Fault diagnosis method for low-voltage flexible DC voltage regulator based on artificial intelligence

Abstract

The invention discloses a low-voltage flexible DC voltage regulator fault diagnosis method based on artificial intelligence, and relates to the technical field of intelligent electric power, and the method comprises the steps: collecting multi-modal operation data through a distributed sensor array, and carrying out the constraint verification and cleaning of the multi-modal data based on the electric power physical law; inputting the multi-modal operation data into a multi-modal fusion network, enabling the voltage-current characteristics to accord with the Ohm law according to a cross-modal attention mechanism, and correcting the characteristic weight through differentiable constraint loss; applying constraint disturbance in the feature space by adopting a fast gradient symbol method and a projection gradient descent method to obtain a countering feature set; constructing a training set, and performing dynamic confrontation defense by adopting a confrontation training framework and using a meta-learning mechanism; and inputting real-time multi-modal data into the fault classification model, outputting fault probability distribution, setting a fault threshold value, and comparing to give an alarm. According to the invention, intelligent fault diagnosis of complex working conditions is realized through combination of adversarial training and a meta-learning mechanism.

Description

Technical Field

[0001] The present invention relates to the field of intelligent power technology, and particularly to a fault diagnosis method for a low-voltage flexible direct current regulator based on artificial intelligence. Background Art

[0002] As a core device of the smart grid, the low-voltage flexible direct current regulator is threatened by faults such as short circuits, overheating, and insulation aging caused by its complex working conditions, while traditional diagnosis methods face certain limitations. Existing detection means based on rule matching, threshold analysis, or manual judgment are difficult to achieve high-precision real-time diagnosis in a complex electromagnetic environment, and the single-sensor mode leads to the lack of dimensions of the state monitoring data. At the same time, the feature conflicts of multi-source heterogeneous data (such as voltage / current / temperature / humidity, etc.) exacerbate the difficulty of multi-modal fusion. Traditional machine learning and single deep learning models are sensitive to feature distribution deviations, have weak anti-environment interference ability, and lack an adversarial defense mechanism, resulting in a sharp drop in the generalization performance of the fault classification model under dynamic working conditions. There is an urgent need to construct an intelligent diagnosis framework that integrates multi-modal perception data and has the ability of adaptive feature learning and robust decision-making. By deeply mining the correlation of multi-dimensional fault features, the dynamic diagnosis bottleneck under complex power scenarios is broken through, so as to improve the operation reliability of the power grid system. Summary of the Invention

[0003] In view of the above existing problems, the present invention is proposed.

[0004] Therefore, the present invention provides a fault diagnosis method for a low-voltage flexible direct current regulator based on artificial intelligence to solve the problem of insufficient generalization ability of traditional fault classification models under complex working conditions.

[0005] To solve the above technical problems, the present invention provides the following technical solutions: In the first aspect, the present invention provides a fault diagnosis method for a low-voltage flexible direct current regulator based on artificial intelligence, which includes collecting multi-modal operation data through a distributed sensor array, where the multi-modal operation data includes the voltage, current, temperature, and humidity of the regulator, and performing constraint verification and cleaning on the multi-modal data based on the laws of electric power physics; inputting the multi-modal operation data into a multi-modal fusion network, extracting high-frequency ripple features, temperature and humidity features, and time-domain statistics, making the voltage-current features conform to Ohm's law according to the cross-modal attention mechanism, and correcting the feature weights through a differentiable constraint loss to generate a multi-modal feature vector; adopting the fast gradient sign method and the projected gradient descent method to apply constraint perturbations in the feature space to obtain an adversarial feature set; mixing the multi-modal feature vector with the adversarial feature set to construct a training set, and using an adversarial training framework and a meta-learning mechanism for dynamic adversarial defense to obtain a fault classification model; inputting the real-time multi-modal data into the fault classification model, outputting a fault probability distribution, setting a fault threshold, and comparing for alarm.

[0006] As a preferred solution of the fault diagnosis method for the low-voltage flexible direct current voltage regulator based on artificial intelligence according to the present invention, wherein: the constraint verification and cleaning of multi-modal data based on electrical physical laws includes the following steps, Based on Kirchhoff's law and the principle of power conservation, perform consistency verification on voltage and current time-series data; Adopt the heat conduction equation constraint, and dynamically calibrate the temperature data in combination with the radiator parameters of the voltage regulator.

[0007] As a preferred solution of the fault diagnosis method for the low-voltage flexible direct current voltage regulator based on artificial intelligence according to the present invention, wherein: the multi-modal fusion network includes, Extract high-frequency ripple features and time-domain statistics from voltage and current data through a one-dimensional convolutional neural network; Capture the dynamic thermal response mode from temperature data using a long short-term memory network; Combine humidity data and the heat dissipation parameters of the voltage regulator to generate a heat dissipation efficiency correction coefficient.

[0008] As a preferred solution of the fault diagnosis method for the low-voltage flexible direct current voltage regulator based on artificial intelligence according to the present invention, wherein: the cross-modal attention mechanism refers to using a physical constraint attention gating unit to achieve a hard constraint of Ohm's law for voltage-current features.

[0009] As a preferred solution of the fault diagnosis method for the low-voltage flexible direct current voltage regulator based on artificial intelligence according to the present invention, wherein: the application of constraint perturbation refers to generating adversarial perturbations in the feature space, and the perturbation range is restricted by physical constraints.

[0010] As a preferred solution of the fault diagnosis method for the low-voltage flexible direct current voltage regulator based on artificial intelligence according to the present invention, wherein: the adversarial training framework refers to adopting a dual-channel adversarial training architecture. The main channel optimizes the fault classifier through cross-entropy loss, and the adversarial channel updates the perturbation generator through a max-min optimization objective. The dynamic balance factor decays exponentially with the number of training rounds.

[0011] As a preferred solution of the fault diagnosis method for the low-voltage flexible direct current voltage regulator based on artificial intelligence according to the present invention, wherein: the meta-learning mechanism includes the following steps, Extract fault scenarios from the historical fault library as the support set and new faults as the query set; Perform fast adaptation through inner-loop updates and combine with outer-loop optimization.

[0012] As a preferred solution of the fault diagnosis method for the low-voltage flexible direct current voltage regulator based on artificial intelligence according to the present invention, wherein: the fault threshold refers to adopting a dynamic threshold adjustment strategy and combining the heat dissipation efficiency correction coefficient to achieve condition-adaptive alarm triggering.

[0013] In a second aspect, the present invention provides a computer device, including a memory and a processor, where the memory stores a computer program, and: when the computer program is executed by the processor, any step of the fault diagnosis method for a low-voltage flexible DC voltage regulator based on artificial intelligence as described in the first aspect of the present invention is implemented.

[0014] In a third aspect, the present invention provides a computer-readable storage medium, on which a computer program is stored, and: when the computer program is executed by the processor, any step of the fault diagnosis method for a low-voltage flexible DC voltage regulator based on artificial intelligence as described in the first aspect of the present invention is implemented.

[0015] The beneficial effects of the present invention are as follows: Using power electronic devices to replace traditional mechanical transmission components reduces noise and vibration and decreases the impact on the environment. The high energy utilization efficiency helps reduce energy waste and carbon emissions. The device has a smaller volume and weight, which is convenient for installation and transportation, and its control algorithm and mechanical structure also have good adaptability. Through the built-in sensors and artificial intelligence control algorithm, real-time monitoring and intelligent fault diagnosis of the device are achieved, improving the maintenance efficiency and operation stability of the device. The low-voltage flexible DC voltage regulator adopts advanced intelligent control and real-time monitoring technologies in core technologies, featuring high efficiency, high protection, and multi-functional integration. BRIEF DESCRIPTION OF THE DRAWINGS

[0016] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings required for use in the description of the embodiments will be briefly introduced below. Obviously, the drawings in the following description are only some embodiments of the present invention, and those of ordinary skill in the art can obtain other drawings based on these drawings without creative efforts.

[0017] Figure 1 It is a schematic diagram of the overall architecture of the fault diagnosis method for a low-voltage flexible DC voltage regulator based on artificial intelligence in Embodiment 1.

[0018] Figure 2 It is a flowchart of data acquisition and preprocessing of the fault diagnosis method for a low-voltage flexible DC voltage regulator based on artificial intelligence in Embodiment 1.

[0019] Figure 3 It is a structural diagram of a multi-modal fusion network of the fault diagnosis method for a low-voltage flexible DC voltage regulator based on artificial intelligence in Embodiment 1. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0020] To make the above objects, features, and advantages of the present invention more obvious and understandable, the following detailed description of the specific embodiments of the present invention will be given in conjunction with the accompanying drawings of the specification.

[0021] In the following description, numerous specific details are set forth to provide a thorough understanding of the present invention. However, the present invention may be practiced in other ways different from those described herein. Persons skilled in the art can make similar extensions without departing from the spirit of the present invention. Therefore, the present invention is not limited by the specific embodiments disclosed below.

[0022] Secondly, as used herein, an "embodiment" or "embodiments" refers to specific features, structures, or characteristics that may be included in at least one implementation of the present invention. The phrase "in one embodiment" appearing in different places in this specification does not necessarily refer to the same embodiment, nor is it an individual or alternative embodiment that is mutually exclusive with other embodiments.

[0023] Embodiment 1, referring to Figures 1 to 3 , is the first embodiment of the present invention. This embodiment provides a fault diagnosis method for a low-voltage flexible direct current voltage regulator based on artificial intelligence, including the following steps: S1: Collect multi-modal operation data through a distributed sensor array. The multi-modal operation data includes the voltage, current, temperature, and humidity of the voltage regulator, and perform constraint verification and cleaning on the multi-modal data based on electrical physics laws.

[0024] Specifically, it includes the following steps: S1.1: Deploy a distributed sensor array.

[0025] Specifically, use a high-frequency voltage sensor to collect voltage data, with a range of ±500 VDC, a sampling rate of 2 MHz (meeting the 50 kHz ripple capture), an accuracy of ±0.1% FS, and deploy it at the positive and negative poles of the DC bus (two redundant sensors are configured for each phase).

[0026] Use a Rogowski coil current meter to collect current data, with a bandwidth of DC - 200 kHz, a linearity error <0.05%, a temperature drift of ±10 ppm / °C, and deploy it at the input / output terminals of the IGBT module.

[0027] Use an optical fiber temperature sensor to collect temperature data, with a measurement range of -40°C to +150°C, an accuracy of ±0.3°C, a response time <10 ms, and deploy it on the power device substrate (configuring three-point temperature measurement).

[0028] Use a MEMS humidity sensor to collect humidity data, with a range of 0 to 100% RH, an accuracy of ±1.5% RH, and deploy it inside the sealed cavity of the device (diagonally arranged dual sensors).

[0029] Adopt the IEEE 1588 Precision Time Protocol (PTP) for data synchronization to achieve multi-channel data time stamp alignment (error <1 μs).

[0030] S1.2: Perform constraint verification on multi-modal data based on electrical physical laws.

[0031] Specifically, use Kirchhoff's law to perform current verification and calculate the impedance deviation rate for each sampling point. The expression is: ; where is the impedance deviation rate, is the voltage data, is the current data, is the calibrated resistance, is the time variable.

[0032] When the impedance deviation rate is greater than 5% and persists for more than 10 μs, it is marked as abnormal data.

[0033] Perform constraints through the three-dimensional heat conduction equation. The expression is: ; where is the temperature field, is the partial derivative identifier, is the thermal diffusivity, determined by the radiator material, is the spatial second derivative term, describing the conduction rate of temperature in three-dimensional space, is the heat loss term, is the volumetric heat capacity of the material, is the temperature, , , are the three orthogonal directions that make up the three-dimensional rectangular coordinate system.

[0034] S1.3: Perform abnormal data discrimination and processing.

[0035] Specifically, for current and voltage data, if 5 consecutive sampling points exceed the range of the average value ± 3 times the historical standard deviation, or the impedance deviation rate is greater than 5% and persists for more than 10 μs, it is determined as abnormal and replaced with the sliding average value of the previous 1 ms.

[0036] For temperature data, if and persists for more than 10 ms, it is determined as abnormal, marked, and switched to the backup sensor.

[0037] For humidity data, if the difference between adjacent sensors > 5%RH, it is determined as abnormal, and the humidity data is fused using Kalman filtering.

[0038] Preferably, by deploying high-frequency voltage sensors, Rogowski coil current meters, fiber optic temperature sensors, and redundant MEMS humidity sensor arrays, and combining with the IEEE 1588 protocol to achieve multi-channel data synchronization at the microsecond level, a high-precision multi-modal data acquisition system is constructed; based on the impedance deviation rate verification of Kirchhoff's law and the dynamic constraint of the three-dimensional heat conduction equation, the circuit physical consistency verification of voltage and current and the matching of the temperature field evolution law are realized; through the sliding window statistical threshold, the thermal change rate threshold, and the redundant sensor difference detection, a multi-dimensional anomaly discrimination mechanism is constructed, and intelligent repair strategies such as sliding mean replacement and Kalman filter fusion are adopted to improve the accuracy, integrity, and real-time performance of data acquisition in complex power environments, providing a highly reliable data basis for power network state monitoring and fault warning.

[0039] S2: Input the multi-modal operation data into the multi-modal fusion network, extract high-frequency ripple features, temperature and humidity features, and time-domain statistics. According to the cross-modal attention mechanism, make the voltage-current features conform to Ohm's law, and correct the feature weights through differentiable constraint loss to generate multi-modal feature vectors.

[0040] Specifically, it includes the following steps: S2.1: Use a 1D-CNN network to extract high-frequency ripple features from voltage and current data.

[0041] Specifically, the one-dimensional convolution is set to have dual input channels for voltage and current, the number of output channels is 32, the convolution kernel size is 5, and the convolution stride is 2.

[0042] Select the ReLU activation function, use the pooling layer for dimensionality reduction, extract more advanced features through the second layer of convolution, and normalize to prevent overfitting.

[0043] The expression for the high-frequency ripple feature is: ; Where is the high-frequency ripple feature, is the convolution kernel index, is the maximum value identifier, is the th time series output of the convolution kernel, is the Sigmoid activation function, is the pooling layer weight coefficient, which is automatically optimized through backpropagation, is the output feature of the pooling layer.

[0044] Extract the voltage high-frequency ripple feature and the current high-frequency ripple feature .

[0045] S2.2: Extract the time-domain statistics of voltage and current data.

[0046] It should be noted that the time-domain statistical quantities include the waveform factor, kurtosis, impulse factor, and margin factor of voltage and current data.

[0047] The waveform factor is calculated by the ratio of the peak value and the root mean square value of voltage and current data, reflecting the relationship between the peak value and the effective value. A larger waveform factor indicates the existence of larger mutations or impulses.

[0048] By calculating the mean value and standard deviation of voltage and current data, calculating the fourth-power deviation of each sample point relative to the mean value, and finally calculating the mean value and normalizing it to obtain the kurtosis. The larger the kurtosis, the more likely there are larger impulses or abnormal data points in the voltage and current data.

[0049] By calculating the peak value and the average absolute value of voltage and current data, calculating the ratio, and obtaining the impulse factor, which reflects whether the voltage and current data has prominent impulse components. A larger impulse factor usually indicates that the voltage and current data has strong short-time shocks.

[0050] By calculating the root mean square value of the envelope of voltage and current data through Hilbert transform, and combining with the peak value to calculate the margin factor, which measures the ratio of the maximum peak value of voltage and current data relative to the envelope RMS. A high margin factor indicates that the voltage and current data may contain high-amplitude transient events.

[0051] S2.3: Use BiLSTM to extract thermal dynamic features from temperature data.

[0052] Specifically, construct a bidirectional LSTM (BiLSTM) network with the input feature dimension set to 1, the hidden layer dimension to 32, and use bidirectional LSTM. Use the attention mechanism with the dimension of the query vector (Query) set to 64, the dimension of the key vector (Key) set to 64, and the normalization coefficient to 0.1.

[0053] The expression of the thermal dynamic feature is: ; where is the thermal dynamic feature, is the humidity change rate, is the natural base, is the device startup timestamp.

[0054] S2.4: Use the cross-modal attention mechanism to correct voltage-current features.

[0055] Specifically, based on the high-frequency ripple features of current and voltage, generate current-voltage query vectors, key vectors, and value vectors through linear transformation.

[0056] Calculate the attention score of voltage to current through scaled dot - product attention, using the calibrated resistance as prior knowledge to constrain the attention output. The expression is: ; where, is to constrain the attention output, is the query vector of current - voltage, is the key vector, is the value vector, is the feature dimension, is the normalization function.

[0057] Fuse the original current - voltage features and the constrained attention output through a gating mechanism. The expression is: ; ; where, is the gating coefficient, is the gating weight matrix, obtained through training, is the gating bias term, is the final fused feature, is the element - wise multiplication.

[0058] The finally corrected voltage - current feature is , whose implicit relationship satisfies Ohm's law and retains high - frequency ripple details.

[0059] S2.5: Correct the feature weights through a differentiable constraint loss.

[0060] Specifically, the physical constraint loss function expression is: ; where, is the physical constraint loss function, is the number of sampling points within the time window, is the sampling point number index, is the th sample's voltage data, is the th sample's current data.

[0061] Strengthen the Ohm's law constraint through the square term, and balance the degrees of freedom and physical consistency through the exponential term.

[0062] Preferably, through the combination of deep learning and physical constraints, multi-modal data fusion is achieved by high-frequency ripple feature extraction, time-domain statistical analysis, BiLSTM thermal dynamic modeling, cross-modal attention to correct voltage-current features, and differentiable constraint loss optimization. This solution can not only accurately extract the operating characteristics of the device and improve the prediction accuracy, but also ensure that the features conform to Ohm's law through physical constraints, thereby enhancing the credibility and generalization ability of the method. At the same time, end-to-end optimization reduces the computational complexity and makes the inference more efficient.

[0063] S3: Apply the fast gradient sign method and the projected gradient descent method to impose constrained perturbations in the feature space to obtain an adversarial feature set.

[0064] Specifically, it includes the following steps: S3.1: Use the fast gradient sign method (FGSM) to generate an initial perturbation.

[0065] Specifically, calculate the loss gradient of the input final fusion feature through backpropagation, use the fast gradient sign method to generate the perturbation direction and apply the perturbation to generate an initial perturbation sample.

[0066] S3.2: Use the projected gradient descent method (PGD) to iteratively optimize the perturbation.

[0067] It should be understood that the initial perturbation sample generated by FGSM is too simple to simulate complex adversarial attacks. Therefore, the projected gradient descent method (PGD) is further used for optimization to make the perturbation more deceptive while performing range limitation.

[0068] Specifically, based on the initial perturbation generated by FGSM, perform multiple iterative updates. Each time, adjust the perturbation direction according to the gradient information of the current high-frequency ripple feature, temperature and humidity feature, and time-domain statistic to obtain the current perturbation feature. Each time during iteration, calculate the loss of the current perturbation feature input and calculate the loss gradient. Based on the loss gradient, update the perturbation amount to make it move in the direction of increasing the error.

[0069] To ensure that the perturbation does not destroy the overall physical characteristics of the feature, after each update, the perturbation needs to be restricted within a reasonable range to ensure that it does not exceed the defined range. The projection method is used to project the perturbation back to the specified defined range to limit the maximum value of the perturbation or ensure that the perturbation does not exceed the statistical deviation.

[0070] Continue to execute the above process until the loss value of the current perturbation feature converges or reaches the maximum number of iterations.

[0071] S3.2: Apply perturbations to the final fusion features respectively to generate an adversarial feature set.

[0072] Preferably, adversarial perturbations are applied in the feature space through FGSM and PGD to enhance the robustness and defense capabilities of the fault classification model. FGSM generates initial perturbations to improve the fault classification model's ability to handle small perturbations, while PGD further optimizes the perturbations to make them more aggressive, simulating complex adversarial attack scenarios. To maintain physical consistency, the perturbations are restricted in range during the PGD process to ensure that the physical meaning of the features remains unchanged. By applying perturbations to different feature modalities separately, an adversarial feature set can be generated, which can be used for fault classification model training, robustness evaluation, and security analysis, enabling the fault classification model to maintain stable and efficient prediction capabilities in complex environments.

[0073] S4: Mix the multi-modal feature vectors with the adversarial feature set to construct a training set, adopt an adversarial training framework, and use a meta-learning mechanism for dynamic adversarial defense to obtain a fault classification model.

[0074] Specifically, it includes the following steps: S4.1: Mix the multi-modal feature vectors with the adversarial feature set to construct a training set.

[0075] Specifically, the multi-modal feature vectors and the adversarial feature set are mixed in a ratio of 3:1 to ensure that the fault classification model can learn normal features and adapt to adversarial interference.

[0076] S4.2: Construct an adversarial training framework.

[0077] Specifically, the cross-entropy loss is used as the loss function, the adversarial perturbation loss is added, and the joint training objective is set.

[0078] The expression of the loss function is: ; Where, is the loss function, is the training set, is the predicted probability distribution.

[0079] The dynamic adversarial training process uses an alternating training strategy: in the first stage, the original data is trained, the adversarial sample generator is fixed, and only the multi-modal feature vectors are used to update the parameters of the fault classification model. In the second stage, adversarial training is carried out, the fault classification model is fixed, and new adversarial samples are generated to update the training set. The first stage and the second stage are alternately executed to gradually improve the robustness of the fault classification model.

[0080] S4.3: Use a meta-learning mechanism for dynamic adversarial defense.

[0081] Specifically, a batch of data is randomly sampled from the training set as the support set, and new generated adversarial perturbations are applied to the support set to obtain the query set.

[0082] The expression of the meta-loss function is as follows: ; Among them, is the meta-loss function, are the parameters of the fault classification model, is the support set, is the cloud learning weight, set to 1.2, which strengthens the generalization ability of the fault classification model for unknown attacks, is the query set.

[0083] The fault classification model parameters are updated by calculating the meta-gradient, and the expression is: ; ; Among them, is the meta-gradient, is the gradient of calculating the meta-loss for the fault classification model parameters, representing the optimization direction of the fault classification model, is the meta-learning rate (set to 0.001), which controls the update step of the fault classification model parameters and affects the speed at which the fault classification model learns new tasks.

[0084] Furthermore, the perturbation statistics of the input features (such as the feature norm deviation) are monitored in real time. If a perturbation is detected (the deviation exceeds the threshold of 0.1), the meta-learning optimizer is activated to quickly adjust the fault classification model parameters.

[0085] Preferably, the multi-modal feature vectors and the adversarial feature set are mixed at a ratio of 3:1 to ensure that the fault classification model can learn both normal operating conditions and adapt to adversarial perturbations. A dynamic adversarial training strategy is adopted, with normal training and adversarial training alternating, so that the fault classification model can still maintain high classification performance when facing unseen attacks. Finally, a meta-learning mechanism is introduced to optimize the fault classification model through the support set and the adversarial query set, enabling the fault classification model to quickly adapt to new types of attacks and adjust parameters in real time when abnormal input features are detected, improving the generalization ability.

[0086] S5: Input the real-time multi-modal data into the fault classification model, output the fault probability distribution, set the fault threshold and compare for alarm.

[0087] Specifically, it includes the following steps: S5.1: Set the fault threshold.

[0088] Specifically, the expression of the fault threshold is: ; Among them, is the fault threshold of the th type of fault, is the fault category index, is the historical average failure probability of the nth type of failure, is the standard deviation of the nth type of failure, is the empirical factor, with a value between 2 and 3.

[0089] S5.2: Compare the failure probability distribution with the failure threshold for alarming.

[0090] It should be understood that if the predicted failure probability is greater than the failure threshold, it is determined that a failure has occurred and an alarm is triggered; otherwise, it is determined that there is no failure and no alarm is triggered.

[0091] The alarm content includes the failure type, the failure occurrence probability, the trigger timestamp, the relevant equipment and area. Local alarm is carried out and a remote notification is sent to the maintenance personnel. The alarm information is recorded in the log for future analysis and inspection.

[0092] Preferably, by setting a failure threshold based on historical data and standard deviation and inputting real-time multimodal data into the failure classification model, the probability of failure occurrence can be accurately judged and an alarm can be triggered. This method effectively reduces false alarms and missed alarms, ensuring the accuracy and real-time response of failure detection. At the same time, through the combination of local alarm and remote notification, the failure information is conveyed to the maintenance personnel in a timely manner, and a detailed alarm log is recorded for subsequent analysis and optimization. The adaptive adjustment of the threshold, the intelligent decision-making mechanism and the failure prediction function help to improve the stability and reliability of the equipment, reduce sudden failures and improve the maintenance efficiency.

[0093] This embodiment also provides a computer device applicable to the situation of the failure diagnosis method of the low-voltage flexible DC voltage regulator based on artificial intelligence, including: a memory and a processor; the memory is used to store computer-executable instructions, and the processor is used to execute the computer-executable instructions to implement the failure diagnosis method of the low-voltage flexible DC voltage regulator based on artificial intelligence as proposed in the above embodiment.

[0094] The computer device can be a terminal, which includes a processor, a memory, a communication interface, a display screen, and an input device connected via a system bus. Among them, the processor of the computer device is used to provide computing and control capabilities. The memory of the computer device includes a non-volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system and computer programs. The internal memory provides an environment for the operation of the operating system and computer programs in the non-volatile storage medium. The communication interface of the computer device is used to communicate with external terminals in a wired or wireless manner, and the wireless manner can be achieved through WIFI, carrier networks, NFC (Near Field Communication), or other technologies. The display screen of the computer device can be a liquid crystal display screen or an electronic ink display screen, and the input device of the computer device can be a touch layer covering the display screen, or buttons, trackballs, or touchpads provided on the outer shell of the computer device, or an external keyboard, touchpad, or mouse, etc.

[0095] This embodiment also provides a storage medium, on which a computer program is stored. When the program is executed by a processor, it implements the method for fault diagnosis of a low-voltage flexible DC voltage regulator based on artificial intelligence as proposed in the above embodiment; the storage medium can be implemented by any type of volatile or non-volatile storage device or a combination thereof, such as static random access memory (abbreviated as SRAM), electrically erasable programmable read-only memory (abbreviated as EEPROM), erasable programmable read-only memory (abbreviated as EPROM), programmable read-only memory (abbreviated as PROM), read-only memory (abbreviated as ROM), magnetic memory, flash memory, magnetic disks, or optical discs.

[0096] In summary, the present invention: replaces traditional mechanical transmission components with power electronic devices, reduces noise and vibration, and reduces the impact on the environment. The high energy utilization efficiency helps to reduce energy waste and carbon emissions. The device has a small volume and weight, which is convenient for installation and transportation, and its control algorithm and mechanical structure also have good adaptability. Through the built-in sensors and artificial intelligence control algorithm, real-time monitoring and intelligent fault diagnosis of the device are realized, improving the maintenance efficiency and operation stability of the device. The low-voltage flexible DC voltage regulator adopts advanced intelligent control and real-time monitoring technologies in core technologies, and has characteristics such as high efficiency, high protection, and multi-functional integration.

[0097] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and not to limit them. Although the present invention has been described in detail with reference to the preferred embodiments, those of ordinary skill in the art should understand that the technical solutions of the present invention can be modified or equivalently replaced without departing from the spirit and scope of the technical solutions of the present invention, and they should all be covered within the scope of the claims of the present invention.

Claims

1. A fault diagnosis method for a low voltage flexible direct current regulator based on artificial intelligence, characterized in that: include, Collecting multimodal operation data through a distributed sensor array, the multimodal operation data including voltage, current, temperature and humidity of the voltage regulator, and performing constraint verification and cleaning on the multimodal data based on the laws of electrical physics; The multimodal operation data is input into the multimodal fusion network to extract high-frequency ripple features, temperature and humidity features, and time domain statistics. The voltage-current features are made to conform to Ohm's law based on the cross-modal attention mechanism. The feature weights are corrected through the differentiable constraint loss to generate a multimodal feature vector. The fast gradient sign method and projected gradient descent method are used to impose constrained perturbations in the feature space to obtain adversarial feature sets; Mix the multimodal feature vector with the adversarial feature set to construct a training set, adopt the adversarial training framework and use the meta-learning mechanism for dynamic adversarial defense to obtain a fault classification model; Input real-time multimodal data into the fault classification model, output the fault probability distribution, set the fault threshold and compare it to issue an alarm.

2. The artificial intelligence-based low voltage flexible direct current regulator fault diagnosis method according to claim 1, characterized in that: The constraint checking and cleaning of multimodal data based on the laws of electrical physics includes the following steps: Based on Kirchhoff's law and the principle of power conservation, the voltage and current time series data are checked for consistency; The temperature data is dynamically calibrated by using heat conduction equation constraints and combining the regulator heat sink parameters.

3. The artificial intelligence-based low voltage flexible direct current regulator fault diagnosis method according to claim 1, characterized in that: The multimodal fusion network comprises: From the voltage and current data, high-frequency ripple features and time-domain statistics are extracted through a one-dimensional convolutional neural network; From the temperature data, a long short-term memory network is used to capture the dynamic thermal response pattern; Combine the humidity data with the regulator heat dissipation parameters to generate a heat dissipation efficiency correction factor.

4. The artificial intelligence-based low voltage flexible direct current regulator fault diagnosis method according to claim 1, characterized in that: The cross-modal attention mechanism refers to the use of a physical constraint attention gating unit to implement Ohm's law hard constraints on voltage-current characteristics.

5. The artificial intelligence-based low voltage flexible direct current regulator fault diagnosis method according to claim 1, characterized in that: The imposing constrained disturbance refers to generating adversarial disturbance in the feature space, and the disturbance range is limited by physical constraints.

6. The artificial intelligence-based low voltage flexible direct current regulator fault diagnosis method according to claim 1, characterized in that: The adversarial training framework refers to a dual-channel adversarial training architecture, in which the main channel optimizes the fault classifier through cross entropy loss, the adversarial channel updates the perturbation generator through the maximum-minimum optimization objective, and the dynamic balance factor decays exponentially with the training rounds.

7. The artificial intelligence-based low voltage flexible direct current regulator fault diagnosis method according to claim 1, characterized in that: The meta-learning mechanism includes the following steps: Extract fault scenarios from the historical fault database as the support set and new faults as the query set; Rapid adaptation is achieved through inner loop updates, combined with outer loop optimization.

8. The artificial intelligence-based low voltage flexible direct current regulator fault diagnosis method according to claim 1, characterized in that: The fault threshold refers to the use of a dynamic threshold adjustment strategy combined with a heat dissipation efficiency correction coefficient to achieve an alarm trigger that is adaptive to the working condition.

9. A computer device comprising a memory and a processor, wherein the memory stores a computer program, wherein: When the processor executes the computer program, the steps of the artificial intelligence-based low-voltage flexible direct current regulator fault diagnosis method described in any one of claims 1 to 8 are implemented.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the steps of the artificial intelligence-based low-voltage flexible direct current regulator fault diagnosis method described in any one of claims 1 to 8 are implemented.

Images