Isolation switch cabinet fault detection method, apparatus and device, and storage medium

Abstract

The invention provides an isolation switch cabinet fault detection method and device, equipment and a storage medium, relates to the technical field of power system monitoring, and solves the problem of low fault detection accuracy of an isolation switch cabinet in the related technology. According to the scheme, the modal features corresponding to the modal data are identified and analyzed by using the physical information neural network, the knowledge base constructed based on the knowledge graph and the fault mapping table, so that the fault condition of the isolation switch cabinet can be cooperatively analyzed in combination with multiple pieces of modal data, the omission ratio and the false alarm rate are reduced, and the reliability of the isolation switch cabinet is improved. The accuracy of fault detection of the isolation switch cabinet is effectively improved.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of power system monitoring, and in particular to a disconnector cabinet fault detection method, device, equipment and storage medium. BACKGROUND

[0002] As an electrical device, the disconnector cabinet can play a role in safe isolation and load protection, and has multiple protection modes such as short circuit, overload and leakage, so that the disconnector cabinet is widely used in power systems in the fields of rail transit and industrial production. Therefore, in order to ensure the stable operation of the power system, the fault detection of the disconnector cabinet becomes an important daily management project.

[0003] In related technologies, manual inspection or low-frequency PSCADA (Power Supervisory Control And Data Acquisition) system detection is usually used. However, the manual inspection method only relies on the visual observation and detection of the disconnector cabinet by the inspector using a handheld detection instrument. This method has strong subjectivity of the detection result and depends on the experience of the inspector, and has low accuracy in fault detection of the disconnector cabinet. The PSCADA system cannot capture high-frequency transient signals, but only analyzes through a single data source, which is prone to missed detection. Moreover, the detection is only a simple threshold judgment, which can only simply determine whether there is an anomaly, so that this method is prone to misjudgment, and lacks physical basis in the judgment process, and cannot determine the fault type based on the analysis of the collected data at the physical mechanism level, so that the accuracy and explainability of fault detection are low, and it is difficult to achieve a practical effect in engineering application. SUMMARY

[0004] The present application provides a disconnector cabinet fault detection method, device, equipment and storage medium, which solves the problem of low accuracy of fault detection of the disconnector cabinet in related technologies. The present application can analyze the fault condition of the disconnector cabinet in combination with multiple modal data, reduce the missed detection rate and false positive rate, and effectively improve the accuracy of fault detection of the disconnector cabinet.

[0005] In a first aspect, the present application provides a disconnector cabinet fault detection method, which comprises: preprocessing the received multiple modal data to determine the modal features corresponding to each modal data; inputting the modal features into a preset physical information neural network, and determining the fault type information and the first probability of the corresponding fault type information based on the physical information neural network; determine a second probability according to case probabilities corresponding to the plurality of similar cases; determine a third probability having a mapping relationship with the plurality of modal features based on a mapping relationship between the feature values and the probability values recorded in the fault mapping table; determine the weighted cumulative sum as the fault probability based on the first probability, the second probability and the third probability.

[0006] In a second aspect, the present application further provides a disconnector cabinet fault detection method and device, which comprises: a feature extraction module configured to pre-process the plurality of received modal data respectively, and determine modal features corresponding to each modal data; a first probability output module configured to input the modal features into a preset physical information neural network, and determine fault type information and a first probability corresponding to the fault type information based on the physical information neural network; a second probability output module configured to determine a plurality of similar cases associated with the modal features in a preset knowledge base, and determine a second probability according to case probabilities corresponding to the plurality of similar cases; a third probability output module configured to determine a third probability having a mapping relationship with the plurality of modal features based on a mapping relationship between the feature values and the probability values recorded in the fault mapping table; a probability fusion module configured to determine a weighted cumulative sum as the fault probability based on the first probability, the second probability and the third probability.

[0007] In a third aspect, the present application further provides an electronic device, which comprises: one or more processors; a storage device configured to store one or more programs, when the one or more programs are executed by the one or more processors, the one or more processors implement the disconnector cabinet fault detection method of the present application.

[0008] In a fourth aspect, the present application further provides a storage medium storing computer executable instructions, when the computer executable instructions are executed by a processor, the computer executable instructions are used to execute the disconnector cabinet fault detection method of the present application.

[0009] The scheme provided in the application determines modal features and fusion features based on multiple different modal data, and then determines a first probability based on a physical information neural network, the first probability being a probability value for a preliminary judgment on the received modal data, and further combines historical cases and a fault mapping table to perform probability fusion on a fault probability to determine a final fault probability. Compared with the related art, the scheme can combine multiple modal data to cooperatively analyze the fault condition of the disconnecting switch cabinet, reduces the missed detection rate and the false positive rate, and thus effectively improves the accuracy of fault detection on the disconnecting switch cabinet. BRIEF DESCRIPTION OF DRAWINGS

[0010] Figure 1 A structural schematic diagram of an edge computing network is provided for an embodiment of the application.

[0011] Figure 2 A step schematic diagram of a disconnecting switch cabinet fault detection method is provided for an embodiment of the application.

[0012] Figure 3 A step schematic diagram of determining a prediction result based on a physical information neural network is provided for an embodiment of the application.

[0013] Figure 4 A step schematic diagram of determining a first probability is provided for an embodiment of the application.

[0014] Figure 5 A structural schematic diagram of a disconnecting switch cabinet fault detection device is provided for an embodiment of the application.

[0015] Figure 6 A structural schematic diagram of an electronic device is provided for an embodiment of the application. DETAILED DESCRIPTION

[0016] The embodiments of the application will be further described in detail below with reference to the drawings and embodiments. It can be understood that the specific embodiments described herein are only used to explain the embodiments of the application, and not to limit the embodiments of the application. In addition, it should be noted that, in order to facilitate description, only parts related to the embodiments of the application are shown in the drawings, and those skilled in the art should understand that, as long as the technical features are not mutually contradictory, any combination of technical features can constitute an optional embodiment.

[0017] The terms "first", "second", etc. in the specification and claims of the present application are used to distinguish similar objects, and are not used to describe a specific order or sequence. It should be understood that the data thus used can be interchanged under appropriate circumstances, so that the embodiments of the present application can be implemented in an order other than those illustrated or described herein, and the objects distinguished by "first", "second", etc. are generally of a kind and do not limit the number of objects, for example, the first object can be one or more. In addition, "and / or" in the specification and claims indicates at least one of the connected objects, and the character " / " generally indicates that the objects before and after are in an "or" relationship. In the description of the present application, "a plurality of" means two or more, and "several" means one or more.

[0018] Disconnecting switch cabinets are widely used in power systems to ensure stable operation of the power system. Therefore, fault detection of disconnecting switch cabinets has attracted more and more attention, and fault detection of disconnecting switch cabinets is added in daily management projects to detect the state of disconnecting switch cabinets in time, so as to eliminate faults in time when faults exist. In the related art, the fault detection of the disconnecting switch cabinet is usually performed by manual inspection, that is, the inspector checks the disconnecting switch cabinet by visual inspection or detects the devices in the disconnecting switch cabinet by hand-held detection instrument to determine whether the disconnecting switch cabinet is faulty by comparing data, but this method relies too much on the experience of the inspector, and the subjectivity of the detection result is strong, which cannot accurately detect the state of the disconnecting switch cabinet, and the method is inefficient in large-scale equipment inspection scenarios.

[0019] In addition, the disconnecting switch cabinet is also detected by a low-frequency PSCADA system in the related art. The low-frequency PSCADA system cannot capture high-frequency signals, resulting in loss of high-frequency signals, and its detection method only relies on a single data source for analysis and uses a simple threshold for judgment, which is easy to miss detection or misjudge, resulting in low accuracy in detecting the state of the disconnecting switch cabinet.

[0020] To this end, the present application provides a disconnecting switch cabinet fault detection method, which can be applied to an edge computing module in an edge computing network, as shown in Figure 1 Figure 1 ​An embodiment of the edge computing network structure provided by the present application is shown in the figure. In the edge computing network, the edge computing module 101 is connected with a plurality of different data acquisition devices, and the edge computing module 101 is also used for receiving a plurality of different modal data. The modal data received by the edge computing module 101 is derived from the connected data acquisition devices. Optionally, the modal data is associated with an isolation switch cabinet, for example, the modal data includes high-frequency current waveform data, infrared images and video images. Correspondingly, the data acquisition devices include a transient current sensor 102, an infrared camera 103 and a visible light camera 104, that is, the high-frequency current waveform data of the isolation switch cabinet is acquired by the transient current sensor 102 at a high frequency, the infrared image of the isolation switch cabinet is acquired by the infrared camera 103, and the video image of the isolation switch cabinet is acquired by the visible light camera 104. The visible light camera 104 is a camera device using a visible light frequency band and used for capturing, recording and transmitting visible light signals. The edge computing module 101 is used for implementing edge computing deployment, and then data processing is completed at the local end, so as to detect the state of the isolation switch cabinet, effectively improve the detection efficiency, and meet the real-time requirement.

[0021] Figure 2 An embodiment of the isolation switch cabinet fault detection method provided by the present application is shown in the figure. In an embodiment, the isolation switch cabinet fault detection method provided by the present application can detect the fault of the isolation switch cabinet based on the acquired plurality of modal data, determine the state of the current isolation switch cabinet from the data of different modalities, and improve the accuracy of the fault detection of the isolation switch cabinet. The specific steps are as follows: Step S110, the received plurality of modal data is respectively preprocessed to determine the modal features corresponding to each modal data.

[0022] Optionally, the electronic device executing the method can acquire the plurality of modal data from the data acquisition device, and further preprocess the modal data. The modal feature reflects the characteristic data of the modal data. For different modal data, different processing methods are adopted according to the data type of the modal data to realize the preprocessing of the modal data, and then the modal features are extracted.

[0023] In an embodiment, the plurality of modal data is high-frequency current waveform data, infrared images and video images. The high-frequency current waveform data includes the waveform of high-frequency transient signals (such as harmonic components, discharge arcs, etc.), which can be used to feed back the electrical data of the isolation switch cabinet. The infrared image is an image acquired by the infrared camera shooting the isolation switch cabinet, which can be used to feed back the temperature information of the isolation switch cabinet. The video image is an image acquired by the visible light camera shooting the isolation switch cabinet, which can be used to feed back the object state of the isolation switch cabinet.

[0024] After obtaining the high-frequency current waveform data, the infrared image and the video image, the received data is denoised, for example, based on wavelet transform, and the high-frequency current waveform data, the infrared image and the video image are denoised respectively. In this regard, in the denoising process, the signal is decomposed by wavelet transform to distinguish the wavelet coefficients of the signal and the noise in different scales, and the noise is removed by threshold processing, and finally the signal is reconstructed by inverse wavelet transform to finally retain the high-frequency current waveform data, the infrared image and the video image after removing the noise. It should be noted that in some embodiments, the received data can also be denoised by mean filtering or median filtering.

[0025] Further, based on Fourier transform, the fundamental component and the harmonic component in the denoised high-frequency current waveform data are determined, and the first feature is determined according to the fundamental component and the harmonic component. The high-frequency current waveform data includes the fundamental component and the harmonic component, the fundamental component is a component with order equal to 1 in the Fourier series of the periodic quantity, and the harmonic component is a component with order greater than 1 in the Fourier series of the periodic quantity. Fourier transform is a harmonic analysis method, and the fundamental component and the harmonic component in the high-frequency current waveform data can be decomposed by Fourier transform. After determining the fundamental component and the harmonic component, the first feature is determined according to the fundamental component and the harmonic component, so that the first feature is taken as the characteristic value of the electrical data of the high-frequency current waveform data, i.e. the first feature is taken as the modal feature corresponding to the high-frequency current waveform data. Optionally, the value of the first feature is the ratio of the fundamental component to the harmonic component.

[0026] As for the modal feature (i.e. the second feature) corresponding to the infrared image, the temperature gradient value in the denoised infrared image is determined, and the temperature gradient value is taken as the second feature. Optionally, the infrared image is converted into a gray matrix, i.e. the gray value is used to represent each pixel point in the infrared image, and the gray matrix is formed according to the gray value to represent the infrared image, and then the temperature gradient value is calculated using an edge detection algorithm such as Sobel operator to reflect the temperature change direction and intensity.

[0027] For the modal feature corresponding to the video image (i.e., the third feature), the pixel offset of the same target object between the denoised video images is determined, and the pixel offset is taken as the third feature. Optionally, the target object in the video image is an electrical switch in the disconnecting cubicle, and in the process of determining the pixel offset, target object recognition is performed on the video image, such as using a Blob analysis method, a template matching method, a target detection algorithm based on a CNN (Convolutional Neural Networks), and the like, to recognize the target object in the image, and then compare the pixel positions of the target object in different video images. For example, since the visible light camera is used to collect video images of the disconnecting cubicle, the camera position is fixed, and accordingly the acquired video images are images based on the same shooting angle. In this case, a same coordinate point (such as the center point of the disconnecting cubicle) is taken as the origin, and the pixel position of the target object on each video image is determined. By comparing the pixel positions of the target object in different video images, the pixel offset of the target object is determined, and the pixel offset is taken as the third feature. The third feature can be used to represent the deformation of the target object.

[0028] In some embodiments, the fault detection of the disconnecting cubicle is performed periodically, and in each period, a plurality of infrared images and video images can be collected. Based on this, the temperature gradient value and the pixel offset value in the period can be determined. In this case, the second feature and the third feature can be represented in the form of a multi-dimensional vector or matrix to represent the value change of the modal feature in the period.

[0029] Therefore, by preprocessing the received modal data, the modal features in the modal data can be extracted to represent the information of the disconnecting cubicle at different levels through the modal features, so that the data sources can be expanded, and the disconnecting cubicle can be more comprehensively analyzed, which helps to improve the accuracy of fault detection.

[0030] In step S120, the modal feature is input into a preset physical information neural network, and based on the physical information neural network, the fault type information and the first probability corresponding to the fault type information are determined.

[0031] The physical information neural network (Physics-Informed Neural NetworKs, PINNs) is a deep learning framework combining a data-driven method and a physical constraint, which directly embeds the physical law (usually in the form of a partial differential equation, an initial condition, and a boundary condition) into the training process of the neural network, so that the neural network approximates the target function while also satisfying the known physical law. The physical information neural network used in the present embodiment is a neural network that has completed model training, and the trained physical information neural network can analyze whether the disconnecting cubicle is faulty and the corresponding probability.

[0032] Further, the modal features are taken as input information of the physical information neural network and input into the physical information neural network. The loss function used in the model training process of the physical information neural network adds a physical information error term, which is used to represent the physical law followed, for example, in the form of a physical equation, and is further embedded in the model training process of the physical information neural network. In this regard, in the model training process, not only the data error is minimized, but also the physical information error is minimized to ensure that the prediction result conforms to the configured physical law.

[0033] Optionally, the physical information neural network is constructed based on a fully connected deep neural network, such as including an input layer, a hidden layer, and an output layer, wherein the hidden layer is implemented based on a fully connected layer, the fully connected layer is connected to all nodes of the previous layer, used to synthesize the features extracted in the front, i.e., the result of the next layer is the weighted accumulation of each neuron of the previous layer, and a nonlinear activation function is used to realize nonlinear mapping, such as ReLU, sigmoid, or tanh. And the physical law is embedded in the loss function in the form of a partial differential equation, the loss function includes a data error term and a physical information error term, and the partial differential equation corresponding to the physical law is taken as the physical information error term, wherein the data error term is used to quantify the difference between the model prediction result and the actual data, and the physical information error term is used to quantify the difference between the model prediction result and the physical equation residual, so as to determine that the model prediction result follows the known physical law. Further, the loss value is determined based on the above loss function in the model training process, and the gradient direction of the loss function is determined by using an optimization algorithm (such as a gradient descent algorithm) to optimize the model parameters.

[0034] In an embodiment, the physical information neural network takes a first physical equation corresponding to power balance as a physical constraint equation, and determines the physical information error term in the loss function based on the physical constraint equation, so that the model prediction result follows the power balance law, which is used to represent that the input power P in in the disconnecting switch cabinet is equal to the sum of the load power P load and the loss power P loss , i.e., P in = P load + P loss , when the power is unbalanced, it usually indicates poor contact or overload failure, and the corresponding physical constraint equation is as follows:

[0035] wherein L represents the physical quantity output by the physical constraint equation, and λ is a coefficient. In this regard, the power imbalance can be determined when L is greater than 0.

[0036] The physical information neural network also uses the second physical equation corresponding to thermal balance as a physical constraint equation. The thermal balance equation describes the balance between heat generated by current and heat dissipation in the disconnector cabinet. When the heat generated and heat dissipated are not balanced, thermal imbalance (i.e., thermal balance is disrupted) is determined. A disruption of thermal balance typically indicates poor heat dissipation or abnormal contact resistance. The corresponding physical constraint equation is as follows:

[0037]

[0038]

[0039] Where ΔT is the temperature change, I is the current through the disconnector, R is the equivalent resistance, which includes contact resistance and conductor resistance, T is the current temperature, T0 is the ambient temperature, K1 is the heat generation coefficient, which is determined by the material properties and structural parameters of the disconnector. The specific value of K1 is determined when the material and structure of the disconnector are determined. ρ is the material density, Cp is the specific heat capacity, and V is the heat generation volume. K2 is the heat dissipation coefficient, which is determined by the heat dissipation area and convective heat transfer coefficient of the disconnector. Similarly, when the disconnector is determined, K2 is also determined. h is the convective heat transfer coefficient, and A is the heat dissipation surface area.

[0040] The physical information neural network also uses the third physical equation corresponding to the contact resistance variation law as the physical constraint equation. The contact resistance variation law is used to describe the relationship between the contact resistance and temperature, oxidation degree and other factors, and is used to predict the degree of contact degradation. The corresponding physical constraint equation is as follows:

[0041] Among them, R j is the contact resistance; R0 is the initial value of the contact resistance, such as the reference value corresponding to the contact resistance of the disconnector cabinet when it leaves the factory or after maintenance as the above-mentioned initial value, which can be obtained by direct measurement or testing; ΔT is the temperature difference of the contact rise, α is the temperature coefficient, which is used to express the rate of change of the contact resistance with temperature, and its specific value is determined according to the contact material. For example, for silver contacts, the value of α is 0.0005-0.002; β is the oxidation coefficient, which is used to express the degree of influence of contact oxidation on the contact resistance; ΔN is the oxidation factor, which is used to express the oxidation influence on the contact resistance, and is determined according to the equipment operation time and environmental conditions. For example, the ratio of the current time to the design life multiplied by the environmental corrosion coefficient is used as the value of the oxidation factor, and the environmental corrosion coefficient is determined according to the humidity and pollution level of the installation environment. The corresponding value range is optionally 0.5-2.0.

[0042] The constructed physical information error term is used to guide the physical information neural network to adjust the model parameters and minimize the physical quantity output by the physical constraint equation in the case of power imbalance, heat imbalance or abnormal contact resistance change. To this end, the physical information neural network is used to predict the fault of the disconnecting switch cabinet to determine the corresponding fault type with a probability value, and the fault type information in the prediction result and the first probability of the corresponding fault type information.

[0043] The physical constraint equation adopted by the embodiments of the present application can directly correspond to the typical fault mode of the disconnecting switch cabinet. The first physical equation corresponding to power balance is used to detect poor contact and overload faults. The second physical equation corresponding to heat balance is used to detect abnormal heat dissipation and temperature faults. The third physical equation corresponding to the contact resistance change rule is used to predict contact degradation and mechanical looseness, thereby guiding fault detection from the physical mechanism level. Compared with the simple threshold judgment method in the prior art which lacks physical basis, the present scheme can distinguish different types of faults and give corresponding physical explanations, thereby improving the accuracy, interpretability and engineering practicability of fault detection.

[0044] In step S130, a plurality of similar cases associated with the modal feature are determined in the preset knowledge base, and a second probability is determined according to case probabilities corresponding to the plurality of similar cases.

[0045] The knowledge base is a database constructed based on knowledge graph technology, which stores corresponding entity relationship information to represent the association relationship between different entities, such as recording the association relationship between the feature value and the fault case through the entity relationship information. Then, a plurality of similar cases associated with the modal feature are determined in the preset knowledge base, such as determining the association degree of each case with the modal feature according to the entity relationship information, so as to determine a plurality of similar cases therefrom. Moreover, each similar case also corresponds to a case probability, which is the occurrence probability of the fault corresponding to the case, and then a second probability is determined according to the case probabilities corresponding to the plurality of similar cases. Optionally, the average value of the case probabilities of the plurality of similar cases is taken as the value of the second probability.

[0046] In some embodiments, the knowledge base can be updated based on historical cases, for example, all parameters (such as the determined modal feature, fault type information and fault probability) judged to exist fault conditions in the detection period are taken as historical cases, and corresponding entity relationship information is constructed to be stored in the knowledge base, thereby enriching the knowledge base, thereby helping to improve the accuracy of fault detection.

[0047] In step S140, a third probability having a mapping relationship with the plurality of modal features is determined based on the mapping relationship between the feature values and the probability values recorded in the fault mapping table.

[0048] The fault mapping table is used to record the mapping relationship between the characteristic value and the probability value. Optionally, the probability value recorded in the fault mapping table is a pre-set empirical value corresponding to the corresponding characteristic value, that is, the characteristic value and the probability value in the fault mapping table are one-to-one corresponding, so that the mapping relationship between the characteristic value and the probability value is recorded in the fault mapping table. Then, in the case of determining the modal characteristic, based on the characteristic value corresponding to the modal characteristic, the fault mapping table is searched to determine the probability value having the mapping relationship with the modal characteristic, and the probability value is taken as the third probability. For example, the modal characteristic includes a first characteristic corresponding to the high-frequency current waveform data, a second characteristic corresponding to the infrared image, and a third characteristic corresponding to the video image, and the mapping relationship between the characteristic value of the modal characteristic and the corresponding probability value is recorded in the fault mapping table, such as the modal characteristic is represented in the form of a multi-dimensional vector (f1, f2, f3) to identify the included first characteristic f1, second characteristic f2 and third characteristic f3, wherein different multi-dimensional vectors of different characteristic values correspond to different probability values, and each multi-dimensional vector (f1, f2, f3) in the fault mapping table has a mapping relationship with a probability value. Then, after determining the characteristic value of the modal characteristic, the multi-dimensional vector matching the characteristic value of the modal characteristic can be determined from the fault mapping table through table lookup, so as to determine the probability value having the mapping relationship with the modal characteristic according to the recorded mapping relationship, and take it as the third probability.

[0049] In step S150, the corresponding weighted cumulative sum is determined as the fault probability based on the first probability, the second probability and the third probability.

[0050] The corresponding probability, that is, the first probability, the second probability and the third probability, is determined by the different ways shown in the above examples, and the fault probability corresponding to the current fault type can be finally determined. Optionally, the first probability, the second probability and the third probability are weighted and accumulated to calculate the corresponding weighted cumulative sum, wherein the weight value set for each probability is a pre-set value, which is a weight value set according to the contribution degree of the physical information neural network, the knowledge base and the fault mapping table.

[0051] As can be seen from the above scheme, the scheme can improve the prediction efficiency by using edge computing deployment, and by processing modal data of multiple different modalities, the modal characteristics corresponding to the modal data are identified and analyzed by using the physical information neural network, the knowledge base constructed based on the knowledge graph and the fault mapping table, thereby improving the accuracy of the fault detection of the disconnecting switch cabinet, and the fault analysis by multiple levels of data can effectively reduce the missed detection rate and the false alarm rate.

[0052] In one embodiment, for the modal characteristics of the high-frequency current waveform data (i.e., the first characteristic described above), the total harmonic distortion rate can be used as the specific value of the first characteristic. After determining the fundamental component and harmonic components based on Fourier transform, the arithmetic square root of the harmonic amplitude in each harmonic component is determined, and then the ratio between the arithmetic square root and the fundamental amplitude in the fundamental component is determined, and the ratio is used as the first characteristic. The specific calculation formula is as follows:

[0053] Among them, THDi is the first characteristic, V1 is the fundamental amplitude, V h is the harmonic amplitude corresponding to the hth harmonic component. The selected total harmonic distortion (THD) represents the distortion rate of the corresponding current, reflecting the harmonic proportion in the current waveform and closely related to load characteristics. By determining the THD and using it as a modal feature of the high-frequency current waveform data, this solution can more accurately extract the electrical data of the disconnector cabinet to determine its electrical characteristics, thereby improving the accuracy of disconnector cabinet fault detection.

[0054] Figure 3 This is a schematic diagram of the steps for determining prediction results based on a physical information neural network, provided in one embodiment of the present application. In one embodiment, the physical information neural network includes an input layer, a hidden layer, and an output layer. The hidden layer performs a nonlinear transformation on the input data based on three fully connected layers to extract the corresponding eigenvalues. The physical information neural network also introduces Kirchhoff's current as a physical constraint, that is, the physical equation corresponding to Kirchhoff's current law is used as the physical constraint equation to determine the physical information error term in the loss function. Based on this, this solution uses this physical information neural network to predict modal features to determine the corresponding fault type information and the first probability. The specific steps are as follows: Step S210: input the modal features into the hidden layer through the input layer.

[0055] Step S220: weighted summing is performed on the modal features according to corresponding weights through the hidden layer, and nonlinear mapping is performed based on the activation function to determine a first probability and output the first probability through the output layer.

[0056] Step S230: Based on the first probability, determine the device state associated with the target probability interval in which the first probability is located, and add the device state to the fault type information.

[0057] The input layer is configured to access the input information and transmit the input information to the hidden layer. The hidden layer is implemented based on three full connection layers. Each node in each full connection layer is connected with all nodes in the previous layer, and is configured to synthesize the features extracted in the previous step, i.e., the result of the next layer is the weighted cumulative sum of the neurons in the previous layer, and a nonlinear mapping is implemented based on a nonlinear activation function. The output layer is configured to output the processing result of the hidden layer. Based on this, the modal features are input to the hidden layer through the input layer, and then the hidden layer performs weighted summation on the modal features according to the corresponding weights to determine the corresponding weighted cumulative sum. A nonlinear mapping is also performed based on an activation function, for example, a ReLU activation function is used to implement the nonlinear mapping, and the determined first probability is output through the output layer.

[0058] Further, after the first probability is determined, the device state of the disconnecting switch cabinet can be determined based on a target probability interval in which the first probability is located. Optionally, the device state of the disconnecting switch cabinet includes a normal state, a warning state and a fault state, and different probability intervals are set for different device states, for example, a probability interval in which the probability value is greater than 0 and less than a first preset probability value is a probability interval corresponding to the normal state, a probability interval in which the probability value is greater than or equal to the first preset probability value and less than a second preset probability value is a probability interval corresponding to the warning state, and a probability interval in which the probability value is greater than the second preset probability value and less than 1 is a probability interval corresponding to the fault state. Therefore, the probability interval in which the first probability is located can be determined according to the value of the first probability, and the probability interval is taken as the target probability interval. Then, the device state corresponding to the target probability interval is taken as the current device state of the disconnecting switch cabinet, and the device state is taken as the fault type information.

[0059] It should be noted that in some embodiments, the modal features and the corresponding feature weights are also added to the fault type information, for example, the modal features include a first feature corresponding to the high-frequency current waveform data, a second feature corresponding to the infrared image and a third feature corresponding to the video image. Then, the first feature, the second feature and the third feature are added to the fault type information, so that it can be determined according to the modal features and the corresponding feature weights which type of fault the current fault is associated with, i.e., the electrical fault, the temperature abnormal fault or the deformation fault. Optionally, the type of fault associated with the current fault is determined based on the maximum value in the product value of the modal features and the corresponding feature weights. Therefore, by using the physical information neural network, the device state of the disconnecting switch cabinet can be determined based on the fault type information, and the fault factor can also be determined, which is beneficial to the subsequent maintenance and repair work.

[0060] Figure 4The step of determining the first probability provided by an embodiment of the present application is schematically shown as follows. In some embodiments, the first probability is further determined by analyzing and judging the modal feature through the physical information neural network, and the specific steps are as follows: In step S310, the modal feature is analyzed abnormally through the physical information neural network, and it is determined whether the modal feature is an abnormal feature.

[0061] In step S320, in the case that the modal feature is an abnormal feature, a target case matching the abnormal feature is found in historical cases based on the abnormal feature, and an initial probability corresponding to the target case and a fault type corresponding to the target case are determined as the fault type information.

[0062] In step S330, the abnormal feature is verified based on a physical constraint mechanism built in the physical information neural network, and a physical constraint compliance degree is determined. The physical constraint compliance degree is used to represent the matching degree of the abnormal feature and a physical equation corresponding to the physical constraint mechanism.

[0063] In step S340, a confidence degree is determined according to the physical constraint compliance degree. The confidence degree is used to represent the increment of the matching degree of the abnormal feature and the physical equation to the initial probability.

[0064] In step S350, a probability sum is determined according to the initial probability and the confidence degree, and is used as the first probability.

[0065] The physical information neural network can be used to analyze the abnormality of the modal feature. It is conceivable that the physical information neural network corresponds to different features and configures corresponding thresholds. For example, the modal feature includes a first feature corresponding to high-frequency current waveform data, a second feature corresponding to an infrared image, and a third feature corresponding to a video image. When the total harmonic distortion rate is taken as a specific value of the first feature, if the total harmonic distortion rate exceeds a first threshold preset for the first feature, it is determined that the first feature is abnormal. When the temperature gradient value is taken as the second feature, if the temperature gradient value exceeds a second threshold preset for the second feature, it is determined that the second feature is abnormal. When the pixel offset is taken as the third feature, if the pixel offset exceeds a third threshold preset for the third feature, it is determined that the third feature is abnormal. In this case, if any one of the first feature, the second feature, and the third feature is abnormal, it is determined that the modal feature is an abnormal feature.

[0066] The database of the device stores historical cases, and the historical cases also store corresponding modal features as fault cases, to record the relationship between the modal features and the historical cases in the database. For example, a historical case is a case of the fault type of contact oxidation, and the corresponding modal data of the historical case has a total harmonic distortion rate (corresponding to the first feature) of 3.2%, a temperature gradient value (corresponding to the second feature) of 4.2°C / cm, and a pixel offset (corresponding to the third feature) of 0.08mm. In this regard, according to the specific values in the abnormal features, all historical cases are traversed, and then the target case corresponding to the matching of the modal features and the abnormal features of the case is found, for example, by calculating the cosine similarity of the modal features and the abnormal features corresponding to each case, to determine that they match when the calculated cosine similarity is less than a preset value.

[0067] And it is also determined that the initial probability corresponding to the target case can be determined according to the target case, for example, according to the probability distribution of the target case in the historical cases, the initial probability corresponding to the target case is determined, that is, the probability value is determined according to the number ratio of the target case in the historical cases, to be used as the initial probability.

[0068] Further, the abnormal features are verified based on the physical constraint mechanism built in the physical constraint neural network. It can be envisaged that the physical constraint mechanism is represented in the form of a physical constraint equation in the physical constraint neural network, such as the first, second and third physical equations provided in the above embodiments. Accordingly, the obtained abnormal features are input into the physical constraint equation to determine whether the abnormal features satisfy the physical constraint mechanism, for example, the physical constraint neural network is configured with a physical equation corresponding to the Kirchhoff's current law, and then the input current sum is verified when the abnormal features are verified to determine whether the output current sum is equal to the output current sum, to determine whether the abnormal features satisfy the physical constraint mechanism. In this regard, the physical constraint compliance degree represents the matching degree of the abnormal features and the physical equation corresponding to the physical constraint mechanism, and then the matching degree is quantified to determine whether the physical constraint mechanism is satisfied.

[0069] And the confidence is determined based on the corresponding physical constraint compliance degree. It can be understood that the confidence is used to represent the increase of the initial probability by the matching degree of the abnormal feature value and the physical equation, that is, the confidence is a quantitative index, which can represent the influence of the corresponding matching degree on the initial probability, and then the probability sum of the initial probability and the confidence is used as the first probability. Therefore, the physical information neural network can analyze the modal features to determine the corresponding probability, thereby increasing the accuracy of the fault detection of the disconnecting switch cabinet.

[0070] In an embodiment, the physical information neural network verifies the abnormal features based on physical equations to determine a matching degree of the abnormal features with the physical equations, wherein the physical equations adopted include a first physical equation corresponding to power balance, a second physical equation corresponding to heat balance, and a third physical equation corresponding to contact resistance change. In the case where input power and load power are determined, the loss power can be determined based on the input power and the load power, i.e., taking the difference between the input power and the load power as the loss power. Wherein, the input power is determined based on the voltage value and the current value in the first feature, i.e., determined by referring to the power calculation formula (i.e., P=UI, P is power, U is voltage, and I is current); and the load power can be measured. And a theoretical loss value is also provided, for which a first difference between the loss power and the theoretical loss value is determined, and in the case where the first difference is greater than a preset threshold, it is determined that the first feature does not match the first physical equation, and then the first compliance degree is taken as the physical constraint compliance degree corresponding to the first physical equation, such as taking 1 minus the ratio of the first difference to the maximum power as the first compliance degree.

[0071] And, according to the current value and the resistance value in the first feature, a theoretical temperature rise value is determined, which corresponds to the temperature change caused by the increase of the energization time, such as referring to the above-mentioned second physical equation corresponding to heat balance, and taking the current value as I in the equation and the resistance value as R in the equation, to take the calculated temperature change as the theoretical temperature rise value. For this, it is determined whether the theoretical temperature rise value is equal to the measured temperature rise value, and when there is a second difference between the theoretical temperature rise value and the measured temperature rise value in the second feature, it means that the theoretical temperature rise value is not equal to the measured temperature rise value, wherein the measured temperature rise value can be determined based on the second feature, such as based on the temperature gradient value determined from the infrared image, to determine the temperature value of the disconnecting switch cabinet based on the temperature gradient value, and then determine the measured temperature rise value. For this, in the case where it is determined that the second difference exists, it is determined that the second feature does not match the second physical equation, and then the second compliance degree is taken as the physical constraint compliance degree corresponding to the second physical equation, such as taking 1 minus the ratio of the second difference to the maximum temperature rise as the second compliance degree.

[0072] In addition, according to the temperature value in the second feature and a preset oxidation coefficient, a predicted contact resistance is determined, such as referring to the above-mentioned third physical equation corresponding to the contact resistance change rule, substituting the temperature value and the oxidation coefficient into the equation, and taking the calculation result as the value of the predicted contact resistance. In the case where there is a third difference between the predicted contact resistance and the measured contact resistance, it is determined that the first feature does not match the third physical equation, and then the third compliance degree is determined, and the third compliance degree is taken as the physical constraint compliance degree corresponding to the third physical equation, such as taking 1 minus the ratio of the third difference to the measured contact resistance as the third compliance degree.

[0073] The first, second and third degrees of coincidence correspond to the matching degrees of the abnormal features and different physical equations respectively. Based on the determined first, second and third degrees of coincidence, the corresponding confidence is further determined to be superimposed with the initial probability, and then the value of the first probability is determined.

[0074] For example, the acquired modal features include first features, second features and third features, wherein the first features are electrical features (such as recording voltage, current, total harmonic distortion rate, etc. through feature values), the second features are temperature features (such as recording temperature, temperature gradient value, etc. through feature values), and the third features are mechanical features (such as recording pixel offset, deformation, etc. through feature values). By comparing the corresponding threshold values, it is determined whether the modal features are abnormal features. For example, the value of the total harmonic distortion rate (corresponding to the first feature) is 3.2%, the value of the temperature gradient value (corresponding to the second feature) is 4.2°C / cm, and the value of the pixel offset (corresponding to the third feature) is 0.08mm. The abnormal analysis determines that the total harmonic distortion rate is greater than the normal value (the threshold value set for the first feature), the temperature gradient value is greater than the warning value (the threshold value set for the second feature), and the pixel offset value is within the normal range and does not exceed the threshold value. Accordingly, the target case is searched in the historical cases. For example, there is a case corresponding to the abnormal features in the historical cases, and the feature of the case is matched with the abnormal features. In this case, the case of this type is taken as the target case, and the probability of the occurrence of the target case of this type is taken as the initial probability. In this way, the fault type can be determined as contact oxidation, and then added to the fault type information.

[0075] Furthermore, the abnormal features are verified by the built-in physical constraint mechanism corresponding to the physical equation to determine the matching degree of the abnormal features and the physical equation, and the matching degree is represented by the physical constraint coincidence degree. For example, the input power is determined by the current value and the voltage value in the first feature, and the loss power is determined by comparing the input power with the load power. If there is a first difference between the loss power and the theoretical loss value, the first coincidence degree corresponding to the first physical equation is determined, such as subtracting the ratio of the first difference to the maximum power from 1 as the first coincidence degree. For other physical equations, the corresponding physical constraint coincidence degree can be determined according to the above embodiments. Moreover, the confidence is determined by the physical constraint coincidence degree, wherein different physical constraint mechanisms are provided with corresponding confidence gain intervals. For example, when the physical constraint coincidence degree is less than the first interval boundary value, the corresponding confidence is a first set value; when the physical constraint coincidence degree is greater than or equal to the first interval boundary value and less than the second interval boundary value, the corresponding confidence is a second set value. Optionally, when the physical constraint coincidence degree is less than 1%, the corresponding confidence is 5%; when the physical constraint coincidence degree is greater than or equal to 1% and less than 3%, the corresponding confidence is 4%. Based on this, the initial probability and the confidence are accumulated to take the corresponding probability accumulation as the first probability.

[0076] In some embodiments, the case vector and the case probability corresponding to each historical case are recorded in the knowledge base through the entity relationship information, and the case vector is associated with the feature vector corresponding to the modal data of the historical case. In this regard, in the process of searching for similar cases in the knowledge base, the feature vector associated with the modal feature is determined, for example, the feature vector is constructed based on the modal feature, that is, the modal feature is represented in the form of a vector. Then, the similarity between the feature vector and each case vector is determined, and optionally, the value determined by different vector similarity calculation methods such as cosine similarity, Euclidean distance or Manhattan distance is taken as the similarity between the feature vector and each case vector. Then, a preset number of historical cases are selected as similar cases in order of similarity from high to low, that is, after the similarity between the feature vector and each case vector is determined, the historical cases with the highest similarity are selected as similar cases, and the number of selected historical cases meets the preset number, such as 3 historical cases with the highest similarity are selected as similar cases in an embodiment.

[0077] In the entity relationship information, the case vector and the case probability corresponding to each historical case are recorded, and after the historical cases as similar cases are determined, the case probability corresponding to the similar cases is determined based on the entity relationship information, that is, the case probability corresponding to the historical cases is taken as the case probability corresponding to the similar cases. In an embodiment, the case probability can be determined based on the number of detection periods and the number of historical cases, that is, the ratio of the number of historical cases of the same fault type to the number of detection periods is taken as the case probability of the historical case. Alternatively, the case probability corresponding to the historical case can be based on the output of a neural network model, which is used to predict the corresponding fault occurrence probability according to the input modal feature. In this regard, the case vector of the feature vector corresponding to the modal data of each historical case is input into the neural network model, and the model prediction result is taken as the corresponding case probability.

[0078] Then, after the case probability corresponding to all similar cases is determined, the weighted average of the case probability corresponding to all similar cases is determined as the value of the second probability. Therefore, the second probability is determined by the knowledge base in this scheme, so that the fault probability can be further determined in combination with the historical cases, which helps to improve the accuracy of the disconnector cabinet fault detection.

[0079] Figure 5 The structure diagram of the disconnector cabinet fault detection device provided by an embodiment of the present application is shown in the figure. The device is used to execute the disconnector cabinet fault detection method provided by the above-mentioned embodiments, and has the corresponding function modules and beneficial effects of the execution method. As shown in the figure, the device includes a feature extraction module 301, a first probability output module 302, a second probability output module 303, a third probability output module 304 and a probability fusion module 305.

[0080] The feature extraction module 301 is configured to pre-process the received multiple modal data respectively, and determine modal features corresponding to each modal data. The first probability output module 302 is configured to input the modal features into a preset physical information neural network, and determine fault type information and a first probability corresponding to the fault type information based on the physical information neural network. The second probability output module 303 is configured to determine multiple similar cases associated with the modal features in a preset knowledge base, and determine a second probability according to case probabilities corresponding to the multiple similar cases. The third probability output module 304 is configured to determine a third probability having a mapping relationship with the modal features based on a mapping relationship between feature values and probability values recorded in a fault mapping table. The probability fusion module 305 is configured to determine a weighted cumulative sum corresponding to the first probability, the second probability and the third probability as a fault probability.

[0081] On the basis of the above embodiment, the multiple modal data are high-frequency current waveform data, infrared images and video images, and the feature extraction module 301 is specifically configured as follows: Based on wavelet transform, the high-frequency current waveform data, the infrared images and the video images are all subjected to denoising processing; Based on Fourier transform, fundamental wave components and harmonic components in the denoised high-frequency current waveform data are determined, and a first feature is determined according to the fundamental wave components and the harmonic components; A temperature gradient value in the denoised infrared images is determined, and the temperature gradient value is taken as a second feature; Pixel offsets of corresponding same target objects between the denoised video images are determined, and the pixel offsets are taken as a third feature.

[0082] On the basis of the above embodiment, the feature extraction module 301 is further configured as follows: An arithmetic square root of a harmonic amplitude in each harmonic component is determined; A ratio between the arithmetic square root and a fundamental wave amplitude in the fundamental wave component is determined, and the ratio is taken as the first feature.

[0083] On the basis of the above embodiment, the physical information neural network takes a first physical equation corresponding to power balance, a second physical equation corresponding to heat balance and a third physical equation corresponding to contact resistance change as physical constraint equations, and determines a physical information error term in a loss function based on the physical constraint equations. The physical information error term is used to guide the physical information neural network to adjust model parameters and minimize physical quantities output by the physical constraint equations in the case of power imbalance, heat imbalance or contact resistance change anomaly.

[0084] On the basis of the above-mentioned embodiments, the physical information neural network comprises an input layer, a hidden layer and an output layer, the hidden layer performs nonlinear transformation on the input data based on a 3-layer fully connected layer to extract corresponding characteristic values, and the first probability output module 302 is specifically configured to: input the modal features into the hidden layer through the input layer; perform weighted summation on the modal features according to corresponding weights through the hidden layer, and perform nonlinear mapping based on an activation function to determine the first probability and output the first probability through the output layer; based on the first probability, determine a device state associated with a target probability interval where the first probability is located, and add the device state to the fault type information.

[0085] On the basis of the above-mentioned embodiments, the first probability output module 302 is specifically configured to: perform abnormality analysis on the modal features through the physical information neural network, and determine whether the modal features are abnormal features; in the case where the modal features are abnormal features, find a target case matching the abnormal features in historical cases based on the abnormal features, and determine an initial probability corresponding to the target case and a fault type corresponding to the target case as the fault type information; verify the abnormal features based on a physical constraint mechanism built in the physical information neural network, determine a physical constraint compliance degree, and the physical constraint compliance degree is used to represent a matching degree of the abnormal features and a physical equation corresponding to the physical constraint mechanism; determine a confidence degree according to the physical constraint compliance degree, and the confidence degree is used to represent an increment of the matching degree of the abnormal features and the physical equation to the initial probability; determine a probability sum according to the initial probability and the confidence degree, and take the probability sum as the first probability.

[0086] On the basis of the above-mentioned embodiments, in the case where the abnormal features comprise a first feature corresponding to high-frequency current waveform data and a second feature corresponding to an infrared image, the first probability output module 302 is specifically configured to: determine a loss power according to the input power and the load power; in the case where a first difference between the loss power and a theoretical loss value is greater than a preset threshold, determine that the first feature does not match the first physical equation, take 1 minus a ratio of the first difference to a maximum power as a first compliance degree, the first compliance degree is a physical constraint compliance degree corresponding to the first physical equation, and the input power is determined based on a voltage value and a current value in the first feature; determine a theoretical temperature rise value according to a current value and a resistance value in the first feature; In a case where a second difference value exists between the theoretical temperature rise value and the measured temperature rise value in the second feature, it is determined that the second feature does not match the second physical equation, and a second compliance degree is determined by subtracting a ratio of the second difference value to the maximum temperature rise from 1, the second compliance degree being a physical constraint compliance degree corresponding to the second physical equation; According to the temperature value in the second feature and a preset oxidation coefficient, a predicted contact resistance is determined. In a case where a third difference value exists between the predicted contact resistance and the measured contact resistance, it is determined that the first feature does not match the third physical equation, and a third compliance degree is determined by subtracting a ratio of the third difference value to the measured contact resistance from 1, the third compliance degree being a physical constraint compliance degree corresponding to the third physical equation.

[0087] On the basis of the above-mentioned embodiments, the case vectors and case probabilities corresponding to each historical case are recorded in the knowledge base through the entity relationship information, the case vector is associated with the feature vector corresponding to the modal data of the historical case, and the second probability output module 303 is specifically configured to: determine the feature vector associated with the modal feature, and determine the similarity between the feature vector and each case vector; select a preset number of historical cases as similar cases in a descending order of the similarity; determine the case probability corresponding to the similar cases based on the entity relationship information, and determine a weighted average value of the case probabilities corresponding to all the similar cases as the value of the second probability.

[0088] It should be noted that in the above-mentioned embodiments of the device, each module is only divided according to the functional logic, but is not limited to the above-mentioned division, as long as the corresponding functions can be realized; in addition, the specific names of the modules are only for mutual differentiation, and do not limit the protection scope of the embodiments of the present application.

[0089] Figure 6A structural schematic diagram of an electronic device for performing the disconnector cabinet fault detection method provided in the embodiments and having corresponding function modules and beneficial effects is provided in an embodiment of the present application. As shown in the figure, the device includes a processor 401, a memory 402, an input device 403, and an output device 404. The number of processors 401 can be one or more, and one processor 401 is taken as an example in the figure; the processor 401, the memory 402, the input device 403, and the output device 404 can be connected through a bus or other means, and connection through a bus is taken as an example in the figure. The memory 402, as a computer-readable storage medium, can be used to store software programs, computer executable programs, and modules, such as program instructions / modules of the disconnector cabinet fault detection method in the embodiments of the present application. The processor 401 performs corresponding various function applications and data processing by running the software programs, instructions, and modules stored in the memory 402, that is, implements the disconnector cabinet fault detection method described above.

[0090] The memory 402 can mainly include a program storage area and a data storage area, wherein the program storage area can store an operating system and at least one application required by a function; the data storage area can store data recorded or created during use, and the like. In addition, the memory 402 can include a high-speed random access memory, and can also include a non-volatile memory, such as at least one magnetic disk storage device, a flash memory device, or other non-volatile solid-state memory device. In some embodiments, the memory 402 can further include a memory remotely arranged with respect to the processor 401, which can be connected to the device through a network. Examples of the above network include but are not limited to the Internet, an intranet, a local area network, a mobile communication network, and a combination thereof.

[0091] The input device 403 can be used to input corresponding digital or character information to the processor 401, and generate key signal inputs related to user settings and function control of the device; the output device 404 can be used to send or display key signal outputs related to user settings and function control of the device.

[0092] The embodiments of the present application also provide a storage medium having computer executable instructions stored therein, which are used to perform related operations in the disconnector cabinet fault detection method provided in any of the embodiments of the present application when executed by a processor.

[0093] Computer-readable storage media includes permanent and non-permanent, removable and non-removable media implemented by any method or technology for information storage. Information can be computer-readable instructions, data structures, program modules or other data. Examples of computer storage media include, but are not limited to, phase-change memory (PRAM), static random access memory (SRAM), dynamic random access memory (DRAM), other types of random access memory (RAM), read-only memory (ROM), electrically erasable programmable read-only memory (EEPROM), flash memory or other memory technologies, compact disc read-only memory (CD-ROM), digital versatile disc (DVD) or other optical storage, magnetic cassette, disk storage or other magnetic storage devices, or any other non-transmission medium that can be used to store information accessible by a computing device.

[0094] It should also be noted that the terms "comprising", "containing", or any other variant thereof are intended to cover a non-exclusive inclusion, such that a process, method, article or apparatus that comprises a list of elements does not include only those elements recited, but can also include other elements not expressly listed or inherent to such process, method, article or apparatus. Without further limitation, an element defined by the statement "comprising a" does not exclude the presence of additional identical elements in the process, method, article or apparatus that includes the element.

[0095] Note that the above are only the preferred embodiments of the present application and the principles of technology used. Those skilled in the art will understand that the present application is not limited to the specific embodiments herein, and that various obvious changes, re-adjustments and substitutions can be made without departing from the scope of the present application. Therefore, although the present application has been described in detail through the above embodiments, the present application is not limited to the above embodiments, and can include more other equivalent embodiments without departing from the concept of the present application, and the scope of the present application is determined by the scope of the appended claims.

Claims

1. A method for detecting faults in an isolation switch cabinet, characterized in that: include: Preprocessing the received multiple modal data respectively to determine the modal features corresponding to each of the modal data; Inputting the modal features into a preset physical information neural network, and determining fault type information and a first probability corresponding to the fault type information based on the physical information neural network; Determining a plurality of similar cases associated with the modal feature in a preset knowledge base, and determining a second probability based on case probabilities corresponding to the plurality of similar cases; Determining a third probability having a mapping relationship with the plurality of modal features based on a mapping relationship between feature values ​​and probability values ​​recorded in the fault mapping table; Based on the first probability, the second probability, and the third probability, a corresponding weighted cumulative sum is determined as a failure probability.

2. The isolation switch cabinet fault detection method according to claim 1, characterized in that: The plurality of modal data are respectively high-frequency current waveform data, infrared images and video images; The preprocessing of the received multiple modal data to determine the modal features corresponding to each of the modal data includes: Based on wavelet transform, denoising is performed on the high-frequency current waveform data, the infrared image and the video image respectively; Determine, based on Fourier transform, a fundamental component and a harmonic component in the denoised high-frequency current waveform data, and determine a first feature based on the fundamental component and the harmonic component; determining a temperature gradient value in the denoised infrared image, and using the temperature gradient value as a second feature; The pixel offset between the denoised video images corresponding to the same target object is determined, and the pixel offset is used as the third feature.

3. The isolating switch cabinet fault detection method according to claim 2, characterized in that: The determining of the first feature according to the fundamental wave component and the harmonic component includes: Determine the arithmetic square root of the harmonic amplitude in each harmonic component; A ratio between the arithmetic square root and the amplitude of the fundamental wave in the fundamental wave component is determined, and the ratio is used as a first feature.

4. The isolating switch cabinet fault detection method according to claim 1, characterized in that: The physical information neural network takes the first physical equation corresponding to power balance, the second physical equation corresponding to thermal balance, and the third physical equation corresponding to contact resistance change as physical constraint equations, and determines the physical information error term in the loss function based on the physical constraint equations. The physical information error term is used to guide the physical information neural network to adjust the model parameters and minimize the physical quantities output by the physical constraint equations in the case of power imbalance, thermal imbalance or abnormal contact resistance change.

5. The isolating switch cabinet fault detection method according to claim 4, characterized in that: The determining, based on the physical information neural network, fault type information and a first probability corresponding to the fault type information includes: Performing an abnormality analysis on the modal feature through the physical information neural network, and determining whether the modal feature is an abnormal feature; In the case where the modal feature is an abnormal feature, searching for a target case matching the abnormal feature in historical cases based on the abnormal feature, determining an initial probability corresponding to the target case, and using the fault type corresponding to the target case as the fault type information; Based on the physical constraint mechanism built into the physical information neural network, the abnormal feature is verified to determine the physical constraint compliance, where the physical constraint compliance is used to indicate the degree of match between the abnormal feature and the physical equation corresponding to the physical constraint mechanism; Determining a confidence level based on the physical constraint compliance, wherein the confidence level is used to indicate an increase in the initial probability due to a degree of matching between the abnormal feature and the physical equation; A probability sum is determined according to the initial probability and the confidence level and is used as the first probability.

6. The isolating switch cabinet fault detection method according to claim 5, characterized in that: In a case where the abnormal feature includes a first feature corresponding to high-frequency current waveform data and a second feature corresponding to an infrared image, the physical constraint mechanism built into the physical information neural network is used to verify the abnormal feature and determine the physical constraint compliance, including: Determine the power loss based on the input power and load power; If a first difference between the power loss and the theoretical power loss value is greater than a preset threshold, determining that the first feature does not match the first physical equation, subtracting a ratio of the corresponding first difference to the maximum power from 1 as a first compliance degree, the first compliance degree being a physical constraint compliance degree corresponding to the first physical equation, and the input power being determined based on the voltage and current values ​​in the first feature; Determining a theoretical temperature rise value based on the current value and the resistance value in the first characteristic; If there is a second difference between the theoretical temperature rise value and the measured temperature rise value in the second feature, determining that the second feature does not match the second physical equation, and subtracting the ratio of the corresponding second difference to the maximum temperature rise from 1 as a second compliance, where the second compliance is a physical constraint compliance of the corresponding second physical equation; Determining a predicted contact resistance based on the temperature value in the second characteristic and a preset oxidation coefficient; When there is a third difference between the predicted contact resistance and the measured contact resistance, it is determined that the first feature does not match the third physical equation, and the ratio of 1 minus the corresponding third difference to the measured contact resistance is taken as the third conformity, and the third conformity is the physical constraint conformity corresponding to the third physical equation.

7. The isolating switch cabinet fault detection method according to claim 1, characterized in that: The knowledge base records case vectors and case probabilities corresponding to each historical case through entity relationship information, wherein the case vector is associated with a feature vector corresponding to the modal data of the historical case; Determining similar cases associated with the modal feature in a preset knowledge base, and determining a second probability based on case probabilities corresponding to a plurality of similar cases, includes: determining a feature vector associated with the modal feature, and determining a similarity between the feature vector and each of the case vectors; Selecting a preset number of historical cases as the similar cases in descending order of similarity; Based on the entity relationship information, the case probabilities corresponding to the similar cases are determined, and a weighted average of the case probabilities corresponding to all the similar cases is determined as the value of the second probability.

8. A method and device for detecting faults in an isolation switch cabinet, characterized in that: include: a feature extraction module configured to pre-process the received multiple modal data respectively and determine the modal features corresponding to each of the modal data; a first probability output module configured to input the modal feature into a preset physical information neural network, and determine fault type information and a first probability corresponding to the fault type information based on the physical information neural network; a second probability output module configured to determine a plurality of similar cases associated with the modal feature in a preset knowledge base, and determine a second probability based on case probabilities corresponding to the plurality of similar cases; a third probability output module configured to determine a third probability having a mapping relationship with the plurality of modal features based on a mapping relationship between feature values ​​and probability values ​​recorded in the fault mapping table; The probability fusion module is configured to determine a corresponding weighted cumulative sum as a fault probability based on the first probability, the second probability and the third probability.

9. An electronic device, characterized in that: include: one or more processors; A storage device for storing one or more programs, wherein when the one or more programs are executed by the one or more processors, the one or more processors implement the isolation switch cabinet fault detection method according to any one of claims 1 to 7.

10. A storage medium storing computer executable instructions, characterized in that: When the computer executable instructions are executed by a processor, they are used to execute the disconnector cabinet fault detection method according to any one of claims 1 to 7.

Images