Power substation fault detection method and device

Abstract

The invention discloses a power substation fault detection method and device, and belongs to the technical field of artificial intelligence, and the method comprises the steps: determining a fault detection position of to-be-detected equipment in an inspection task; based on the SLAM technology, the fault detection robot is controlled to arrive at the fault detection position and collect real-time operation data; performing lightweight processing on the real-time operation data by using an edge computing technology to form key operation data, and transmitting the key operation data to a cloud computing platform; inputting key operation data into the fault detection deep neural network, and performing feature extraction on the key operation data to obtain shared feature data; respectively determining the aging degree and fault type probability distribution of the to-be-tested equipment according to the shared feature data; calculating a dynamic fault threshold according to the aging degree and a preset fault threshold; and comparing the fault type probability distribution with the dynamic fault threshold, and determining a fault detection result of the to-be-detected equipment. The problems of low fault data transmission efficiency and low fault detection accuracy can be solved.

Description

Technical Field

[0001] This invention relates to the field of artificial intelligence technology, and in particular to a method and apparatus for detecting faults in power substations. Background Technology

[0002] Ensuring the safe and stable operation of equipment is crucial in the operation and maintenance of power substations. Traditional manual inspection methods are inefficient and pose safety hazards, failing to meet the demands of modern smart grids for efficient and accurate inspections. With the development of robotics technology, intelligent inspection and maintenance robots for power substations have emerged.

[0003] However, using robots to acquire equipment information in complex substation environments results in large amounts of data. Transmitting this data back to the cloud platform for fault detection leads to increased transmission latency and a higher risk of network congestion. Furthermore, current fault detection methods rely on uniform thresholds, which fail to adapt to dynamic changes in equipment performance over time, significantly reducing the accuracy of fault detection. Summary of the Invention

[0004] This invention provides a method and apparatus for fault detection in power substations, which can solve the problems of low efficiency in fault data transmission and low accuracy in fault detection in the prior art.

[0005] To address the aforementioned technical problems, this invention provides a method for detecting faults in power substations, comprising:

[0006] During inspection tasks, determine the location of fault detection in the equipment under test;

[0007] Based on SLAM technology, the fault detection robot is controlled to reach the fault detection location.

[0008] The fault detection robot is controlled to collect real-time operating data of the device under test; wherein, the real-time operating data includes temperature data, vibration data, current data and noise data;

[0009] Using edge computing technology, the fault detection robot is controlled to perform lightweight processing on the real-time operating data to form key operating data, and the key operating data is transmitted to the cloud computing platform.

[0010] On a cloud computing platform, the key operational data is input into a fault detection deep neural network, which extracts features from the key operational data to obtain shared feature data. The aging degree of the device under test is determined based on the shared feature data. The probability distribution of fault types of the device under test is determined based on the shared feature data. A dynamic fault threshold is calculated based on the aging degree and a preset fault threshold. The fault detection result of the device under test is determined by comparing the probability distribution of fault types and the dynamic fault threshold.

[0011] As a preferred embodiment, the step of controlling the fault detection robot to reach the fault detection location based on SLAM technology includes:

[0012] Based on SLAM technology, the current pose of the fault detection robot is determined in real time;

[0013] Based on the fault detection location, determine the target map coordinates in the pre-built power substation map;

[0014] Based on the fault detection location, the target map coordinates, and the pre-built power substation map, a feasible optimal path is planned.

[0015] Based on the feasible optimal path, the fault detection robot is controlled to reach the fault detection location.

[0016] As a preferred embodiment, the use of edge computing technology to control the fault detection robot to perform lightweight processing on the real-time operating data to form key operating data includes:

[0017] The Kalman filter algorithm is used to fuse the temperature data, vibration data, current data and noise data in the real-time operation data to form sensor fusion data;

[0018] By utilizing edge computing technology, the statistical features, time-domain features, and frequency-domain features of the sensor fusion data are extracted to form key operational data.

[0019] As a preferred embodiment, the step of extracting features from the key operational data to obtain shared feature data includes:

[0020] By utilizing the shared feature extraction layer in the fault detection deep neural network, temperature features, vibration features, current features, and noise features are extracted from the key operational data, respectively.

[0021] The temperature characteristics, vibration characteristics, current characteristics, and noise characteristics are fused to form shared feature data.

[0022] As a preferred embodiment, determining the aging degree of the device under test based on the shared feature data includes:

[0023] The nonlinear degradation function in the fault detection deep neural network is used to determine the data acquisition time based on the shared feature data;

[0024] Based on the data acquisition time and the model initial time, determine the time characteristics;

[0025] The aging degree of the device under test is calculated using the following formula:

[0026]

[0027] In the formula, denoted as aging degree; f(·) is a nonlinear function; T is the temperature characteristic; V is the vibration characteristic; I is the current characteristic; N is the noise characteristic; t is the time characteristic; θ is the model parameter; ε is the random error constant.

[0028] As a preferred embodiment, determining the probability distribution of fault types of the device under test based on the shared feature data includes:

[0029] Using a fault diagnostic tool in a fault detection deep neural network, the fault probability of various preset fault types is determined based on the shared feature data and feature fault mapping relationship; wherein, the preset fault types include electrical faults, mechanical faults, environmental impact faults, and system faults.

[0030] The failure probabilities of each preset failure type are summarized to form a failure type probability distribution.

[0031] As a preferred embodiment, the calculation of the dynamic fault threshold based on the aging degree and the preset fault threshold includes:

[0032] Obtain the preset fault threshold for each of the preset fault types;

[0033] Based on the degree of aging, each of the preset fault thresholds is adjusted to a dynamic fault threshold.

[0034] The dynamic fault threshold is calculated using the following formula:

[0035] T dynamic =T base ×(1-α·A)

[0036] In the formula, T dynamic For dynamic fault thresholds; T base α is the preset fault threshold; A is the preset sensitivity coefficient; A is the degree of aging.

[0037] As a preferred embodiment, determining the fault detection result of the device under test by comparing the fault type probability distribution and the dynamic fault threshold includes:

[0038] The fault probabilities of each preset fault type in the fault type probability distribution are compared with the corresponding dynamic fault thresholds.

[0039] Preset fault types with a fault probability greater than the dynamic fault threshold are identified as predicted fault types.

[0040] Summarize all predicted fault types to determine the fault detection result of the device under test.

[0041] As a preferred embodiment, the training process of the fault detection deep neural network is as follows:

[0042] Acquire data samples from various devices within the power substation; wherein, the data samples include temperature data, vibration data, current data, and noise data under various preset fault types and normal operating conditions, as well as aging labels and fault labels;

[0043] Based on the data samples, a shared feature extraction layer in a deep neural network is trained to learn and extract deep features;

[0044] Based on the data samples, a nonlinear degrader in a deep neural network is trained to learn the mapping relationship between deep features and aging labels;

[0045] Based on the data samples, a fault diagnostic tool in a deep neural network is trained to learn the mapping relationship between deep features and fault labels;

[0046] Once the shared feature extraction layer, the nonlinear degradation generator, and the fault diagnosticer have been trained, the deep neural network is identified as a fault detection deep neural network.

[0047] Accordingly, the present invention provides a power substation fault detection device, comprising: a location determination module, a robot control module, a data acquisition module, an edge computing module, and a fault detection module;

[0048] The location determination module is used to determine the fault detection location of the device under test during the inspection task;

[0049] The robot control module is used to control the fault detection robot to reach the fault detection location based on SLAM technology;

[0050] The data acquisition module is used to control the fault detection robot to collect real-time operating data of the device under test; wherein, the real-time operating data includes temperature data, vibration data, current data and noise data;

[0051] The edge computing module is used to control the fault detection robot to perform lightweight processing on the real-time operating data using edge computing technology, form key operating data, and transmit the key operating data to the cloud computing platform.

[0052] The fault detection module is used to input the key operating data into a fault detection deep neural network on a cloud computing platform, so that the fault detection deep neural network can extract features from the key operating data to obtain shared feature data; determine the aging degree of the device under test based on the shared feature data; determine the fault type probability distribution of the device under test based on the shared feature data; calculate a dynamic fault threshold based on the aging degree and a preset fault threshold; and determine the fault detection result of the device under test by comparing the fault type probability distribution and the dynamic fault threshold.

[0053] Compared with the prior art, the embodiments of the present invention have the following beneficial effects:

[0054] This invention provides a fault detection method for power substations. During an inspection, the method determines the fault location of the equipment under test (EUT). Based on SLAM technology, it controls a fault detection robot to reach the fault location. The robot collects real-time operational data from the EUT. Using edge computing technology, the robot performs lightweight processing on the real-time operational data to generate key operational data, which is then transmitted to a cloud computing platform. On the cloud computing platform, the key operational data is input into a fault detection deep neural network, enabling the network to extract features and obtain shared feature data. The aging degree of the EUT is determined based on the shared feature data. The probability distribution of fault types is determined based on the shared feature data. A dynamic fault threshold is calculated based on the aging degree and a preset fault threshold. Finally, the fault detection result is determined by comparing the probability distribution of fault types with the dynamic fault threshold. This invention utilizes a fault detection robot to collect real-time operational data, and then uses edge computing technology to perform lightweight processing on the robot locally. This effectively reduces the amount of data transmitted and the data transmission latency, thereby improving the efficiency of fault data transmission. When performing fault detection on a cloud computing platform, the fault threshold is also dynamically adjusted based on the aging degree of the equipment, so that the fault threshold can adapt to the dynamic changes in equipment performance over time, thereby effectively improving the accuracy of fault detection. Attached Figure Description

[0055] To more clearly illustrate the technical solution of this application, the drawings used in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained from these drawings without creative effort.

[0056] Figure 1 A flowchart illustrating one embodiment of the power substation fault detection method provided by the present invention;

[0057] Figure 2 A schematic diagram illustrating the workflow of a fault detection deep neural network provided by the present invention;

[0058] Figure 3 This is a schematic diagram of one embodiment of the power substation fault detection device provided by the present invention. Detailed Implementation

[0059] To make the objectives, technical solutions, and advantages of this application clearer, the technical solutions of this application will be clearly and completely described below with reference to the accompanying drawings of the embodiments. Obviously, the described embodiments are only some embodiments of this application, not all embodiments. Based on the embodiments of this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0060] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this application pertains; the terminology used herein is for the purpose of describing particular embodiments only and is not intended to limit the application; the terms “comprising” and “having”, and any variations thereof, in the specification, claims, and foregoing description of the drawings are intended to cover non-exclusive inclusion.

[0061] In the description of the embodiments of this application, technical terms such as "first" and "second" are used only to distinguish different objects and should not be construed as indicating or implying relative importance or implicitly specifying the number, specific order, or primary and secondary relationship of the indicated technical features. In the description of the embodiments of this application, "multiple" means two or more, unless otherwise explicitly defined.

[0062] In this document, the term "embodiment" means that a particular feature, structure, or characteristic described in connection with an embodiment may be included in at least one embodiment of this application. The appearance of this phrase in various places throughout the specification does not necessarily refer to the same embodiment, nor is it a separate or alternative embodiment mutually exclusive with other embodiments. It will be explicitly and implicitly understood by those skilled in the art that the embodiments described herein can be combined with other embodiments.

[0063] In the description of the embodiments in this application, the term "and / or" is merely a description of the relationship between related objects, indicating that three relationships can exist. For example, A and / or B can represent: A existing alone, A and B existing simultaneously, and B existing alone. Additionally, the character " / " in this document generally indicates that the preceding and following related objects have an "or" relationship.

[0064] In the description of the embodiments of this application, the term "multiple" refers to two or more (including two), similarly, "multiple sets" refers to two or more (including two sets), and "multiple pieces" refers to two or more (including two pieces).

[0065] In the description of the embodiments of this application, unless otherwise expressly specified and limited, technical terms such as "installation," "connection," "joining," and "fixing" should be interpreted broadly. For example, they can refer to a fixed connection, a detachable connection, or an integral part; they can refer to a mechanical connection or an electrical connection; they can refer to a direct connection or an indirect connection through an intermediate medium; they can refer to the internal communication of two components or the interaction between two components. For those skilled in the art, the specific meaning of the above terms in the embodiments of this application can be understood according to the specific circumstances.

[0066] See Figure 1 To address the problems of low data transmission efficiency and low detection accuracy in existing technologies, an embodiment of the present invention provides a fault detection method for power substations. This method includes steps 101 to 105, each step of which is detailed below:

[0067] Step 101: Determine the fault detection location of the equipment under test during the inspection task.

[0068] In this embodiment of the invention, the safe and stable operation of a power substation highly depends on the quality of inspection tasks. Inspection tasks are a core component of the substation operation and maintenance system. Their function is to promptly identify and address potential problems by monitoring and inspecting the condition of substation equipment, thereby preventing faults from occurring or escalating and ensuring the reliable operation of substation equipment. Inspection tasks include the locations of equipment requiring fault detection. Therefore, upon receiving an inspection task, the fault detection location of the equipment under test can be determined by parsing the task.

[0069] Step 102: Based on SLAM technology, control the fault detection robot to reach the fault detection location.

[0070] In this embodiment of the invention, SLAM (Simultaneous Localization and Mapping) technology is a technique that enables robots or mobile devices to perceive their surroundings in real time through sensors in unknown environments, simultaneously determining their own position and building an environmental map. After obtaining the fault detection location of the device under test, SLAM can be used to navigate the fault detection robot to the fault detection location, allowing the robot to collect relevant real-time operational data at the fault detection location.

[0071] As a preferred embodiment, based on SLAM technology, controlling the fault detection robot to reach the fault detection location includes:

[0072] Based on SLAM technology, the current pose of the fault detection robot is determined in real time;

[0073] Based on the fault detection location, determine the target map coordinates in the pre-built power substation map;

[0074] Based on the fault detection location, the target map coordinates, and the pre-built power substation map, a feasible optimal path is planned.

[0075] Based on the feasible optimal path, the fault detection robot is controlled to reach the fault detection location.

[0076] In this embodiment of the invention, when the fault detection robot performs fault detection on a power substation, sensors such as lidar and vision cameras are first installed on the fault detection robot so that when the fault detection robot enters the power substation for the first time, it can scan and collect data on the surrounding environment based on the installed sensors, and use the SLAM algorithm to construct a two-dimensional or three-dimensional map of the substation. The constructed map is determined as a pre-built power substation map, which can be used in specific inspection tasks.

[0077] In this embodiment of the invention, a lidar is mounted on the fault detection robot, adapted to the robot's overall operating system. This provides the robot with precise environmental perception capabilities in complex substation environments, assisting it in autonomous navigation and accurate equipment detection. This effectively improves the efficiency and quality of the robot's inspection and maintenance, ensuring the stable operation of the power system. The fault detection robot also features a rotating lidar support. This support can be driven by a motor to rotate gears, allowing for adjustment of the rotation and pitch axes. This enables the lidar to scan equipment within the power substation from different directions and angles, significantly increasing the detection range and coverage area. This allows for more comprehensive acquisition of equipment information and timely detection of potential fault points.

[0078] In this embodiment of the invention, the visual camera of the fault detection robot has image enhancement and low light compensation functions, which can clearly acquire equipment image information under low light or complex lighting conditions, thereby improving the accuracy of equipment appearance defect detection.

[0079] In this embodiment of the invention, when the fault detection robot receives a fault detection location, it first determines its current pose and, in conjunction with a previously constructed power substation map, determines the location of the fault detection location on the map, thereby determining the target map coordinates corresponding to the fault detection location. Based on the fault detection location, target map coordinates, and the pre-built power substation map, a collision-free feasible optimal path from the fault detection robot's current pose to the fault detection location can be calculated using a path planning algorithm, thereby controlling the fault detection robot to reach the fault detection location along this feasible optimal path.

[0080] In this embodiment of the invention, the shell of the fault detection robot can be made of high-strength metal materials, such as aluminum alloy or stainless steel, to ensure that the shell has sufficient mechanical strength to withstand external forces such as collisions and compression that may be encountered in the substation. Electromagnetic shielding technology is integrated inside the metal shell, such as using a multi-layer shielding structure, including conductive layers and magnetic layers, to effectively block the interference of high-voltage electric fields and electromagnetic radiation on the internal electronic components of the robot. The internal electronic circuits of the robot are optimized and designed, and measures such as shielded cables and filtering circuits are used to further reduce the impact of electromagnetic interference on the circuits. At the same time, key electronic components are specially treated to improve their anti-electromagnetic interference capabilities, ensuring that the robot can operate stably in a high-voltage and strong electromagnetic field environment.

[0081] Step 103: Control the fault detection robot to collect real-time operating data of the device under test; wherein, the real-time operating data includes temperature data, vibration data, current data and noise data.

[0082] In this embodiment of the invention, the fault detection robot is also equipped with various types of sensors to collect real-time operating data of the device under test. For example, a high-precision infrared thermal imaging sensor is used to detect the surface temperature distribution of the device to generate temperature data, so as to detect potential faults caused by overheating in a timely manner; a vibration sensor is used to monitor the vibration of the device during operation to generate vibration data, and to determine whether the mechanical parts of the device are normal; a current transformer is used to convert the alternating current of the tested conductor into a small current output through electromagnetic induction to generate current data; and an ultrasonic sensor is used to detect abnormal sounds such as partial discharge inside the device to generate noise data.

[0083] Step 104: Using edge computing technology, control the fault detection robot to perform lightweight processing on the real-time operating data to form key operating data, and transmit the key operating data to the cloud computing platform.

[0084] In this embodiment of the invention, edge computing technology allows for preliminary processing and analysis of collected data locally on the robot, generating key operational data. This reduces the amount of data uploaded to the cloud computing platform, effectively minimizing signal loss and remote diagnostic delays, thereby improving fault handling efficiency. Specifically, the data upload delay can be simplified as: D = S / B; where D is the transmission delay, S is the uploaded data size (in bits), and B is the link bandwidth (in bps). Edge computing technology effectively reduces S, thus decreasing data transmission delay.

[0085] As a preferred embodiment, edge computing technology is used to control the fault detection robot to perform lightweight processing on the real-time operating data, forming key operating data, including:

[0086] The Kalman filter algorithm is used to fuse the temperature data, vibration data, current data and noise data in the real-time operation data to form sensor fusion data;

[0087] By utilizing edge computing technology, the statistical features, time-domain features, and frequency-domain features of the sensor fusion data are extracted to form key operational data.

[0088] In this embodiment of the invention, data collected by various sensors are fused to form sensor fusion data, which enables the acquisition of device information from multiple dimensions and improves the accuracy of fault detection. Advanced sensor fusion algorithms are employed to fuse data from different sensors; for example, a Kalman filter algorithm is used to perform weighted fusion of data from multiple sensors.

[0089] The Kalman filter algorithm consists of two steps: prediction and update.

[0090] Prediction phase:

[0091]

[0092] P k|k-1 =AP k-1|k-1 A T +Q

[0093] Update phase:

[0094] K k =P k|k-1 H T HP k|k-1 H T +R) -1

[0095]

[0096] P k|k =(IK kH)P k|k-1

[0097] In the formula, This represents the predicted state at time k. Let A represent the optimal state estimate at time k-1, B represent the state transition matrix, and u represent the control input matrix. k P represents the control quantity at time k; k|k-1 Let P represent the predicted covariance matrix. k-1|k-1 K represents the state covariance at the previous moment; Q represents the process noise covariance; k Let H represent the Kalman gain, H represent the observation matrix, R represent the measurement noise covariance, and z represent the measurement noise covariance. k Let I represent the observed value, and let I represent the identity matrix.

[0098] In this embodiment of the invention, after fusing data collected from various sensors to form sensor fusion data, edge computing technology can be used to extract key features from the sensor fusion data, thereby forming key operational data. Subsequent transmission of this key operational data to the cloud computing platform effectively reduces the amount of data transmitted, thus reducing transmission latency. The key features extracted using edge computing technology include: statistical features reflecting data distribution characteristics such as mean, variance, peak value, and kurtosis; time-domain features such as rate of change and duration; and frequency-domain features such as extracting the main frequency components using a sliding window FFT. After extracting the key features to form key operational data, the key operational data is uploaded to the cloud computing platform via high-speed wireless communication modules such as 5G.

[0099] Step 105: On the cloud computing platform, the key operational data is input into the fault detection deep neural network, so that the fault detection deep neural network extracts features from the key operational data to obtain shared feature data; the aging degree of the device under test is determined based on the shared feature data; the fault type probability distribution of the device under test is determined based on the shared feature data; a dynamic fault threshold is calculated based on the aging degree and a preset fault threshold; the fault type probability distribution and the dynamic fault threshold are compared to determine the fault detection result of the device under test.

[0100] In this embodiment of the invention, key operational data received by the cloud computing platform is input into a pre-trained fault detection deep neural network. Shared feature data can be obtained through feature extraction. Based on the shared feature data, the aging degree and fault type probability distribution of the device under test are analyzed and obtained. The preset static fault threshold is corrected based on the aging degree of the device under test to form a dynamic fault threshold. Thus, the fault detection result of the device under test is obtained according to the fault type probability distribution and the dynamic fault threshold.

[0101] As a preferred embodiment, the training process of the fault detection deep neural network is as follows:

[0102] Acquire data samples from various devices within the power substation; wherein, the data samples include temperature data, vibration data, current data, and noise data under various preset fault types and normal operating conditions, as well as aging labels and fault labels;

[0103] Based on the data samples, a shared feature extraction layer in a deep neural network is trained to learn and extract deep features;

[0104] Based on the data samples, a nonlinear degrader in a deep neural network is trained to learn the mapping relationship between deep features and aging labels;

[0105] Based on the data samples, a fault diagnostic tool in a deep neural network is trained to learn the mapping relationship between deep features and fault labels;

[0106] Once the shared feature extraction layer, the nonlinear degradation generator, and the fault diagnosticer have been trained, the deep neural network is identified as a fault detection deep neural network.

[0107] In this embodiment of the invention, the structure of the fault detection deep neural network mainly includes a shared feature extraction layer, a nonlinear degradation unit, and a fault diagnostic unit. The training process of the fault detection deep neural network first involves acquiring the data samples required for model training. These data samples cover the operational status data of various equipment in a power substation under various preset fault types and normal operating conditions. The operational status data includes temperature data, vibration data, current data, and noise data. Aging labels and fault labels corresponding to each operational status data are then obtained from a database. The operational status data, aging labels, and fault labels are combined to form a model training dataset. The deep neural network is trained using this dataset, enabling the shared feature extraction layer to learn to extract deep features, the nonlinear degradation unit to learn the mapping relationship between deep features and aging labels, and the fault diagnostic unit to learn the mapping relationship between deep features and fault labels. During model training, the cross-entropy loss function can be used to adjust the model parameters. The cross-entropy loss function is:

[0108]

[0109] In the formula, y is the loss function value; N is the total number of training samples; i Let i be the true label of the i-th sample; The model predicts the probability value for the i-th sample.

[0110] After the shared feature extraction layer, nonlinear degradation unit, and fault diagnostic unit are trained, the trained deep neural network is identified as the fault detection deep neural network, and the fault probability is calculated using the following formula during the inference phase:

[0111]

[0112] In the formula, σ represents the predicted probability of the model output; w represents the weight vector; x represents the input feature vector; and b represents the model bias term.

[0113] As a preferred embodiment, feature extraction is performed on the key operational data to obtain shared feature data, including:

[0114] By utilizing the shared feature extraction layer in the fault detection deep neural network, temperature features, vibration features, current features, and noise features are extracted from the key operational data, respectively.

[0115] The temperature characteristics, vibration characteristics, current characteristics, and noise characteristics are integrated to form shared feature data.

[0116] In this embodiment of the invention, a shared feature extraction layer in a fault detection deep neural network can be used to automatically extract data features of key operating data. Since key operating data includes temperature data, vibration data, current data, and noise data, the shared feature extraction layer can extract temperature features, vibration features, current features, and noise features. To facilitate subsequent analysis of potential relationships between data, the extracted temperature features, vibration features, current features, and noise features can be fused to form shared feature data.

[0117] As a preferred embodiment, determining the aging degree of the device under test based on the shared feature data includes:

[0118] The nonlinear degradation function in the fault detection deep neural network is used to determine the data acquisition time based on the shared feature data;

[0119] Based on the data acquisition time and the model initial time, determine the time characteristics;

[0120] The aging degree of the device under test is calculated using the following formula:

[0121]

[0122] In the formula, denoted as aging degree; f(·) is a nonlinear function; T is the temperature characteristic; V is the vibration characteristic; I is the current characteristic; N is the noise characteristic; t is the time characteristic; θ is the model parameter, updated through online adaptive calibration; ε is the random error constant.

[0123] In this embodiment of the invention, the nonlinear degradation model is a mathematical model that describes the nonlinear decline trend of things (such as equipment performance) over time, number of uses, or other cumulative factors. Since the aging (performance degradation) of equipment is the result of multiple mechanisms such as physical wear, chemical corrosion, material fatigue, and thermal aging, and the nature of these processes is often nonlinear, using a nonlinear degradation model to predict the degree of equipment aging can accurately characterize the complex degradation laws.

[0124] As a preferred embodiment, determining the fault type probability distribution of the device under test based on the shared feature data includes:

[0125] Using a fault diagnostic tool in a fault detection deep neural network, the fault probability of various preset fault types is determined based on the shared feature data and feature fault mapping relationship; wherein, the preset fault types include electrical faults, mechanical faults, environmental impact faults, and system faults.

[0126] The failure probabilities of each preset failure type are summarized to form a failure type probability distribution.

[0127] In this embodiment of the invention, the fault diagnostician in the fault detection deep neural network learns the mapping relationship between the features of the device and the fault. Therefore, in practical applications, after extracting the shared feature data of the device under test, the fault probability of each preset fault type can be output respectively, and then the fault probabilities can be summarized to form a fault type probability distribution.

[0128] As a preferred embodiment, the dynamic fault threshold is calculated based on the aging degree and the preset fault threshold, including:

[0129] Obtain the preset fault threshold for each of the preset fault types;

[0130] Based on the degree of aging, each of the preset fault thresholds is adjusted to a dynamic fault threshold.

[0131] The dynamic fault threshold is calculated using the following formula:

[0132] T dynamic =T base ×(1-α·A)

[0133] In the formula, T dynamic For dynamic fault thresholds; T base α is the preset fault threshold; A is the preset sensitivity coefficient; A is the degree of aging.

[0134] In this embodiment of the invention, the degree of aging is a proportional value ranging from 0% to 100%. The fault judgment threshold can be dynamically adjusted based on the degree of aging, making the fault threshold more adaptable to the dynamic changes in equipment performance over time. Therefore, by combining preset sensitivity coefficients and using the above formula, the preset fault thresholds corresponding to each preset fault type can be adjusted to dynamic fault thresholds. These dynamic fault thresholds can then be used to obtain the fault detection results of the device under test. Adjusting the winter threshold lowers the fault threshold of highly aging equipment, making it easier to trigger alarms; while maintaining a high fault threshold for new equipment avoids false alarms. By improving diagnostic accuracy and reducing false alarms and missed alarms, the service life of equipment is extended, and the replacement and maintenance costs are reduced.

[0135] As a preferred embodiment, the fault detection result of the device under test is determined by comparing the fault type probability distribution and the dynamic fault threshold, including:

[0136] The fault probabilities of each preset fault type in the fault type probability distribution are compared with the corresponding dynamic fault thresholds.

[0137] Preset fault types with a fault probability greater than the dynamic fault threshold are identified as predicted fault types.

[0138] Summarize all predicted fault types to determine the fault detection result of the device under test.

[0139] See Figure 2 This invention provides a schematic diagram of the workflow of a fault detection deep neural network. When raw sensor data is input into the fault detection deep neural network (DNN model), shared feature extraction layers within the DNN extract shared feature data. This shared feature data is then input into a multi-task output layer, which includes a nonlinear degradation unit and a fault diagnostic unit. The nonlinear degradation unit predicts aging to determine the degree of aging, while the fault diagnostic unit diagnoses discrete fault types and probabilities. Subsequently, the fault threshold is dynamically adjusted based on the degree of aging, and the fault detection result is derived from the dynamic fault threshold and the discrete fault type probabilities.

[0140] In this embodiment of the invention, after dynamically adjusting the fault threshold, the presence of preset fault types in the fault type probability distribution can be determined by comparing the fault probabilities of various preset fault types with the corresponding dynamic fault thresholds. For example, when the fault probability of preset fault type A is greater than the dynamic fault threshold of preset fault type A, it is determined that preset fault type A exists in the device under test, and preset fault type A is identified as the predicted fault type of the device under test. Similarly, the fault probabilities of other preset fault types are compared with the dynamic fault thresholds to determine several predicted fault types present in the device under test. These predicted fault types are then summarized to form the fault detection result of the device under test.

[0141] In this embodiment of the invention, by predicting the aging degree of the device under test, a gradient early warning mechanism can be employed to adjust the fault judgment control method and inspection frequency according to the aging degree. The gradient early warning mechanism is shown in the table below:

[0142]

[0143] In the table, Y represents the degree of aging. th1 and Y th2 These are the first aging threshold and the second aging threshold, respectively, where Y th1 <Y th2 ;I max This is the maximum current value of the device.

[0144] Implementing the above embodiments has the following effects:

[0145] This invention provides a fault detection method for power substations. During an inspection, the method determines the fault location of the equipment under test (EUT). Based on SLAM technology, it controls a fault detection robot to reach the fault location. The robot collects real-time operational data from the EUT. Using edge computing technology, the robot performs lightweight processing on the real-time operational data to generate key operational data, which is then transmitted to a cloud computing platform. On the cloud computing platform, the key operational data is input into a fault detection deep neural network, enabling the network to extract features and obtain shared feature data. The aging degree of the EUT is determined based on the shared feature data. The probability distribution of fault types is determined based on the shared feature data. A dynamic fault threshold is calculated based on the aging degree and a preset fault threshold. Finally, the fault detection result is determined by comparing the probability distribution of fault types with the dynamic fault threshold. This invention utilizes a fault detection robot to collect real-time operational data, and then uses edge computing technology to perform lightweight processing on the robot locally. This effectively reduces the amount of data transmitted and the data transmission latency, thereby improving the efficiency of fault data transmission. When performing fault detection on a cloud computing platform, the fault threshold is also dynamically adjusted based on the aging degree of the equipment, so that the fault threshold can adapt to the dynamic changes in equipment performance over time, thereby effectively improving the accuracy of fault detection.

[0146] like Figure 3 As shown, based on the above method embodiments, corresponding apparatus embodiments are provided;

[0147] One embodiment of the present invention provides a fault detection device for power substations, comprising: a location determination module, a robot control module, a data acquisition module, an edge computing module, and a fault detection module;

[0148] The location determination module is used to determine the fault detection location of the device under test during the inspection task;

[0149] The robot control module is used to control the fault detection robot to reach the fault detection location based on SLAM technology;

[0150] The data acquisition module is used to control the fault detection robot to collect real-time operating data of the device under test; wherein, the real-time operating data includes temperature data, vibration data, current data and noise data;

[0151] The edge computing module is used to control the fault detection robot to perform lightweight processing on the real-time operating data using edge computing technology, form key operating data, and transmit the key operating data to the cloud computing platform.

[0152] The fault detection module is used to input the key operating data into a fault detection deep neural network on a cloud computing platform, so that the fault detection deep neural network can extract features from the key operating data to obtain shared feature data; determine the aging degree of the device under test based on the shared feature data; determine the fault type probability distribution of the device under test based on the shared feature data; calculate a dynamic fault threshold based on the aging degree and a preset fault threshold; and determine the fault detection result of the device under test by comparing the fault type probability distribution and the dynamic fault threshold.

[0153] It is understood that the above-described device embodiments correspond to the method embodiments of the present invention, and can implement the power substation fault detection method provided by any of the above-described method embodiments of the present invention.

[0154] It should be noted that the device embodiments described above are merely illustrative, and some or all of the modules can be selected to achieve the purpose of this embodiment according to actual needs. Furthermore, in the accompanying drawings of the device embodiments provided by this invention, the connection relationships between modules indicate that they have communication connections, which can specifically be implemented as one or more communication buses or signal lines. Those skilled in the art can understand and implement this without any creative effort.

[0155] The specific embodiments described above further illustrate the purpose, technical solution, and beneficial effects of the present invention. It should be understood that the above descriptions are merely specific embodiments of the present invention and are not intended to limit the scope of protection of the present invention. In particular, it should be noted that any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the scope of protection of the present invention for those skilled in the art.

Claims

1. A method for fault detection in a power substation, characterized in that, include: During inspection tasks, determine the location of fault detection in the equipment under test; Based on SLAM technology, the fault detection robot is controlled to reach the fault detection location. The fault detection robot is controlled to collect real-time operating data of the device under test; wherein, the real-time operating data includes temperature data, vibration data, current data and noise data; Using edge computing technology, the fault detection robot is controlled to perform lightweight processing on the real-time operating data to form key operating data, and the key operating data is transmitted to the cloud computing platform. On a cloud computing platform, the key operational data is input into a fault detection deep neural network, which extracts features from the key operational data to obtain shared feature data. The aging degree of the device under test is determined based on the shared feature data. The probability distribution of fault types of the device under test is determined based on the shared feature data. A dynamic fault threshold is calculated based on the aging degree and a preset fault threshold. The fault detection result of the device under test is determined by comparing the probability distribution of fault types and the dynamic fault threshold.

2. The power substation fault detection method according to claim 1, characterized in that, The method of controlling the fault detection robot to reach the fault detection location based on SLAM technology includes: Based on SLAM technology, the current pose of the fault detection robot is determined in real time; Based on the fault detection location, determine the target map coordinates in the pre-built power substation map; Based on the fault detection location, the target map coordinates, and the pre-built power substation map, a feasible optimal path is planned. Based on the feasible optimal path, the fault detection robot is controlled to reach the fault detection location.

3. The power substation fault detection method according to claim 1, characterized in that, The use of edge computing technology to control the fault detection robot to perform lightweight processing on the real-time operating data, forming key operating data, includes: The Kalman filter algorithm is used to fuse the temperature data, vibration data, current data and noise data in the real-time operation data to form sensor fusion data; By utilizing edge computing technology, the statistical features, time-domain features, and frequency-domain features of the sensor fusion data are extracted to form key operational data.

4. The power substation fault detection method according to claim 1, characterized in that, The step of extracting features from the key operational data to obtain shared feature data includes: By utilizing the shared feature extraction layer in the fault detection deep neural network, temperature features, vibration features, current features, and noise features are extracted from the key operational data, respectively. The temperature characteristics, vibration characteristics, current characteristics, and noise characteristics are integrated to form shared feature data.

5. The power substation fault detection method according to claim 4, characterized in that, Determining the aging degree of the device under test based on the shared feature data includes: The nonlinear degradation function in the fault detection deep neural network is used to determine the data acquisition time based on the shared feature data; Based on the data acquisition time and the model initial time, determine the time characteristics; The aging degree of the device under test is calculated using the following formula: In the formula, denoted as aging degree; f(·) is a nonlinear function; T is the temperature characteristic; V is the vibration characteristic; I is the current characteristic; N is the noise characteristic; t is the time characteristic; θ is the model parameter; ε is the random error constant.

6. The power substation fault detection method according to claim 5, characterized in that, Determining the probability distribution of fault types of the device under test based on the shared feature data includes: Using a fault diagnostic tool in a fault detection deep neural network, the fault probability of various preset fault types is determined based on the shared feature data and feature fault mapping relationship; wherein, the preset fault types include electrical faults, mechanical faults, environmental impact faults, and system faults. The failure probabilities of each preset failure type are summarized to form a failure type probability distribution.

7. The power substation fault detection method according to claim 6, characterized in that, The calculation of the dynamic fault threshold based on the aging degree and the preset fault threshold includes: Obtain the preset fault threshold for each of the preset fault types; Based on the degree of aging, each of the preset fault thresholds is adjusted to a dynamic fault threshold. The dynamic fault threshold is calculated using the following formula: T dynamic =T base ×(1-α·A) In the formula, T dynamic For dynamic fault thresholds; T base α is the preset fault threshold; A is the preset sensitivity coefficient; A is the degree of aging.

8. The power substation fault detection method according to claim 7, characterized in that, The step of comparing the fault type probability distribution and the dynamic fault threshold to determine the fault detection result of the device under test includes: The fault probabilities of each preset fault type in the fault type probability distribution are compared with the corresponding dynamic fault thresholds. Preset fault types with a fault probability greater than the dynamic fault threshold are identified as predicted fault types. Summarize all predicted fault types to determine the fault detection result of the device under test.

9. The power substation fault detection method according to claim 6, characterized in that, The training process of the fault detection deep neural network is as follows: Acquire data samples from various devices within the power substation; wherein, the data samples include temperature data, vibration data, current data, and noise data under various preset fault types and normal operating conditions, as well as aging labels and fault labels; Based on the data samples, a shared feature extraction layer in a deep neural network is trained to learn and extract deep features; Based on the data samples, a nonlinear degrader in a deep neural network is trained to learn the mapping relationship between deep features and aging labels; Based on the data samples, a fault diagnostic tool in a deep neural network is trained to learn the mapping relationship between deep features and fault labels; Once the shared feature extraction layer, the nonlinear degradation generator, and the fault diagnosticer have been trained, the deep neural network is identified as a fault detection deep neural network.

10. A fault detection device for power substations, characterized in that, include: The module includes a location determination module, a robot control module, a data acquisition module, an edge computing module, and a fault detection module. The location determination module is used to determine the fault detection location of the device under test during the inspection task; The robot control module is used to control the fault detection robot to reach the fault detection location based on SLAM technology; The data acquisition module is used to control the fault detection robot to collect real-time operating data of the device under test; wherein, the real-time operating data includes temperature data, vibration data, current data and noise data; The edge computing module is used to control the fault detection robot to perform lightweight processing on the real-time operating data using edge computing technology, form key operating data, and transmit the key operating data to the cloud computing platform. The fault detection module is used to input the key operating data into a fault detection deep neural network on a cloud computing platform, so that the fault detection deep neural network can extract features from the key operating data to obtain shared feature data; determine the aging degree of the device under test based on the shared feature data; determine the fault type probability distribution of the device under test based on the shared feature data; calculate a dynamic fault threshold based on the aging degree and a preset fault threshold; and determine the fault detection result of the device under test by comparing the fault type probability distribution and the dynamic fault threshold.

Images