect to the power grid. The voltage at the point of common coupling is... This means that the power grid can be considered as an ideal voltage source. With resistors ,inductance Line impedance The system uses a series structure. The entire control system consists of two core modules: a phase-locked loop (PLL) for grid phase tracking and an alternating current loop (ACC) for power decoupling control. Together, they achieve the overall control objective of grid synchronization and power regulation. The mathematical modeling equations for the PLL are as follows: (2.1) In the formula, It is the synchronization phase angle of the phase-locked loop output; By adjusting the grid voltage Axial components The angular frequency obtained after performing proportional-integral control; It is the angular frequency of the power grid; These are the proportional parameters of the phase-locked loop; These are the integral parameters of the phase-locked loop; For grid voltage d Axial components; For grid voltage q Axial components.

[0027] Phase-locked loop control block diagram as follows Figure 3 The diagram shows the overall control structure of the phase-locked loop (PLL). This PLL uses the three-phase grid voltage... As the input signal, it is first transformed into a voltage component in the dq rotating coordinate system through an abc / dq coordinate transformation. and By maintaining the q-axis voltage component at zero through closed-loop control, the phase and frequency information of the grid voltage can be tracked in real time, ultimately outputting the real-time phase angle of the grid voltage. .

[0028] 2. Establish a grid-type converter model based on virtual synchronous generator control. The control system of a Virtual Synchronous Generator (VSG) mainly consists of an active power control loop, a reactive power control loop, a voltage and current dual closed-loop controller, and a Space Vector Pulse Width Modulation (SVPWM) signal generation and modulation module. The core of this control method focuses on the active and reactive power control loops. By simulating the rotor motion equations of the synchronous generator, the converter achieves dynamic synchronization with the power grid. Grid-connected converters using VSG technology can exhibit external characteristics similar to traditional synchronous generators. The principles of the active-frequency and reactive-voltage control loops are as follows: Figure 4 As shown. The mathematical model of the active-frequency loop is as follows: (2.2) The mathematical model for the reactive power loop-voltage loop is as follows: (2.3) In the formula, For the virtual rotational inertia parameter of VSG; The system's rated angular frequency parameter; This is the actual angular frequency; For virtual mechanical power input; Reflects the actual active power output of the VSG; This represents the damping coefficient of the VSG system; This is the active frequency droop factor; The power angle between the VSG output voltage and the mains voltage; This indicates the preset active power reference value; This represents the actual output reactive power of the inverter. This is a preset reactive power reference value; This is the reference amplitude of the VSG internal potential; The amplitude of the internal potential of the VSG; This is the integral gain.

[0029] 3. Dimensionality reduction of aggregate models In practical power systems, multiple grid-connected converters and multiple network-connected converters often operate in a coupled state. To reduce the complexity of system analysis and control, it is necessary to first aggregate converters with consistent control modes. The aggregation process strictly follows the principle of dynamic characteristic consistency, and achieves a centralized equivalent representation of similar converters by implementing equivalent mapping of converter power and impedance parameters.

[0030] Specifically, the current loop bandwidth of the grid-connected converter is much higher than that of the phase-locked loop, and its current loop dynamic response process can be ignored. Therefore, after aggregation, it is equivalent to a current source. The voltage and current inner loop regulation rate of the grid-connected converter is significantly faster than that of the active power control loop and the reactive power control loop. The dynamic process of the voltage and current inner loop has little impact on the overall characteristics of the system and can be ignored. Therefore, after aggregation, it is equivalent to a voltage source.

[0031] It should be clarified that after aggregation, the converter's control parameters and power reference values ​​will be adjusted accordingly, but the control strategy adopted by the converter remains unchanged, and its core electrical characteristics remain stable. The converter aggregation model is as follows: Figure 5 As shown, the equivalent model after aggregation is as follows: Figure 6 As shown in the figure, This represents the effective value of the output current of a current-controlled converter. This indicates the effective value of the output voltage of the grid-connected converter. yes The phase angle difference between the same grid and the corresponding synchronous generator power angle is referred to as the power angle below for ease of description. For voltage and current The phase angle difference; This represents the effective value of the output voltage of the grid-type converter. The power angle of the grid-type converter; This is the effective value of the grid voltage.

[0032] Step 2: Fault Data Acquisition 1. Set up faults This invention analyzes seventeen typical microgrid faults, selecting their locations around the common coupling point (CCP) of the grid-connected converters. The aim is to deeply explore the impact of different fault types on the voltage and current at the CCP. The fault types covered include single-phase grounding faults, single-phase open-circuit faults, two-phase grounding faults, two-phase open-circuit faults, two-phase short-circuit faults, three-phase open-circuit faults, and three-phase short-circuit faults. Specifically, single-phase grounding faults can be further subdivided into A-phase grounding faults, B-phase grounding faults, and C-phase grounding faults, etc. Other fault types are subdivided in a similar manner, ultimately forming seventeen different fault scenarios.

[0033] 2. Fault Database Construction: To clearly understand the changing characteristics of the electrical parameters at the common coupling point during various faults, a corresponding simulation program was built on the Simulink platform based on the aforementioned mathematical model of the coupled system. To closely approximate actual fault scenarios, 200 sets of gradient simulations were designed for each type of fault. Specifically, the fault resistance for grounding and short-circuit faults was set in an equal increment from 0 to 2 ohms; the fault resistance for open-circuit faults was set to 1 ohm. ~100 A predictable incremental pattern is used. Within a 0.1s time interval following a fault occurrence, voltage and current time-series data at the common coupling point are collected to construct a fault database. The main circuit simulation parameters are detailed in Table 1.

[0034] Table 1: Simulation parameters of the root / network coupled system

[0035] Step 3: Design a fault diagnosis strategy based on FFT-ResNet It should be noted that FFT-ResNet refers to Fast Fourier Transform and Residual Network.

[0036] 1. Frequency domain analysis of voltage and current amplitude extracted by FFT In the field of power electronic grid-connected systems and their fault research, the fundamental frequency components of voltage and current signals contain key information about the system's operating state and fault characteristics. Actual time-domain signals often contain harmonic components and noise interference, making it difficult to accurately represent their steady-state characteristics through direct time-domain analysis. Therefore, frequency-domain analysis methods are essential for signal processing. The Fast Fourier Transform (FFT), as a highly efficient spectrum analysis tool, is widely used in power system signal processing, capable of extracting the fundamental amplitude and phase information of voltage and current signals at the power frequency. Assuming the sampled discrete voltage or current time-domain signal is... The sampling frequency is The number of sampling points is N The FFT can be used to transform a time-domain signal into the frequency domain, and its spectral expression is as follows: (3.1) Among them, frequency index k With actual frequency The correspondence is In power systems, the power frequency is typically 50 Hz, therefore, the frequency index closest to 50 Hz can be used to determine the frequency. Obtain the spectral components of the signal at the power frequency. Based on this spectral component, the fundamental amplitude and phase of the voltage or current signal can be further calculated. The fundamental amplitude is given by the spectral magnitude, and its calculation formula is: (3.2) By performing FFT on the three-phase voltage and current signals, the amplitude and phase information of the fundamental frequency components of each phase can be accurately extracted. Under different fault conditions such as ground fault, short circuit fault, and open circuit fault, the aforementioned fundamental frequency characteristics show significant differences with the change of fault impedance. Generally, ground fault and short circuit faults cause a significant increase in the fundamental current amplitude and a corresponding decrease in the voltage amplitude, while open circuit faults show a sharp decay in the current amplitude and a significant distortion in the phase characteristics. Therefore, extracting the fundamental frequency amplitude and phase characteristics of voltage and current signals based on the FFT method can not only effectively suppress harmonic and noise interference, but also obtain key parameters reflecting the system operating status with low computational complexity, providing reliable data support for fault diagnosis of power electronic systems. Given the complexity of fault types and data volume, this invention selects the most representative single-phase ground fault as the analysis object, and conducts feature analysis on the fundamental frequency amplitude and phase of its A-phase voltage and current under power frequency. The extraction results are as follows: Figure 7 As shown.

[0037] 2. Using ResNet for fault diagnosis Traditional deep neural networks are prone to gradient vanishing or exploding problems as the number of layers increases, making model training difficult. To address this issue, ResNet introduces residual blocks. The key concept is that for each layer, the network learns the residual—the difference between the expected output and the input—rather than directly learning the target output. Residual blocks, through skip connections, allow the input to "skip" certain layers and connect directly to the output of deeper layers, thus preserving input information and mitigating the gradient problem in deep networks. A typical ResNet network mainly consists of an input layer, stacked residual blocks, a global pooling layer, and an output classification layer. The residual block is the core unit of the network, containing a main feature branch composed of convolutions, batch normalization, and activation functions, as well as shortcut branches implemented through identity or 1×1 convolutions. The residuals are learned element-wise by adding these branches together. The residual structure effectively alleviates the gradient vanishing and model degradation problems in deep network training, improving feature extraction capabilities and generalization performance.

[0038] This invention proposes a fault classification method for root / network coupled systems based on Fast Fourier Transform (FFT) fundamental characteristics and Residual Network (ResNet). This method can accurately identify 17 typical power electronic faults, including phase-to-phase short circuits, single-phase grounding, multi-phase grounding, single-phase open circuits, and multi-phase open circuits. The overall process consists of six core stages: data acquisition and FFT fundamental extraction, data standardization preprocessing, dataset partitioning and enhancement, small-scale ResNet network construction, model training, and multi-dimensional performance evaluation. This enables diagnosis from raw electrical signals to fault type output. In the fault diagnosis model, the amplitude and phase angle of the bus voltage and current obtained after FFT processing within a short period after a system fault occur are used as input, with the fault type as the label. The complete data arrangement within the observation window is as follows: (3.3) In the formula, I This refers to the current amplitude. V Voltage amplitude; For the first i Group data j Phase current amplitude; For the first i Group data j Phase current phase angle; No. i Group data j Phase voltage amplitude; For the first i Group data j Phase voltage and phase angle; 1-200 indicates that 200 sets of data have been collected.

[0039] To eliminate dimensional differences and highlight fault-sensitive components, Z-Score normalization is applied to the fundamental characteristics of voltage and current: (3.4) In the formula, The mean and standard deviation of the global voltage samples; The mean and standard deviation of the global current samples are given. As shown in the fault output characteristic diagram above, the current characteristics are more pronounced than the voltage characteristics. Therefore, to enhance the abrupt change characteristics of current in faults, the current standardization results are weighted: (3.5) Finally, the processed voltage feature matrix and current feature matrix are concatenated to form a 12-dimensional feature vector: (3.6) Subsequently, the fault samples of each class were randomly divided into three sets at a ratio of 7:1.5:1.5:Train: 70% for network parameter updates; Validation: 15% for monitoring overfitting and parameter tuning; and Test: 15% for final performance evaluation. To adapt to the input format of the convolutional neural network, the 12-dimensional features were reshaped into a one-dimensional sequence feature map with a height of 12 × width of 1 × channel of 1.

[0040] This invention employs a lightweight residual network, which reduces computational load while maintaining accuracy. The structure is as follows: Figure 8 As shown. The training objective is to minimize the loss function. When applying ResNet to multi-class classification problems, the classification cross-entropy loss function is used: (3.7) In the formula, y For real labels, The model predicts probabilities. After data preprocessing, the merged feature matrix is ​​used as input features for training the neural network. During subsequent training, the neural network performs fault diagnosis based on these 12 features for each sample.

[0041] The accuracy and loss value of performing fault diagnosis tasks are as follows: Figure 9 As shown in the figure. Accuracy represents the diagnostic accuracy, Epoch represents the number of iterations, and Loss represents the loss value. Accuracy rates for various fault tests are shown in the figure. Figure 10As shown in the figure, the power grid fault diagnosis model based on FFT-ResNet exhibits good convergence characteristics and stable generalization ability during training. The training curves show that in the initial training stage (0–200 Epochs), both the training loss and validation loss decrease rapidly, while the training accuracy and validation accuracy increase rapidly, indicating that the network can effectively learn the feature information from the 50 Hz fundamental amplitude and phase features extracted by FFT. As the training progresses (approximately 200–600 Epochs), the loss function continues to decrease slowly, and the model enters a stable convergence phase. The training accuracy gradually approaches 100%, and the validation accuracy also increases to a high level, indicating that the constructed ResNet network structure can fully extract the nonlinear mapping relationship in voltage and current features. In the later training stage (after 600 Epochs), both the training loss and validation loss tend to stabilize and remain at a low level, while the training accuracy and validation accuracy remain basically consistent, without significant divergence or overfitting. This indicates that by using unified normalization preprocessing, low-noise data augmentation, and residual structure design, the model effectively improves its generalization ability to different fault conditions. At the end of training, the model loss value stabilized in a low range, and the validation accuracy was close to 98%–100%, indicating that the network had fully converged.

[0042] The model was further validated using a test set, yielding statistical results on the accuracy of fault identification for 17 categories. The bar chart shows that the accuracy of fault identification for all categories remained above 95%, with most categories approaching 100%. The highest accuracy was achieved for typical short-circuit faults (such as three-phase ABC short circuits) and ground faults (such as ACG, ABG, etc.), indicating that the model can accurately capture the significant differences in electrical quantities under short-circuit and ground fault conditions. The accuracy for some open-circuit faults (such as two-phase AC open circuits and two-phase BC open circuits) was slightly lower, but still remained at a high level, indicating that the model still has good identification capabilities under complex open-circuit fault conditions.

[0043] The combined training process and test results demonstrate that the fault diagnosis method based on the combination of FFT feature extraction and ResNet residual network can effectively extract fault features from the fundamental amplitude and phase information of voltage and current, achieving high-precision identification of 17 types of power grid faults. This method not only has high classification accuracy but also exhibits good stability and convergence during training, providing an effective solution for multi-type fault diagnosis in complex power electronic systems.

[0044] This embodiment provides a computer program product, including a computer program that, when executed by a processor, implements the steps of the above-described method for fault diagnosis of multi-source heterogeneous new energy power systems based on FFT-ResNet.

[0045] The embodiments of the present invention have been described above with reference to the accompanying drawings. However, the present invention is not limited to the specific embodiments described above. The specific embodiments described above are merely illustrative and not restrictive. Those skilled in the art can make many other forms under the guidance of the present invention without departing from the spirit and scope of the claims. All of these forms are within the protection scope of the present invention.

Claims

1. A fault diagnosis method for multi-source heterogeneous new energy power systems based on FFT-ResNet, characterized in that, Includes the following steps: S1: Based on the grid-connected converter model with current control and the grid-connected converter model with virtual synchronous generator control, a new power system model with grid-connected / grid-connected coupling is constructed. S2: Based on the new power system model coupled with the grid, set up various faults and perform simulations, and construct a fault database based on the simulation results; S3: Based on the fault database, a lightweight residual network is used to diagnose the electrical signals to be diagnosed, and the diagnostic results are obtained.

2. The fault diagnosis method for multi-source heterogeneous new energy power systems based on FFT-ResNet according to claim 1, characterized in that, The process of constructing a novel power system model coupled with the grid includes: Establish a grid-connected converter model based on current control; Establish a grid-type converter model based on virtual synchronous generator control; By performing dimensionality reduction on the grid-connected converter model and the grid-connected converter model, a new power system model with grid-connected / grid-connected coupling is obtained.

3. The fault diagnosis method for multi-source heterogeneous new energy power systems based on FFT-ResNet according to claim 1, characterized in that, The dimensionality reduction process of the aggregation model includes: aggregating converters with consistent control modes based on the grid-connected converter model and the grid-connected converter model. The aggregation process follows the principle of dynamic characteristic consistency. By implementing equivalent mapping on the power and impedance parameters of the converters, a centralized equivalent representation of similar converters is achieved. The grid-connected converter is equivalent to a current source, and the grid-connected converter is equivalent to a voltage source. The control strategy adopted by the converter remains unchanged. The aggregated equivalent model is constructed to obtain a new power system model coupled with the grid.

4. The fault diagnosis method for multi-source heterogeneous new energy power systems based on FFT-ResNet according to claim 1, characterized in that, The grid-connected converter model includes a phase-locked loop (PLL) and an AC current loop. The mathematical modeling equations for the PLL are as follows: In the formula, It is the synchronization phase angle of the phase-locked loop output; By adjusting the grid voltage Axial components The angular frequency obtained after performing proportional-integral control; It is the angular frequency of the power grid; These are the proportional parameters of the phase-locked loop; These are the integral parameters of the phase-locked loop; For grid voltage q Axial components.

5. The fault diagnosis method for multi-source heterogeneous new energy power systems based on FFT-ResNet according to claim 1, characterized in that, The grid-type converter model includes an active-frequency control loop and a reactive-voltage control loop. The mathematical model of the active-frequency control loop is as follows: The mathematical model for the reactive power-voltage control is as follows: In the formula, The virtual moment of inertia parameter for the virtual synchronous generator; The system's rated angular frequency parameter; This is the actual angular frequency; For virtual mechanical power input; It reflects the actual active power output of the virtual synchronous generator; This represents the damping coefficient of the virtual synchronous generator system; This is the active frequency droop factor; The power angle between the output voltage of the virtual synchronous generator and the grid voltage; This indicates the preset active power reference value; This represents the actual output reactive power of the inverter. This is a preset reactive power reference value; This is the reference amplitude of the internal potential of the virtual synchronous generator; The amplitude of the internal potential of the virtual synchronous generator; This is the integral gain.

6. The fault diagnosis method for multi-source heterogeneous new energy power systems based on FFT-ResNet according to claim 1, characterized in that, The various faults include single-phase ground fault, single-phase open circuit fault, two-phase ground fault, two-phase open circuit fault, two-phase short circuit fault, three-phase open circuit fault, and three-phase short circuit fault.

7. The fault diagnosis method for multi-source heterogeneous new energy power systems based on FFT-ResNet according to claim 1, characterized in that, The fault database includes fault type labels and time series data of voltage and current at common coupling points.

8. The fault diagnosis method for multi-source heterogeneous new energy power systems based on FFT-ResNet according to claim 1, characterized in that, Step S3 specifically includes: S31: Based on the fault database, the fundamental frequency is extracted and the data is standardized and preprocessed using the Fast Fourier Transform method to obtain the fault dataset; S32: Use the fault dataset to train the lightweight residual network to obtain the trained lightweight residual network; S33: Use the trained lightweight residual network to diagnose the electrical signal to be diagnosed and obtain the diagnostic result.

9. The fault diagnosis method for multi-source heterogeneous new energy power systems based on FFT-ResNet according to claim 1, characterized in that, The lightweight residual network includes an input layer, multiple stacked residual blocks, a global average pooling layer, a fully connected output layer, and a Softmax function. The multiple stacked residual blocks are used for feature extraction and nonlinear transformation. The global average pooling layer is used to compress the feature map into a vector. The fully connected output layer includes 17 nodes corresponding to 17 fault types. The Softmax function is used to output the probability of each fault.

10. A computer program product, comprising a computer program, characterized in that, When the computer program is executed by the processor, it implements the steps of the fault diagnosis method for multi-source heterogeneous new energy power systems based on FFT-ResNet as described in any one of claims 1-9.

Images