Transformer substation fault detection method and related equipment

Abstract

The invention discloses a transformer substation fault detection method and related equipment, and the method comprises the steps: obtaining the real-time data of a transformer substation secondary system, forming a scene prediction model through a deep learning model and a dynamic Bayesian network, and determining a fault detection result in combination with a digital twin system. The deep learning model can deeply mine multi-source real-time data features, accurately recognize fault information in a complex scene, make up for the problem of insufficient information mining of a traditional data driving method, and improve the accuracy of fault detection and positioning. The dynamic Bayesian network carries out probability modeling on the dynamic change of the system state, and the problems that a traditional method is difficult to adapt to complex working conditions and cannot accurately simulate system evolution are solved. Meanwhile, the digital twin system realizes deep fusion analysis of multi-source data, constructs an intelligent processing closed loop from data acquisition to maintenance decision, remarkably improves the timeliness, accuracy and intelligent level of fault detection, and provides reliable guarantee for stable operation of a transformer substation.

Description

Technical Field

[0001] The present application relates to the technical field of intelligent detection of substations, and more specifically, to a substation fault detection method and related equipment. Background Art

[0002] As power systems continue to expand and become increasingly complex, the requirements for accurate, timely, and intelligent substation fault detection are becoming increasingly stringent. Traditional substation fault detection relies primarily on manual inspections and scheduled maintenance. This approach is not only inefficient but also prone to human error, making it difficult to meet the high standards of reliability and stability required by modern power systems.

[0003] While some existing data-driven approaches have been applied to substation fault detection, they have exposed numerous practical challenges. For example, these approaches lack adaptability to the complex operational scenarios of substations, failing to accurately capture dynamic changes in system status and effectively model them. Furthermore, due to a lack of efficient fusion and full utilization of multi-source data, a wealth of valuable information remains untapped, significantly reducing the accuracy of fault detection and location, making it difficult to meet the growing operational and maintenance demands of power systems.

[0004] Therefore, a new substation fault detection method is urgently needed to avoid the defects of existing technologies and realize intelligent detection of substation faults. Summary of the Invention

[0005] This application provides a substation fault detection method and related equipment. By acquiring real-time data from the secondary system, using a scenario prediction model composed of a deep learning model and a dynamic Bayesian network to extract features and predict scenario probabilities, and combining a digital twin system to determine maintenance needs, it solves the problems of low efficiency, poor adaptability, and insufficient data fusion of traditional methods, and achieves accurate, timely, and intelligent fault detection.

[0006] A substation fault detection method, comprising:

[0007] Acquire real-time data of the substation secondary system and input the real-time data into a scenario prediction model, wherein the scenario prediction model includes a deep learning model and a dynamic Bayesian network;

[0008] Extracting features from the real-time data using the deep learning model to obtain state change features of the secondary system, inputting the state change features into the dynamic Bayesian network to obtain at least one prediction scenario information of the substation secondary system at a target time and the probability corresponding to each prediction scenario information;

[0009] Inputting the real-time data and the predicted scenario information into a pre-established digital twin system to obtain target maintenance requirements under each predicted scenario;

[0010] The substation fault detection result is determined based on the probability corresponding to each of the prediction scenario information and the target maintenance requirements under each of the prediction scenarios.

[0011] Optionally, also include:

[0012] Performing preliminary simulation based on the real-time data and the digital twin system to generate target simulation data;

[0013] Fusing the real-time data with the target simulation data to obtain fusion features;

[0014] The fusion features are input into the deep learning model to obtain the state change features of the secondary system.

[0015] Optionally, the dynamic Bayesian network is composed of state variable nodes set based on each state variable of the substation secondary system, and connection edges set based on the causal relationship between the state variable nodes;

[0016] Inputting the state change characteristics into the dynamic Bayesian network to obtain at least one prediction scenario information of the substation secondary system at a target time and the probability corresponding to each prediction scenario information includes:

[0017] Inputting the state change characteristics into the observation node of the dynamic Bayesian network, and obtaining at least one prediction scenario information of the substation secondary system at the target time through inference calculation, wherein the observation node is a node in the state variable node that directly reflects the system state change;

[0018] According to the probability distribution of each state variable node, the probability corresponding to each predicted scene information is determined.

[0019] Optionally, fusing the real-time data with the target simulation data to obtain fusion features includes:

[0020] Using the electrical parameters in the real-time data as a first mode and the fault simulation data in the target simulation data as a second mode;

[0021] Encode the features of each modal data and adjust the modality weight through the attention mechanism;

[0022] The encoded features are weightedly fused according to the modal weights to generate fused features.

[0023] Optionally, determining a substation fault detection result according to the probability corresponding to each prediction scenario information and the target maintenance requirement under each prediction scenario includes:

[0024] Quantify the urgency, complexity, and maintenance cost of the target maintenance requirements under each of the forecast scenarios, and multiply them by the probability corresponding to each of the forecast scenario information to obtain a risk value;

[0025] The faults having a risk value higher than a threshold are sorted and filtered to determine the fault detection result of the substation.

[0026] A substation fault detection device, comprising:

[0027] A real-time data unit, configured to obtain real-time data of a substation secondary system and input the real-time data into a scenario prediction model, wherein the scenario prediction model includes a deep learning model and a dynamic Bayesian network;

[0028] a scenario prediction unit, configured to extract features from the real-time data using the deep learning model to obtain state change features of the secondary system, input the state change features into the dynamic Bayesian network, and obtain at least one predicted scenario information of the substation secondary system at a target time and a probability corresponding to each predicted scenario information;

[0029] A maintenance requirement unit, configured to input the real-time data and the predicted scenario information into a pre-established digital twin system to obtain target maintenance requirements under each predicted scenario;

[0030] The fault result unit is used to determine the substation fault detection result according to the probability corresponding to each of the prediction scenario information and the target maintenance requirements under each of the prediction scenarios.

[0031] Optionally, a preliminary simulation unit is also included;

[0032] The preliminary simulation unit is used to perform preliminary simulation based on the real-time data and the digital twin system, generate target simulation data, fuse the real-time data with the target simulation data to obtain fusion features, and input the fusion features into the deep learning model to obtain state change features of the secondary system.

[0033] A substation fault detection device, comprising a memory and a processor;

[0034] The memory is used to store programs;

[0035] The processor is configured to execute the program to implement each step of the substation fault detection method as described in any one of the above items.

[0036] A readable storage medium having a computer program stored thereon, characterized in that when the computer program is executed by a processor, each step of the substation fault detection method as described in any one of the above items is implemented.

[0037] A computer program product includes a computer program, wherein the computer program is configured to execute the steps of any of the above-mentioned substation fault detection methods when the computer program is executed by a processor.

[0038] It can be seen from the above technical solutions that the embodiment of the present application provides a substation fault detection method and related equipment, which obtains real-time data of the substation secondary system, inputs the real-time data into a scenario prediction model composed of a deep learning model and a dynamic Bayesian network, uses the deep learning model to extract state change characteristics, obtains predicted scenario information and probability through the dynamic Bayesian network, and combines the digital twin system to determine the target maintenance requirements. Finally, according to the probability corresponding to each predicted scenario information and the target maintenance requirements under each predicted scenario, the substation fault detection result is determined. The deep learning model in this application has a powerful multi-source data processing capability, which can fully mine the potential information in the real-time data and realize the efficient extraction of fault characteristics in complex scenarios, making up for the defect of insufficient information mining in the existing data-driven method and improving the accuracy of fault detection and positioning. The dynamic Bayesian network can perform probabilistic modeling of the dynamic changes of the system, accurately characterize the state evolution during the operation of the substation, and effectively solve the problem that the traditional data-driven method is difficult to adapt to complex scenarios and cannot accurately model the dynamic changes of the system. In addition, combined with the digital twin system, deep integration and comprehensive analysis of multi-source data are achieved, forming an integrated intelligent processing flow from data collection, fault prediction to maintenance decision-making, which greatly improves the accuracy, timeliness and intelligence level of fault detection, and provides strong guarantees for the reliable and stable operation of substations. BRIEF DESCRIPTION OF THE DRAWINGS

[0039] In order to more clearly illustrate the embodiments of the present application or the technical solutions in the prior art, the following briefly introduces the drawings required for use in the embodiments or the description of the prior art. Obviously, the drawings described below are merely embodiments of the present application. For ordinary technicians in this field, other drawings can be obtained based on the provided drawings without any creative work.

[0040] Figure 1 This is a flow chart of a substation fault detection method disclosed in an embodiment of the present application;

[0041] Figure 2 A schematic diagram of a substation fault detection device disclosed in an embodiment of the present application;

[0042] Figure 3This is a hardware structure block diagram of a substation fault detection device disclosed in an embodiment of the present application. DETAILED DESCRIPTION

[0043] The following will be combined with the drawings in the embodiments of this application to clearly and completely describe the technical solutions in the embodiments of this application. Obviously, the embodiments described are only part of the embodiments of this application, not all of the embodiments. Based on the embodiments in this application, all other embodiments obtained by ordinary technicians in this field without making creative efforts are within the scope of protection of this application.

[0044] The present application can be used in a variety of general or special computing device environments or configurations, such as personal computers, server computers, handheld or portable devices, tablet devices, multi-processor devices, and distributed computing environments including any of the above devices or devices.

[0045] Next, we will introduce the application scheme. This application proposes the following technical scheme, please see below for details.

[0046] Figure 1 This is a flow chart of a substation fault detection method disclosed in an embodiment of the present application.

[0047] like Figure 1 As shown, the method may include:

[0048] Step S1: Acquire real-time data of the substation secondary system and input the real-time data into a scenario prediction model, wherein the scenario prediction model includes a deep learning model and a dynamic Bayesian network.

[0049] Specifically, during the operation of the substation, real-time data of the secondary system is collected in all directions through various sensors, monitoring equipment and communication networks, covering electrical parameters, equipment status information, communication status data and environmental parameters, etc., which are not limited here. Electrical parameters can be voltage, current, power, frequency, etc., which can intuitively reflect the electrical operating status of the system; equipment status information can be the action signal of the protection device, the working status of the measurement and control device, the open and close position of the switch, etc., which are used to determine whether the equipment is working normally; communication status data can be the signal strength, data transmission rate, bit error rate, etc. of the communication link to ensure smooth information exchange between systems; environmental parameters can be temperature, humidity, air pressure, etc. After the collection is completed, the above real-time data is transmitted to the scenario prediction model, which is composed of a deep learning model and a dynamic Bayesian network, providing a data basis for subsequent feature extraction and scenario prediction.

[0050] Step S2: extract features from the real-time data through the deep learning model to obtain state change features of the secondary system, input the state change features into the dynamic Bayesian network, and obtain at least one prediction scenario information of the substation secondary system at the target time and the probability corresponding to each prediction scenario information.

[0051] Specifically, real-time data is fed into a deep learning model. Leveraging its multi-layer neural network structure and nonlinear activation functions, the model automatically extracts features and learns representations of the data. This model captures complex patterns and underlying relationships from the raw data, generating feature vectors that effectively reflect changes in the secondary system state (e.g., equipment operating trends, parameter correlations, etc.). These state change features are then fed into a dynamic Bayesian network, constructed using the substation secondary system's state variables (e.g., electrical parameter nodes, equipment status nodes, environmental parameter nodes, etc.) as nodes and causal relationships between nodes as edges. Through inference calculations such as belief propagation algorithms, the model combines observed data to update the node probability distribution, thereby inferring at least one predicted scenario at the target time (e.g., equipment failure, external interference, system operating state change, etc.) and determining the corresponding probability of occurrence for each scenario.

[0052] Specifically, the possible scenarios of the substation secondary system are as follows:

[0053] 1. Equipment failure

[0054] (1) Failure of protective device

[0055] Scenarios of malfunction caused by errors in constant value calculation: New power sources are connected to the power system without recalculating the constant values; after the grid topology changes due to line maintenance, the constant value calculation model is not updated, and normal parameter fluctuations trigger malfunctions.

[0056] Sampling circuit failure causes protection failure scenarios: CT secondary side wiring screws are loose; PT fuses are blown; sampling resistors are damaged, causing gas volume data deviation or loss, and protection failure occurs in the event of a fault.

[0057] Abnormal operation scenarios caused by damage to logic components: overheating and burning of logic operation chips; adhesion of relay contacts; long-term electromagnetic interference causing the solder joints of logic circuits to loosen, resulting in chaotic operation logic.

[0058] (2) Failure of measurement and control device

[0059] Scenarios where data deviation is caused by analog-to-digital conversion module failure include: aging of the analog-to-digital conversion chip and decreased accuracy; capacitor leakage in the conversion circuit and zero point drift; and failure of the reference voltage source, causing the collected data to deviate significantly from the actual value.

[0060] Communication interface failures can cause data transmission interruptions: the communication interface chip is damaged; the interface pins are broken; or the baud rate and parity bit settings in the communication protocol configuration are incorrect, preventing data transmission.

[0061] Clock synchronization anomalies can cause data timestamp errors: The GPS antenna is blocked; the clock module crystal oscillator is aging, resulting in frequency deviation; the NTP server is faulty, preventing time synchronization and causing data timestamp errors.

[0062] (3) Communication equipment failure

[0063] Switch port failures can cause local communication interruptions: physical damage to the port; a switch software failure that causes the port to become disabled; or a malfunctioning power supply module on the port that causes communication interruption between connected devices.

[0064] Scenarios where fiber optic link breakage leads to long-distance communication interruption include: fiber optic digging during construction; fiber optic damage from rat bites and insect infestation in natural environments; and fiber optic fatigue fracture caused by long-term bending, which hinders long-distance communication.

[0065] Scenarios where network protocol incompatibility causes packet loss: Mixing new and old devices with inconsistent protocol versions; differences in customized protocols between different manufacturers; and failure to adapt the protocol after a network upgrade, leading to packet loss.

[0066] 2. External interference

[0067] (1) Electromagnetic interference

[0068] Electromagnetic induction interference scenario of adjacent high-voltage transmission lines: The transmission line and the secondary system line are too close in parallel; the transmission line load suddenly changes, the magnetic field changes drastically, and interferes with the secondary system equipment.

[0069] Harmonic interference scenarios for electrical equipment in substations: Transformer core saturation generates harmonics; power electronic equipment such as SVGs and inverters operate, injecting harmonics into the grid and interfering with secondary equipment.

[0070] Wireless communication equipment signal interference scenarios: The mobile base station antenna is facing the substation; the wireless communication equipment transmit power is too high; the frequency band is close to the secondary system communication frequency band, causing signal interference.

[0071] (2) Lightning interference

[0072] Scenario of direct lightning overvoltage damage to substation equipment: Lightning strikes the substation lightning rod, and the powerful lightning current strikes directly on electrical equipment, causing the instantaneous high voltage to break through the insulation and burn out components.

[0073] Scenarios where induced lightning overvoltage affects communications and equipment operation: Lightning strikes nearby overhead lines, inducing overvoltage along the lines. Lightning strikes the ground, raising the ground potential and coupling it to the secondary system, affecting equipment.

[0074] Lightning electromagnetic pulses interfere with secondary system electronic components: The strong electromagnetic pulses generated by lightning strikes radiate through space and are conducted through power lines and signal lines, interfering with the operation of electronic components.

[0075] 3. System operation status

[0076] (1) Normal operation

[0077] Stable operation scenarios when the load is stable: during the off-peak period of industrial electricity consumption and the stable period of residential electricity consumption at night, system parameters are stable and equipment works well together.

[0078] Adaptive adjustment scenarios when load changes gradually: seasonal changes, air conditioning load changes; peak power consumption in the morning and evening on weekdays, the system automatically adjusts parameters to adapt.

[0079] (2) System upgrade or maintenance

[0080] Compatibility issues after software updates: The new software does not match the old hardware driver; changes in interfaces between software modules are not adapted, causing the device to freeze or malfunction.

[0081] Parameter configuration errors after hardware replacement: When a protection plug-in is replaced but the rated current ratio is not modified; when the measurement and control main board is replaced but the voltage transformer ratio is not calibrated, the device may malfunction.

[0082] Improper temporary wiring during system maintenance: The following scenarios may occur: insulation damage and short circuits in temporary test cables; reverse polarity of power cables; loose terminals with poor contact, which may introduce interference or cause equipment failure.

[0083] 4. Multiple fault complex type

[0084] Joint failure scenario of protection device and measurement and control device: The protection device malfunctions due to incorrect setting. At the same time, the analog-to-digital conversion failure of the measurement and control device causes data deviation, making fault diagnosis and handling more difficult.

[0085] Cascading failure scenarios of communication equipment and protection devices: Fiber optic link breakage causes communication interruption. The protection device cannot receive remote tripping signals and cannot promptly eliminate the fault, causing a large-scale power outage.

[0086] External interference triggers multiple equipment failure scenarios: Lightning strikes cause strong electromagnetic pulses in the substation, interfering with the logic elements of the protection device, damaging the interface chips of the communication equipment, and causing data disorder in the measurement and control device.

[0087] System upgrade and device failure concurrent scenario: During the software upgrade process, hardware failures occur due to aging. For example, a protection plug-in hardware failure is incompatible with the software upgrade, causing the system to crash.

[0088] Multiple equipment aging and compound failure scenarios: protection device logic elements, measurement and control device analog-to-digital conversion chips, and communication equipment ports age due to long-term operation and fail at the same time, seriously affecting system operation.

[0089] It is understandable that possible scenarios of the substation secondary system should include but not be limited to the above scenarios.

[0090] Step S3: input the real-time data and the predicted scenario information into a pre-established digital twin system to obtain the target maintenance requirements under each predicted scenario.

[0091] Specifically, the pre-established digital twin system is a precise virtual mapping of the substation secondary system, encompassing its physical structure, electrical connections, operating logic, and historical data models. Various data collected in real time from the substation secondary system, such as electrical parameters like voltage, current, and power, are rapidly transmitted to the digital twin system via a communication network. This real-time data reflects the system's current operating status and serves as the foundation for the digital twin system's real-time simulation and analysis. System synchronization based on real-time data: After receiving real-time data, the digital twin system first performs system synchronization. By comparing and calibrating the real-time data with its internal virtual model, it ensures that the operating status of the virtual model is completely consistent with the actual substation secondary system. Based on system synchronization, the digital twin system combines the input scenario information for simulation and analysis. Different simulation strategies are employed for different scenario information.

[0092] When inputting information about a potential misoperation scenario for a protective device, the digital twin system simulates the impact of the misoperation on related equipment based on the protective device's operating logic and its connections to other devices. For example, if a protective device malfunctions and disconnects a line, the system simulates the voltage and current changes in connected equipment (such as transformers and loads) after the outage, as well as the impact on the overall system power distribution. This simulation analyzes potential issues such as overloads and voltage anomalies in other equipment.

[0093] For sudden load changes, the digital twin system simulates the dynamic changes in electrical parameters in the substation's secondary system when the load suddenly increases or decreases, based on historical data and load change models. For example, it simulates how voltage fluctuates with load changes, how current is redistributed among different devices and lines, and how protection devices and measurement and control equipment respond to these load changes. This simulation assesses the system's stability and reliability under sudden load changes.

[0094] After simulating and analyzing equipment failure scenarios, the digital twin system determines the maintenance requirements for the faulty equipment based on the simulation results. For any faulty equipment, the system analyzes the cause, type, and severity of the failure. For example, if a simulation reveals that a communications device has experienced a communication interruption due to aging hardware, the system will determine the hardware component that needs to be replaced based on the device model and fault condition, and recommend a replacement timeframe. The system also considers the potential impact of hardware component replacement on other parts of the system, such as the duration of communication interruption and data transmission recovery time, to develop a comprehensive maintenance plan, including the tools required for maintenance and the required technician qualifications.

[0095] For scenarios with operational status changes, the digital twin system determines operational status adjustment and maintenance requirements based on simulation results. For example, if system voltage fluctuations exceed normal ranges during a sudden load change simulation, the system analyzes the causes (such as insufficient reactive power compensation or changes in line impedance) and recommends appropriate maintenance measures. These may include adjusting reactive power compensation device parameters, conducting line inspections and maintenance to reduce line impedance, and so on. The system also provides a sequence and timeline for implementing maintenance measures to ensure that the system is restored to a stable state as quickly as possible without disrupting normal operation.

[0096] In addition to determining maintenance requirements for equipment and systems that have experienced failures or changes in operating status, the digital twin system also determines preventive maintenance needs based on simulation analysis results. Through in-depth analysis of historical data and simulation results, the system can predict the probability and risk of failure of certain equipment in the future. For example, by analyzing factors such as equipment operating time, temperature changes, and electrical parameter fluctuations, it predicts that a certain protective device has a high probability of failure due to overheating within the next three months. Based on this, the system will develop a preventive maintenance plan, such as recommending inspection and maintenance of the protective device's cooling system within one month and replacing aging cooling components in advance to reduce the risk of equipment failure and ensure the long-term stable operation of the substation's secondary system.

[0097] Step S4: Determine the substation fault detection result according to the probability corresponding to each prediction scenario information and the target maintenance requirement under each prediction scenario.

[0098] Specifically, for each maintenance-required fault in each predicted scenario, the urgency, complexity, and maintenance cost of the target maintenance need are analyzed and quantified. Urgency can be measured based on how quickly the fault impacts the substation's normal operations. For example, if a fault could cause a widespread power outage in a short period of time, the urgency is high and can be quantified as a higher value, such as 8-10. If the fault's impact on system operation is more gradual, the urgency is low and can be quantified as a lower value, such as 1-3. Complexity can be assessed based on the technical difficulty, manpower, and material resources required for troubleshooting and repair. Complex faults requiring specialized technical teams and specialized equipment are considered high complexity and can be quantified as 7-9. Simple faults involving equipment parameter adjustments are considered low complexity and can be quantified as 1-3. Maintenance costs include the cost of parts required for repair, labor costs, and the cost of power loss caused by the power outage. By estimating these costs, the maintenance costs are converted into numerical values. For example, if repairing a fault requires replacing expensive equipment parts and the power outage is long, resulting in large power loss costs, the maintenance cost can be quantified as a higher value, such as 7-10 points. If it only requires a simple replacement of a low-cost component with almost no impact on power supply, the maintenance cost can be quantified as 1-3 points.

[0099] Each maintenance-required fault has a corresponding probability of occurrence within the target time period, which is calculated using a dynamic Bayesian network. The quantified target maintenance requirement value is multiplied by the fault occurrence probability to obtain the risk value of each maintenance-required fault within the target time period. For example, the urgency of a protective device malfunction is quantified as 8 points, the complexity is quantified as 6 points, and the maintenance cost is quantified as 7 points. Taking these three factors into consideration and taking the average value, the target maintenance requirement is quantified as (8+6+7) / 3=7 points. The probability of occurrence of this fault is calculated as 0.2 using a dynamic Bayesian network. Therefore, the risk value of the protection device malfunction within the target time period is 7×0.2=1.4 points. The risk value is calculated in this way for all maintenance-required faults.

[0100] The calculated risk values for various faults to be maintained are ranked from high to low. Faults with high risk values indicate a high probability of occurrence within the target period and, if they occur, will have a serious impact on substation operations, requiring priority attention and resolution. Faults with low risk values have a relatively low probability of occurrence and a relatively small impact on system operations. Combined with a pre-set risk threshold, faults above this threshold are identified as maintenance targets, completing the intelligent substation fault detection process.

[0101] Therefore, the process of determining the substation fault detection result according to the probability corresponding to each prediction scenario information and the target maintenance requirement under each prediction scenario may specifically include:

[0102] ① Quantify the urgency, complexity, and maintenance cost of the target maintenance requirements under each forecast scenario, and multiply them by the probability corresponding to each forecast scenario information to obtain a risk value;

[0103] ② Sort by the risk value and filter faults above the threshold to determine them as substation fault detection results.

[0104] It can be seen from the above technical solutions that the embodiment of the present application provides a substation fault detection method and related equipment, which obtains real-time data of the substation secondary system, inputs the real-time data into a scenario prediction model composed of a deep learning model and a dynamic Bayesian network, uses the deep learning model to extract state change characteristics, obtains predicted scenario information and probability through the dynamic Bayesian network, and combines the digital twin system to determine the target maintenance requirements. Finally, according to the probability corresponding to each predicted scenario information and the target maintenance requirements under each predicted scenario, the substation fault detection result is determined. The deep learning model in this application has a powerful multi-source data processing capability, which can fully mine the potential information in the real-time data and realize the efficient extraction of fault characteristics in complex scenarios, making up for the defect of insufficient information mining in the existing data-driven method and improving the accuracy of fault detection and positioning. The dynamic Bayesian network can perform probabilistic modeling of the dynamic changes of the system, accurately characterize the state evolution during the operation of the substation, and effectively solve the problem that the traditional data-driven method is difficult to adapt to complex scenarios and cannot accurately model the dynamic changes of the system. In addition, combined with the digital twin system, deep integration and comprehensive analysis of multi-source data are achieved, forming an integrated intelligent processing flow from data collection, fault prediction to maintenance decision-making, which greatly improves the accuracy, timeliness and intelligence level of fault detection, and provides strong guarantees for the reliable and stable operation of substations.

[0105] In some embodiments of the present application, step S2 is introduced, in which the real-time data is subjected to feature extraction through the deep learning model to obtain state change characteristics of the secondary system, the state change characteristics are input into the dynamic Bayesian network, and at least one prediction scenario information of the substation secondary system at the target time and the probability corresponding to each prediction scenario information are obtained.

[0106] Deep learning model processing part:

[0107] There are two optional implementations for extracting features from the real-time data using a deep learning model to obtain state change features of the secondary system:

[0108] The first one is to directly extract features from real-time data.

[0109] Real-time data is input into a deep learning model. This real-time data contains various operating parameters and status information of the substation's secondary system, such as electrical parameters. The deep learning model possesses powerful feature learning capabilities. It automatically extracts features and learns representations from the input real-time data through a multi-layer neural network structure. Each neuron in the model processes data using nonlinear activation functions, capturing complex patterns and underlying relationships within the data. In this process, the model gradually learns features at different levels, from low-level features of the raw data, such as instantaneous changes in voltage and current, to higher-level features, such as trends in equipment operating status and correlations between different parameters. Through continuous training and optimization, the deep learning model is able to map the input real-time data into a feature vector that effectively represents changes in the secondary system state. This feature vector contains key information reflecting the current state of the secondary system and changes compared to its historical state, providing a foundation for subsequent input into a dynamic Bayesian network for further analysis and reasoning.

[0110] The second method is to extract features by fusing real-time data with target simulation data to obtain fusion features.

[0111] Specifically include:

[0112] ① Performing preliminary simulation based on the real-time data and the digital twin system to generate target simulation data;

[0113] ② Fusing the real-time data with the target simulation data to obtain fusion features;

[0114] ③ Input the fusion features into the deep learning model to obtain the state change features of the secondary system.

[0115] A digital twin system is a precise virtual representation of a substation's secondary system. Its construction is based on a deep understanding of the substation's physical structure, electrical connections, operating logic, and historical data. After acquiring real-time data, it is input into the digital twin system. Based on pre-defined operating rules, physical models, and various algorithms, the system performs a preliminary simulation of the future trends of the substation's secondary system under its current real-time state. For example, based on real-time collected electrical parameters, combined with the equipment's operating characteristics and load variations, the system simulates the possible changes in these parameters over a period of time. Based on the equipment's current state information, the system simulates the possibility of equipment failures or state changes at a target time. These simulation results, known as target simulation data, cover the possible evolution scenarios of the substation's secondary system across different dimensions, providing rich reference information for subsequent analysis. The fusion process can be accomplished through either concatenated fusion or weighted fusion. The fused features are then input into the deep learning model. After processing, the real-time data and target simulation data are converted into state change features of the secondary system. These features represent a highly abstract and condensed representation of the original data, effectively capturing the dynamic changes in the system's operating state and providing critical data support for subsequent scenario analysis.

[0116] Furthermore, the real-time data is fused with the target simulation data to obtain fusion features, including:

[0117] Using the electrical parameters in the real-time data as a first mode and the fault simulation data in the target simulation data as a second mode;

[0118] Encode the features of each modal data and adjust the modality weight through the attention mechanism;

[0119] The encoded features are weightedly fused according to the modal weights to generate fused features.

[0120] For the electrical parameters of the first mode, considering its time series characteristics, a recurrent neural network can be used. For the fault simulation data of the second mode, due to its diverse data forms and certain structure, a convolutional neural network can be used for encoding.

[0121] The core goal of the attention mechanism is to allow the model to automatically learn the importance of the two modal data to the final fusion features under different circumstances. To this end, an attention module is constructed, which takes the feature encoding of the first modal electrical parameters and the feature encoding of the second modal fault simulation data as input. The attention module is usually composed of a multi-layer perceptron. It first maps the feature encoding of the two modalities to a new feature space through linear transformation, and then calculates the similarity score between the two modal features. For example, the dot product operation can be used to calculate the similarity between the feature vectors to obtain a similarity matrix. Next, the similarity matrix is normalized and converted into an attention weight matrix. Each element in this matrix represents the degree of association and importance weight between a feature of the first modality and a feature of the second modality in the current situation.

[0122] Based on the weight matrix calculated by the attention mechanism, the feature encodings of the two modalities are weightedly fused. For each corresponding feature vector in the feature encoding of the first modal electrical parameters and the feature encoding of the second modal fault simulation data, the corresponding attention weights are multiplied. These weighted feature vectors are then added together to produce a fused feature vector. For example, assuming a feature vector of the first modality is [0.2, 0.3, 0.4] and a feature vector of the corresponding position in the second modality is [0.1, 0.5, 0.2], and the weights calculated by the attention mechanism are 0.6 and 0.4, respectively, the fused feature vector is [0.16, 0.38, 0.32]. By performing this weighted fusion operation on the feature vectors at all positions, a complete fused feature is ultimately obtained. These fused features combine the electrical parameter information in the real-time data with the fault simulation information in the target simulation data, providing a more comprehensive and accurate reflection of the operating status and potential risks of the substation secondary system, providing more valuable input for subsequent deep learning models.

[0123] Dynamic Bayesian network processing part:

[0124] The dynamic Bayesian network is composed of state variable nodes set based on each state variable of the substation secondary system, and connection edges set based on the causal relationship between the state variable nodes.

[0125] The dynamic Bayesian network comprehensively covers the state variable nodes set for each state variable in the substation secondary system. These nodes are specifically set for various electrical parameters in the substation secondary system, such as voltage nodes, which can be subdivided into bus voltage nodes and line voltage nodes to accurately reflect the voltage status at different locations. Current nodes include current nodes for each transmission line and equipment branch current nodes to monitor current changes. Power nodes further distinguish between active power nodes and reactive power nodes to evaluate the system's power distribution. Nodes are set for different devices in the substation secondary system. Protection device action nodes clearly distinguish the status of different types of protection devices (such as differential protection device action nodes and overcurrent protection device action nodes). Measurement and control device operating status nodes specifically monitor the operating, fault, and alarm status of each measurement and control device. Communication equipment status nodes cover switch operating status nodes, fiber optic communication link status nodes, and wireless communication module status nodes, comprehensively monitoring the operation of communication equipment. Taking into account the impact of the environment on the secondary system of the substation, temperature nodes are set up and subdivided according to different areas of the substation (such as equipment room temperature nodes and outdoor equipment temperature nodes); humidity nodes also distinguish between indoor and outdoor humidity nodes; air pressure nodes, etc., are used to monitor changes in environmental factors.

[0126] After setting up the state variable nodes, the connections based on physical principles and the connections mined from historical data are defined as network edges, completing the construction of the Bayesian network. The state change features output by the deep learning model are used as observation data and accurately input into the pre-set observation nodes of the dynamic Bayesian network. These observation nodes typically correspond to variables that can directly reflect changes in system states, such as certain key electrical parameter nodes and equipment status nodes that are susceptible to faults. For example, if the state change features extracted by the deep learning model include information about an abnormal increase in current on a certain line, then this feature will be input into the corresponding line current node as observation data.

[0127] In an embodiment of the present invention, a belief propagation algorithm is selected as the inference algorithm. After multiple iterative calculations of the inference algorithm, the dynamic Bayesian network can infer at least one scenario information of the substation secondary system at the target time based on the causal relationship between the nodes and the updated probability distribution. After the inference calculation is completed, each node in the dynamic Bayesian network has a corresponding probability distribution. For equipment failure scenarios, taking the protection device malfunction scenario as an example, by analyzing the probability distribution of the protection device action node, combined with the probability distribution of the related electrical parameter nodes and environmental parameter nodes, the probability of the protection device malfunction scenario is comprehensively calculated. For example, if the probability of the protection device action node being in the action state under the current probability distribution is 0.3, the probability of the current of the current node closely related to it exceeding the protection set value is 0.8, and taking into account environmental factors, the final probability of the protection device malfunction scenario is calculated through the Bayesian formula and the conditional probability relationship in the network structure.

[0128] On this basis, in step S2, the state change characteristics are input into the dynamic Bayesian network to obtain at least one prediction scenario information of the substation secondary system at the target time and the probability corresponding to each prediction scenario information, including:

[0129] ① Inputting the state change characteristics into the observation node of the dynamic Bayesian network, and obtaining at least one prediction scenario information of the substation secondary system at the target time through inference calculation. The observation node is the node in the state variable node that directly reflects the system state change;

[0130] ② According to the probability distribution of each state variable node, determine the probability corresponding to each predicted scene information.

[0131] The state change features extracted by the deep learning model are used as observation data and input into the observation nodes of the dynamic Bayesian network. These observation nodes are key nodes within the state variable nodes that directly reflect changes in the system state, such as voltage over-limit nodes that indicate electrical parameter anomalies and protective device action nodes that indicate sudden changes in equipment status. Using a Bayesian network's probabilistic reasoning algorithm (such as the belief propagation algorithm), the system state evolution is deduced based on the causal relationships (directed edges) between nodes and the conditional probability distribution, identifying at least one possible scenario at the target moment, such as equipment failure, parameter over-limit, or external interference. After scenario reasoning is complete, the probability of occurrence of each predicted scenario is calculated based on the joint probability distribution of each state variable node in the dynamic Bayesian network and the node combinations involved in the predicted scenario. This probability, obtained by updating the conditional dependencies between nodes and the observed data, reflects the likelihood of the scenario occurring and provides a quantitative basis for subsequent risk assessment.

[0132] A substation fault detection device provided in an embodiment of the present application is described below. The substation fault detection device described below and the substation fault detection method described above can refer to each other.

[0133] See also Figure 2 , Figure 2 This is a schematic diagram of a substation fault detection device disclosed in an embodiment of the present application.

[0134] like Figure 2 As shown, the substation fault detection device may include:

[0135] A real-time data unit 110 is configured to obtain real-time data of the substation secondary system and input the real-time data into a scenario prediction model, wherein the scenario prediction model includes a deep learning model and a dynamic Bayesian network;

[0136] A scenario prediction unit 120 is configured to extract features from the real-time data using the deep learning model to obtain state change features of the secondary system, input the state change features into the dynamic Bayesian network, and obtain at least one predicted scenario information of the substation secondary system at a target time and a probability corresponding to each predicted scenario information;

[0137] The maintenance requirement unit 130 is configured to input the real-time data and the predicted scenario information into a pre-established digital twin system to obtain target maintenance requirements under each predicted scenario;

[0138] The fault result unit 140 is configured to determine a substation fault detection result based on the probability corresponding to each prediction scenario information and the target maintenance requirement under each prediction scenario.

[0139] It can be seen from the above technical solutions that the embodiment of the present application provides a substation fault detection method and related equipment, which obtains real-time data of the substation secondary system, inputs the real-time data into a scenario prediction model composed of a deep learning model and a dynamic Bayesian network, uses the deep learning model to extract state change characteristics, obtains predicted scenario information and probability through the dynamic Bayesian network, and combines the digital twin system to determine the target maintenance requirements. Finally, according to the probability corresponding to each predicted scenario information and the target maintenance requirements under each predicted scenario, the substation fault detection result is determined. The deep learning model in this application has a powerful multi-source data processing capability, which can fully mine the potential information in the real-time data and realize the efficient extraction of fault characteristics in complex scenarios, making up for the defect of insufficient information mining in the existing data-driven method and improving the accuracy of fault detection and positioning. The dynamic Bayesian network can perform probabilistic modeling of the dynamic changes of the system, accurately characterize the state evolution during the operation of the substation, and effectively solve the problem that the traditional data-driven method is difficult to adapt to complex scenarios and cannot accurately model the dynamic changes of the system. In addition, combined with the digital twin system, deep integration and comprehensive analysis of multi-source data are achieved, forming an integrated intelligent processing flow from data collection, fault prediction to maintenance decision-making, which greatly improves the accuracy, timeliness and intelligence level of fault detection, and provides strong guarantees for the reliable and stable operation of substations.

[0140] Optionally, a preliminary simulation unit is also included;

[0141] The preliminary simulation unit is used to perform preliminary simulation based on the real-time data and the digital twin system, generate target simulation data, fuse the real-time data with the target simulation data to obtain fusion features, and input the fusion features into the deep learning model to obtain state change features of the secondary system.

[0142] Optionally, the dynamic Bayesian network is composed of state variable nodes set based on each state variable of the substation secondary system, and connection edges set based on the causal relationship between the state variable nodes;

[0143] Inputting the state change characteristics into the dynamic Bayesian network to obtain at least one prediction scenario information of the substation secondary system at a target time and the probability corresponding to each prediction scenario information includes:

[0144] Inputting the state change characteristics into the observation node of the dynamic Bayesian network, and obtaining at least one prediction scenario information of the substation secondary system at the target time through inference calculation, wherein the observation node is a node in the state variable node that directly reflects the system state change;

[0145] According to the probability distribution of each state variable node, the probability corresponding to each predicted scene information is determined.

[0146] Optionally, fusing the real-time data with the target simulation data to obtain fusion features includes:

[0147] Using the electrical parameters in the real-time data as a first mode and the fault simulation data in the target simulation data as a second mode;

[0148] Encode the features of each modal data and adjust the modality weight through the attention mechanism;

[0149] The encoded features are weightedly fused according to the modal weights to generate fused features.

[0150] Optionally, determining a substation fault detection result according to the probability corresponding to each prediction scenario information and the target maintenance requirement under each prediction scenario includes:

[0151] Quantify the urgency, complexity, and maintenance cost of the target maintenance requirements under each of the forecast scenarios, and multiply them by the probability corresponding to each of the forecast scenario information to obtain a risk value;

[0152] The faults having a risk value higher than a threshold are sorted and filtered to determine the fault detection result of the substation.

[0153] The substation fault detection device provided in the embodiment of the present application can be applied to substation fault detection equipment. Figure 3 The hardware structure diagram of the substation fault detection equipment is shown. Figure 3 ,The hardware structure of the substation fault detection device may include: at least one processor 1, at least one communication interface 2, at least one memory 3 and at least one communication bus 4;

[0154] In the embodiment of the present application, the number of the processor 1, the communication interface 2, the memory 3, and the communication bus 4 is at least one, and the processor 1, the communication interface 2, and the memory 3 communicate with each other through the communication bus 4;

[0155] The processor 1 may be a central processing unit (CPU), or an application-specific integrated circuit (ASIC), or one or more integrated circuits configured to implement the embodiments of the present invention;

[0156] The memory 3 may include a high-speed RAM memory, and may also include a non-volatile memory (non-volatile memory), such as at least one disk memory;

[0157] The memory stores a program, and the processor can call the program stored in the memory, wherein the program is used to:

[0158] Acquire real-time data of the substation secondary system and input the real-time data into a scenario prediction model, wherein the scenario prediction model includes a deep learning model and a dynamic Bayesian network;

[0159] Extracting features from the real-time data using the deep learning model to obtain state change features of the secondary system, inputting the state change features into the dynamic Bayesian network to obtain at least one prediction scenario information of the substation secondary system at a target time and the probability corresponding to each prediction scenario information;

[0160] Inputting the real-time data and the predicted scenario information into a pre-established digital twin system to obtain target maintenance requirements under each predicted scenario;

[0161] The substation fault detection result is determined based on the probability corresponding to each of the prediction scenario information and the target maintenance requirements under each of the prediction scenarios.

[0162] Optionally, the refined functions and extended functions of the program may refer to the above description.

[0163] The present application also provides a readable storage medium, which may store a program suitable for execution by a processor, wherein the program is used to:

[0164] Acquire real-time data of the substation secondary system and input the real-time data into a scenario prediction model, wherein the scenario prediction model includes a deep learning model and a dynamic Bayesian network;

[0165] Extracting features from the real-time data using the deep learning model to obtain state change features of the secondary system, inputting the state change features into the dynamic Bayesian network to obtain at least one prediction scenario information of the substation secondary system at a target time and the probability corresponding to each prediction scenario information;

[0166] Inputting the real-time data and the predicted scenario information into a pre-established digital twin system to obtain target maintenance requirements under each predicted scenario;

[0167] The substation fault detection result is determined based on the probability corresponding to each of the prediction scenario information and the target maintenance requirements under each of the prediction scenarios.

[0168] Optionally, the refined functions and extended functions of the program may refer to the above description.

[0169] The present application also provides a computer program product, including a computer program. When the computer program is executed by a processor, the execution method is as follows:

[0170] Acquire real-time data of the substation secondary system and input the real-time data into a scenario prediction model, wherein the scenario prediction model includes a deep learning model and a dynamic Bayesian network;

[0171] Extracting features from the real-time data using the deep learning model to obtain state change features of the secondary system, inputting the state change features into the dynamic Bayesian network to obtain at least one prediction scenario information of the substation secondary system at a target time and the probability corresponding to each prediction scenario information;

[0172] Inputting the real-time data and the predicted scenario information into a pre-established digital twin system to obtain target maintenance requirements under each predicted scenario;

[0173] The substation fault detection result is determined based on the probability corresponding to each of the prediction scenario information and the target maintenance requirements under each of the prediction scenarios.

[0174] Optionally, the refined functions and extended functions of the program may refer to the above description.

[0175] Finally, it should be noted that, in this document, relational terms such as first and second, etc., are used only to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any actual relationship or order between these entities or operations. Moreover, the terms "comprises," "comprising," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or device comprising a series of elements includes not only those elements, but also other elements not explicitly listed, or elements inherent to such process, method, article, or device. In the absence of further limitations, an element defined by the phrase "comprising a ..." does not exclude the presence of additional identical elements in the process, method, article, or device comprising the element.

[0176] The various embodiments in this specification are described in a progressive manner, and each embodiment focuses on the differences from other embodiments. The same or similar parts between the various embodiments can be referenced to each other.

[0177] The above description of the disclosed embodiments is intended to enable one skilled in the art to implement or use the present application. Various modifications to these embodiments will be readily apparent to one skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of the present application. Therefore, the present application is not limited to the embodiments shown herein, but is intended to conform to the widest scope consistent with the principles and novel features disclosed herein.

Claims

1. A substation fault detection method, characterized in that: include: Acquire real-time data of the substation secondary system and input the real-time data into a scenario prediction model, wherein the scenario prediction model includes a deep learning model and a dynamic Bayesian network; Extracting features from the real-time data using the deep learning model to obtain state change features of the secondary system, inputting the state change features into the dynamic Bayesian network to obtain at least one prediction scenario information of the substation secondary system at a target time and the probability corresponding to each prediction scenario information; Inputting the real-time data and the predicted scenario information into a pre-established digital twin system to obtain target maintenance requirements under each predicted scenario; The substation fault detection result is determined based on the probability corresponding to each of the prediction scenario information and the target maintenance requirements under each of the prediction scenarios.

2. The method according to claim 1, characterized in that Also includes: Performing preliminary simulation based on the real-time data and the digital twin system to generate target simulation data; Fusing the real-time data with the target simulation data to obtain fusion features; The fusion features are input into the deep learning model to obtain the state change features of the secondary system.

3. The method according to claim 1, characterized in that The dynamic Bayesian network is composed of state variable nodes set based on each state variable of the substation secondary system, and connection edges set based on the causal relationship between the state variable nodes; Inputting the state change characteristics into the dynamic Bayesian network to obtain at least one prediction scenario information of the substation secondary system at a target time and the probability corresponding to each prediction scenario information includes: Inputting the state change characteristics into the observation node of the dynamic Bayesian network, and obtaining at least one prediction scenario information of the substation secondary system at the target time through inference calculation, wherein the observation node is a node in the state variable node that directly reflects the system state change; According to the probability distribution of each state variable node, the probability corresponding to each predicted scene information is determined.

4. The method according to claim 2, characterized in that The real-time data is fused with the target simulation data to obtain fusion features, including: Using the electrical parameters in the real-time data as a first mode and the fault simulation data in the target simulation data as a second mode; Encode the features of each modal data and adjust the modality weight through the attention mechanism; The encoded features are weightedly fused according to the modal weights to generate fused features.

5. The method according to claim 1, characterized in that Determining a substation fault detection result based on the probability corresponding to each prediction scenario information and the target maintenance requirements under each prediction scenario includes: Quantify the urgency, complexity, and maintenance cost of the target maintenance requirements under each of the forecast scenarios, and multiply them by the probability corresponding to each of the forecast scenario information to obtain a risk value; The faults having a risk value higher than a threshold are sorted and filtered to determine the fault detection result of the substation.

6. A substation fault detection device, characterized in that: include: A real-time data unit, configured to obtain real-time data of a substation secondary system and input the real-time data into a scenario prediction model, wherein the scenario prediction model includes a deep learning model and a dynamic Bayesian network; a scenario prediction unit, configured to extract features from the real-time data using the deep learning model to obtain state change features of the secondary system, input the state change features into the dynamic Bayesian network, and obtain at least one predicted scenario information of the substation secondary system at a target time and a probability corresponding to each predicted scenario information; A maintenance requirement unit, configured to input the real-time data and the predicted scenario information into a pre-established digital twin system to obtain target maintenance requirements under each predicted scenario; The fault result unit is used to determine the substation fault detection result according to the probability corresponding to each of the prediction scenario information and the target maintenance requirements under each of the prediction scenarios.

7. The device according to claim 6, characterized in that It also includes a preliminary simulation unit; The preliminary simulation unit is used to perform preliminary simulation based on the real-time data and the digital twin system, generate target simulation data, fuse the real-time data with the target simulation data to obtain fusion features, and input the fusion features into the deep learning model to obtain state change features of the secondary system.

8. A substation fault detection device, characterized in that: including memory and processor; The memory is used to store programs; The processor is configured to execute the program to implement each step of the substation fault detection method according to any one of claims 1 to 5.

9. A readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, each step of the substation fault detection method according to any one of claims 1 to 5 is implemented.

10. A computer program product comprising a computer program, characterized in that When the computer program is executed by a processor, each step of the substation fault detection method according to any one of claims 1 to 5 is executed.

Images