Secondary system fault positioning method and system based on dual-channel convolutional neural network

Abstract

The invention discloses a secondary system fault positioning method and system based on a dual-channel convolutional neural network. Relates to the technical field of fault location. Historical fault data of the secondary system are obtained, a secondary system fault positioning model is trained based on the historical fault data of the secondary system, and the secondary system fault positioning model is a two-channel convolutional neural network model; monitoring a secondary system alarm signal in real time, and preprocessing the secondary system alarm signal; and inputting the pre-processed secondary system alarm signal into a secondary system fault positioning model to obtain a positioning result. According to the scheme, on the basis of historical fault feature information of the secondary system of the transformer substation, a feature representation method which does not need to consider physical port information of secondary equipment is constructed through pre-processing; then, secondary system fault positioning is realized on the basis of the two-channel convolutional neural network; the whole secondary system fault positioning process does not need to depend on the physical port of the secondary equipment.

Description

Technical Field

[0001] The present invention relates to the technical field of fault location, and in particular to a secondary system fault location method and system based on a dual-channel convolutional neural network. Background Art

[0002] With the continuous advancement of smart grid construction, this comprehensive system including power system and communication network has played a huge role in the green and intelligent transformation of power construction. As the core part of smart grid, smart substation gradually replaces the functions of traditional substation in smart grid by relying on advanced automatic control technology, information technology, computer technology and sensor technology.

[0003] Each device in the secondary system of a traditional substation is connected through hard wiring, and the input and output of the signal correspond one to one. The secondary system of a smart substation follows the IEC61850 standard, which is characterized by the physical connection and information exchange between each intelligent electronic device (IED) through optical fiber and switches, which makes the signal mapping relationship more complicated; and after a networked secondary system fails, a large amount of alarm information will flow into the dispatching center in a short time. For the above reasons, it is difficult to accurately locate the fault location by manual experience after a fault occurs in the secondary system of a smart substation.

[0004] In terms of traditional fault location, there are state scoring and Weibull distribution methods used to evaluate the failure rate of the secondary system, but their effectiveness is affected by the operating environment parameters and equipment history parameters, and their applicability is limited in some smart substations with imperfect hardware conditions. There are also bad data identification and fault diagnosis methods based on secondary system information redundancy, but this method is limited by the influence of residual drowning, and it is inevitable that there will be problems such as missed detection and false detection. There is also a combination of the Skyline query method and the adjacency table, which can determine the fault range in the absence of an alarm signal, but it is difficult to directly give the fault location.

[0005] In existing secondary system fault location research, the constructed fault features usually rely on the physical port information of secondary devices. However, some smart substation configuration files (SCDs) do not fully provide this key information, which limits the application scope of existing methods. Summary of the invention

[0006] The technical problem to be solved by the present invention is: traditional fault location methods usually rely on the physical port information of secondary equipment, but since some SCD files fail to provide this information completely, the universality of traditional methods is limited; the purpose of the present invention is to provide a secondary system fault location method and system based on a dual-channel convolutional neural network, and based on relevant fault feature information, a feature matrix is ​​constructed without considering the physical port information of the secondary equipment; the constructed feature matrix is ​​used as the input of the dual-channel convolutional neural network model, and the fault location result can be obtained without relying on the physical port information of the secondary equipment.

[0007] The present invention is achieved through the following technical solutions: This solution provides a secondary system fault location method based on a dual-channel convolutional neural network, including: Acquire historical fault data of the secondary circuit, and train a secondary system fault location model based on the historical fault data of the secondary system, wherein the secondary system fault location model is a dual-channel convolutional neural network model; Real-time monitoring of secondary system alarm signals generated during the process of substation access to the dispatching control system and the centralized monitoring system, and pre-processing of the secondary system alarm signals; The preprocessed secondary system alarm signal is input into the secondary system fault location model to obtain the fault location result.

[0008] Working principle of this scheme: Traditional fault location methods usually rely on the physical port information of secondary equipment, but since some SCD files fail to provide this information completely, the universality of traditional methods is limited; this scheme is based on the historical fault feature information of the substation secondary system, and after preprocessing, a feature representation method is constructed that does not need to consider the physical port information of the secondary equipment; then, the secondary system fault location is realized based on the dual-channel convolutional neural network; during the entire secondary system fault location process, there is no need to rely on the physical port information of the secondary equipment.

[0009] A further optimization scheme is that before preprocessing the secondary system alarm signal, it also includes: A preset threshold A is set to determine whether the number N of the secondary system alarm signals is greater than or equal to A. If so, the secondary system alarm signals are preprocessed; otherwise, the secondary system alarm signals are tracked and repaired, and the process ends.

[0010] A further optimization scheme is that the acquisition of secondary system historical fault data includes: Divide the intervals of the secondary system; Acquire monitoring signals of secondary equipment in each interval; the secondary equipment includes: a merging unit, an intelligent terminal, a switch, a measurement and control device, and a protection device.

[0011] A further optimization scheme is that the method for dividing the intervals includes: dividing the substation secondary system into multiple intervals according to line intervals, bus intervals and main transformer intervals.

[0012] A further optimization scheme is that the pretreatment includes the following method: Construct an alarm signal matrix based on the monitoring signal M A :

[0013] Code n Indicates suspected faulty device n; N A Indicates the number of alarm signals related to message sending and receiving generated by the suspected faulty device n in the time period T after the fault occurs; N T Indicates the number of abnormal traffic packets detected by the switch in the T period after the fault occurs; P D and P R Indicates suspected faulty equipment n Send and receive power alarm status. If a suspected faulty device generates a send or receive power alarm, P D or P R will be set to 1, otherwise it will be set to 0; S Represents the overall alarm status of the suspected faulty device. When the suspected faulty device generates any alarm signal, S Set to 1, otherwise S Set to 0.

[0014] A further optimization scheme is that the dual-channel convolutional neural network model includes: The first convolutional layer, the second convolutional layer, the first pooling layer, the second pooling layer, the concatenation layer and the fully connected layer; The first convolutional layer is connected to the first pooling layer, and the first pooling layer is connected to the splicing layer; The second convolutional layer is connected to the second pooling layer, and the second pooling layer is connected to the concatenation layer.

[0015] A further optimization scheme is that the first convolution layer includes a first convolution kernel, and the first convolution kernel is used to extract global features; The second convolutional layers each include a second convolutional kernel, and the second convolutional kernel is used to extract local features; The size of the first convolution kernel is larger than that of the second convolution kernel.

[0016] A further optimization scheme is that the first pooling layer includes a first pooling window, and the output features of the first convolution kernel are input into the first pooling window; The second pooling layer includes a second pooling window, and the output features of the second convolution kernel are input into the second pooling window; The size of the first pooling window is larger than the second pooling window.

[0017] A further optimization scheme is that the input dimension of the fully connected layer is N; The input dimension N is determined according to the total number of devices in the secondary system, the first convolution layer parameters, the second convolution layer parameters, the first pooling layer parameters, and the second pooling layer parameters; When the output feature size of the concatenated layer is smaller than the input dimension N, the features output by the first and second pooling layers are padded with 0 before being input into the fully connected layer.

[0018] The present solution also provides a secondary system fault location system based on a dual-channel convolutional neural network, which is used to implement the secondary system fault location method based on a dual-channel convolutional neural network. The system includes: An acquisition module is used to obtain historical fault data of the secondary system, and train a secondary system fault location model based on the historical fault data of the secondary system, wherein the secondary system fault location model is a dual-channel convolutional neural network model; A preprocessing module is used to monitor in real time the secondary system alarm signals generated during the process of substation access to the dispatching control system and the centralized monitoring system, and to preprocess the secondary system alarm signals; The positioning module is used to input the pre-processed secondary system alarm signal into the secondary system fault positioning model to obtain the positioning result.

[0019] Compared with the prior art, the present invention has the following advantages and beneficial effects: The present invention provides a secondary system fault location method and system based on a dual-channel convolutional neural network; this solution is based on the historical fault feature information of the substation secondary system, and constructs a feature representation method that does not need to consider the physical port information of the secondary equipment after preprocessing; then the secondary system fault location is realized on the basis of the dual-channel convolutional neural network; during the entire secondary system fault location process, there is no need to rely on the physical port information of the secondary equipment. BRIEF DESCRIPTION OF THE DRAWINGS

[0020] In order to more clearly illustrate the technical solutions of the exemplary embodiments of the present invention, the following briefly introduces the drawings required for use in the embodiments. It should be understood that the following drawings only illustrate certain embodiments of the present invention and should not be regarded as limiting the scope. For those of ordinary skill in the art, other relevant drawings can be obtained based on these drawings without creative work. In the drawings: Figure 1 It is a flow chart of the secondary system fault location method based on dual-channel convolutional neural network; Figure 2 Schematic diagram of the secondary system fault location process based on dual-channel convolutional neural network; Figure 3 This is a schematic diagram of the dual-channel convolutional neural network model architecture; Figure 4 This is a schematic diagram of the interval division results; Figure 5 This is a schematic diagram of the main body failure of a merging unit in the main transformer bay of Example 3. DETAILED DESCRIPTION

[0021] In order to make the objectives, technical solutions and advantages of the present invention more clearly understood, the present invention is further described in detail below in conjunction with embodiments and drawings. The exemplary embodiments of the present invention and their description are only used to explain the present invention and are not intended to limit the present invention.

[0022] The secondary system of smart substation plays a key role in maintaining the stable operation of the power system. Therefore, it is very important to accurately locate the fault of the secondary system in real time. Traditional fault location methods usually rely on the physical port information of the secondary equipment, but because some SCD files fail to provide this information completely, the universality of traditional methods is limited. In order to solve this problem, the present invention provides the following embodiments: Example 1 This embodiment provides a secondary system fault location method based on a dual-channel convolutional neural network, such as Figure 1 and Figure 2 As shown, including: Step 1: Acquire historical fault data of the secondary system, and train a secondary system fault location model based on the historical fault data of the secondary system, wherein the secondary system fault location model is a dual-channel convolutional neural network model; The obtaining of historical fault data of the secondary system includes: Divide the intervals of the secondary system; Acquire monitoring signals of secondary equipment in each interval; the secondary equipment includes: a merging unit, an intelligent terminal, a switch, a measurement and control device, and a protection device.

[0023] The methods for dividing the interval include: Figure 4 As shown, the substation secondary system is divided into multiple bays according to the line bay, bus bay and main transformer bay.

[0024] The secondary equipment of smart substations includes merging units (MU), measurement and control devices (M&C), intelligent terminals (IT), and protection devices (P). These secondary devices use optical fibers and switches (SW) as carriers to realize message interaction between sampled values ​​SV and logical information GOOSE. The online monitoring system of smart substations based on IEC61850 can collect various information including SV and GOOSE. After a secondary system fault occurs, the message receiving status, message flow size, and operating status of each secondary device may change. These changes can serve as important data support for locating secondary system faults.

[0025] When a switch in the line interval area fails, all other secondary devices that receive messages sent by the switch generate alarm signals. It can be seen that the scope of the alarm signal is not limited to other devices directly connected to the faulty device through optical fiber. This shows that in a highly networked secondary system, once a device fails, all devices associated with it may be affected and trigger a fault alarm. In addition, the alarm signals generated by the faulty device and other secondary devices associated with it will flood into the online monitoring system in a short time. At this time, it is difficult to accurately locate the fault by manual experience analysis alone.

[0026] According to the protection sampling and tripping mode, the secondary system of the smart substation can generally be divided into four networking modes: "direct sampling and direct tripping", "network sampling and direct tripping", "direct sampling and network tripping", and "network sampling and network tripping". Since the "direct sampling and direct tripping" networking method significantly reduces the risk of communication message loss by reducing the use of relay equipment, this feature makes it gradually become the dominant choice in the construction planning of the secondary system of the smart substation. Based on the above reasons, this embodiment is based on the direct sampling and direct tripping secondary system data of a 220kV smart substation used for fault location research in the early stage, which is used to train and verify the fault location method proposed in the present invention; the secondary system consists of a busbar interval, two line intervals and two main transformer intervals. The secondary equipment between different intervals completes information exchange through the central switch.

[0027] Step 2: real-time monitoring of secondary system alarm signals generated during the process of substation access to the dispatching control system and the centralized monitoring system, and pre-processing of the secondary system alarm signals; Before preprocessing the secondary system alarm signal, the method further includes: A preset threshold A is set to determine whether the number N of the secondary system alarm signals is greater than or equal to A. If so, the secondary system alarm signals are preprocessed; otherwise, the secondary system alarm signals are tracked and repaired, and the process ends.

[0028] In order to prevent the wrong alarm signals from being input into the secondary system fault location model during secondary system operation, this paper sets a trigger condition for the fault location process. When the number of alarm signals N detected by the background is greater than the threshold A, the fault location process will be triggered. Otherwise, the alarm source device will be tracked and repaired. After the fault location process is successfully triggered, the relevant alarm information will be used to construct the alarm signal matrix M A , and input it into the dual-channel convolutional neural network model. Finally, the model will output the results and inform the operation and maintenance personnel.

[0029] The pretreatment comprises the following methods: Construct an alarm signal matrix based on the monitoring signal M A :

[0030] Code n Indicates suspected faulty device n; N A Indicates the number of alarm signals related to message sending and receiving generated by the suspected faulty device n in the time period T after the fault occurs; N T Indicates the number of abnormal traffic packets detected by the switch in the T period after the fault occurs; P D and P R Indicates suspected faulty equipment n Send and receive power alarm status. If a suspected faulty device generates a send or receive power alarm, P D or P R will be set to 1, otherwise it will be set to 0; S Represents the overall alarm status of the suspected faulty device. When the suspected faulty device generates any alarm signal, S Set to 1, otherwise S Set to 0.

[0031] The existing methods rely on the physical port information of the secondary equipment when constructing fault features. Although the IEC 61850 protocol clearly stipulates the description of the physical port of the device in the SCD file, there are still some smart substations whose SCD files do not fully provide this information, which limits the application scope of the existing methods. In response to the above problems, this solution proposes a feature extraction method that does not need to consider the physical port information of the secondary equipment. The alarm signal matrix constructed in this embodiment M A , used to indicate the characteristics of the alarm signal generated by the suspected faulty device. The suspected faulty device refers to the secondary device that generates the alarm signal.

[0032] The suspected faulty device numbers in the fault sample set constructed in this paper are shown in Table 1: Table 1 Number of suspected faulty equipment

[0033] Step 3: Input the preprocessed secondary system alarm signal into the secondary system fault location model to obtain the location result.

[0034] Since convolutional neural networks have advantages such as local connection, parameter sharing and spatial invariance, they often perform well in two-dimensional data classification tasks. Therefore, this solution provides a dual-channel convolutional neural network model for secondary system fault location, such as Figure 3 As shown, the dual-channel convolutional neural network model includes: The first convolution layer, the second convolution layer, the first pooling layer, the second pooling layer, the splicing layer and the fully connected layer; the first convolution layer is connected to the first pooling layer, the first pooling layer is connected to the splicing layer; the second convolution layer is connected to the second pooling layer, and the second pooling layer is connected to the splicing layer. The first convolution layer includes a first convolution kernel, which is used to extract global features; the second convolution layer includes a second convolution kernel, which is used to extract local features; the size of the first convolution kernel is larger than that of the second convolution kernel. The first pooling layer includes a first pooling window, and the output features of the first convolution kernel are input into the first pooling window; the second pooling layer includes a second pooling window, and the output features of the second convolution kernel are input into the second pooling window; the size of the first pooling window is larger than that of the second pooling window.

[0035] The input dimension of the fully connected layer is N; the input dimension N is determined according to the total number of devices in the secondary system, the parameters of the first convolutional layer, the parameters of the second convolutional layer, the parameters of the first pooling layer, and the parameters of the second pooling layer; when the output feature size of the concatenation layer is smaller than the input dimension N, the features output by the first pooling layer and the second pooling layer are padded with 0 before being input into the fully connected layer.

[0036] Alarm signal matrix M A The features are used as input 1 and input 2 and are input into different convolutional layers and pooling layers respectively. After being processed by the pooling layer, the output features are linearly expanded and concatenated in the concatenation layer, and finally the output result is obtained by passing through the fully connected layer. In order to make the model better adapt to input data of different dimensions, this paper uses convolution kernels of different sizes in the convolution layer of the dual-channel convolutional neural network. Small-sized convolution kernels are more concerned with the extraction of local features, while large-sized convolution kernels are more concerned with the extraction of global features. The sizes of the two convolution kernels are 4×4 and 3×3 respectively.

[0037] In order to make the model better adapt to convolutional features of different dimensions, this solution designs pooling windows of different sizes to pool convolutional features. The sizes of the two pooling windows are 4×2 and 4×1. The output features of the large convolution kernel are input to the large pooling window, while the output features of the small convolution kernel are input to the small pooling window. In addition, all pooling layers use the maximum pooling method.

[0038] Since the input dimension of the fully connected layer is a fixed value, this paper will calculate the value based on the total number of devices in the secondary system of the smart substation, the parameters of the convolution layer and the pooling layer. When the output feature size of the concatenation layer is smaller than the input dimension of the fully connected layer, the features output by the pooling layer will be padded with 0 before being input to the fully connected layer.

[0039] In existing secondary system fault location research, the constructed fault features usually rely on the physical port information of the secondary equipment. However, this key information is not fully provided in some smart substation configuration files (SCD), which limits the scope of application of existing methods. In response to the above problems in secondary system fault location, this embodiment proposes a secondary system fault location method based on a dual-channel convolutional neural network. First, based on the secondary system fault feature information of a 220kV smart substation, a feature representation method that does not need to consider the physical port information of the secondary equipment is proposed. Then, based on the dual-channel convolutional neural network, this embodiment designs a secondary system fault location process.

[0040] Example 2 This embodiment provides a secondary system fault location system based on a dual-channel convolutional neural network, which is used to implement the secondary system fault location method based on a dual-channel convolutional neural network in Example 1. The system includes: An acquisition module is used to obtain historical fault data of the secondary system, and train a secondary system fault location model based on the historical fault data of the secondary system, wherein the secondary system fault location model is a dual-channel convolutional neural network model; A preprocessing module is used to monitor in real time the secondary system alarm signals generated during the process of substation access to the dispatching control system and the centralized monitoring system, and to preprocess the secondary system alarm signals; The positioning module is used to input the pre-processed secondary system alarm signal into the secondary system fault positioning model to obtain the positioning result.

[0041] Example 3 This embodiment provides a computer-readable medium on which a computer program is stored. The computer program is executed by a processor to implement the secondary system fault location method based on a dual-channel convolutional neural network as in Embodiment 1. The total number of sample classifications of the fault sample set used in this embodiment is 13, and the total number of samples is 1620. As shown in Table 2, the fault sample set used in this embodiment is divided into two major fault categories, optical fiber fault and body fault. The major optical fiber fault category contains 8 minor fault categories, and the major body fault category contains 5 minor fault categories.

[0042] Table 2 Number of fault sample sets

[0043] In order to find the optimal parameters, the fault sample set was input into the model for training and testing multiple times. During the experiment, the multi-category cross entropy loss function was used to evaluate the model performance. The model is based on the Pytorch framework, and the training platform uses the RTX3050TI GPU. Finally, the parameter settings of the dual-channel convolutional neural network model in this embodiment are shown in Table 3: Table 3 Optimal hyperparameters of the model

[0044] It has been verified that the fault location accuracy of the dual-channel convolutional neural network model of this embodiment is 98.35%; as shown in Table 4, the dual-channel convolutional neural network model proposed in this embodiment has achieved good location accuracy in all fault subcategories. This is due to the convolution kernels and pooling windows of different sizes.

[0045] Table 4 Fault location accuracy

[0046] However, for the same fault subclass with different data dimensions, such as MU-SW, convolution kernels and pooling windows of different sizes may amplify the impact of noise and information loss in the model learning process to a certain extent, limiting the learning effect of the model on this fault subclass. Ultimately, the fault location accuracy of the dual-channel convolutional neural network model in MU-SW is only 88.64%.

[0047] After a fault occurs in the secondary system of an actual smart substation, the alarm signal generated by the equipment will be subject to various interferences, resulting in information distortion and other problems. Therefore, in order to verify the robustness of the proposed model under interference conditions, this embodiment sets interference on the test set, randomly selects half of the samples from the test set to input interference information, that is, the feature information represented by "0" or "1" in the alarm signal matrix of these samples, such as power alarm P and total alarm S, is reversed by artificial methods, and the number of changed feature information is limited to 2.

[0048] Table 5 Robustness test

[0049] As shown in Table 5, after the fault sample set is interfered with by the above method, the average fault location accuracy of the dual-channel convolutional neural network model of this scheme is 98.14% for the fault sample set under interference. The test result of the fault sample set without interference only drops by 0.21%, and the robustness of the model proposed in this paper is verified.

[0050] In order to further illustrate the model's location process, this embodiment selects a case from the fault sample set for analysis and explanation. Figure 5 As shown in the figure, the merging unit of the main transformer interval cannot normally complete the sending and receiving functions of GOOSE messages and SV messages due to physical damage to the internal hardware, thus generating an SV alarm and a general device alarm signal. Subsequently, other devices associated with the faulty merging unit also generate related message alarm signals and general device alarm signals because they cannot normally receive SV messages from the merging unit. The alarm signals generated by the above-mentioned suspected faulty devices and the number of abnormal traffic messages monitored by the switch are counted, and finally a good alarm signal matrix is ​​constructed. M A As shown below:

[0051] The above constructed M A Sent to the dual-channel convolutional neural network model, when the output threshold is set to 0.9, the output result of the model is:

[0052] According to the above formula, the output of the dual-channel convolutional neural network model is MU, which means that the model correctly determines that the fault is a merging unit fault. After the dual-channel convolutional neural network model outputs the fault type, the operation and maintenance personnel can use the network analyzer to accurately locate the fault based on this result.

[0053] In view of the low accuracy of traditional fault location methods and the problem that feature construction depends on the physical port information of secondary devices, this solution designs an alarm signal matrix based on the fault feature information generated by secondary devices. M A Subsequently, a secondary equipment fault location process was proposed based on the construction of a dual-channel convolutional neural network. Finally, the alarm signal matrix M A Input into the Dual-channel CNN model to realize secondary system fault location, which can achieve better fault location effect.

[0054] The specific implementation methods described above further illustrate the objectives, technical solutions and beneficial effects of the present invention in detail. It should be understood that the above description is only a specific implementation method of the present invention and is not intended to limit the scope of protection of the present invention. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principles of the present invention should be included in the scope of protection of the present invention.

Claims

1. A secondary system fault location method based on a dual-channel convolutional neural network, characterized in that: include: Acquire historical fault data of the secondary circuit, and train a secondary system fault location model based on the historical fault data of the secondary system, wherein the secondary system fault location model is a dual-channel convolutional neural network model; Real-time monitoring of secondary system alarm signals generated during the process of substation access to the dispatching control system and the centralized monitoring system, and pre-processing of the secondary system alarm signals; The preprocessed secondary system alarm signal is input into the secondary system fault location model to obtain the fault location result.

2. The secondary system fault location method based on dual-channel convolutional neural network according to claim 1 is characterized in that: Before preprocessing the secondary system alarm signal, the method further includes: A preset threshold A is set to determine whether the number N of the secondary system alarm signals is greater than or equal to A. If so, the secondary system alarm signals are preprocessed; otherwise, the secondary system alarm signals are tracked and repaired, and the process ends.

3. The secondary system fault location method based on a dual-channel convolutional neural network according to claim 1 is characterized in that: The obtaining of historical fault data of the secondary system includes: Divide the intervals of the secondary system; Acquire monitoring signals of secondary equipment in each interval; the secondary equipment includes: a merging unit, an intelligent terminal, a switch, a measurement and control device, and a protection device.

4. The secondary system fault location method based on a dual-channel convolutional neural network according to claim 3 is characterized in that: The method for dividing the bay intervals includes: dividing the substation secondary system into a plurality of bay intervals according to the line interval, bus interval and main transformer interval.

5. The secondary system fault location method based on dual-channel convolutional neural network according to claim 1 is characterized in that: The pretreatment comprises the following methods: Construct an alarm signal matrix based on the monitoring signal M A : Code n Indicates suspected faulty device n; N A Indicates the number of alarm signals related to message sending and receiving generated by the suspected faulty device n in the time period T after the fault occurs; N T Indicates the number of abnormal traffic packets detected by the switch in the T period after the fault occurs; P D and P R Indicates suspected faulty equipment n Send and receive power alarm status. If a suspected faulty device generates a send or receive power alarm, P D or P R will be set to 1, otherwise it will be set to 0; S Represents the overall alarm status of the suspected faulty device. When the suspected faulty device generates any alarm signal, S Set to 1, otherwise S Set to 0.

6. The secondary system fault location method based on dual-channel convolutional neural network according to claim 1 is characterized in that: The dual-channel convolutional neural network model includes: The first convolutional layer, the second convolutional layer, the first pooling layer, the second pooling layer, the concatenation layer and the fully connected layer; The first convolutional layer is connected to the first pooling layer, and the first pooling layer is connected to the splicing layer; The second convolutional layer is connected to the second pooling layer, and the second pooling layer is connected to the concatenation layer.

7. The secondary system fault location method based on dual-channel convolutional neural network according to claim 6 is characterized in that: The first convolution layer includes a first convolution kernel, and the first convolution kernel is used to extract global features; The second convolutional layers each include a second convolutional kernel, and the second convolutional kernel is used to extract local features; The size of the first convolution kernel is larger than that of the second convolution kernel.

8. The secondary system fault location method based on dual-channel convolutional neural network according to claim 7 is characterized in that: The first pooling layer includes a first pooling window, and the output features of the first convolution kernel are input into the first pooling window; The second pooling layer includes a second pooling window, and the output features of the second convolution kernel are input into the second pooling window; The size of the first pooling window is larger than the second pooling window.

9. The secondary system fault location method based on dual-channel convolutional neural network according to claim 6, characterized in that: The input dimension of the fully connected layer is N; The input dimension N is determined according to the total number of devices in the secondary system, the first convolution layer parameters, the second convolution layer parameters, the first pooling layer parameters, and the second pooling layer parameters; When the output feature size of the concatenated layer is smaller than the input dimension N, the features output by the first and second pooling layers are padded with 0 before being input into the fully connected layer.

10. A secondary system fault location system based on a dual-channel convolutional neural network, used to implement the secondary system fault location method based on a dual-channel convolutional neural network according to any one of claims 1 to 9, characterized in that: The system comprises: An acquisition module is used to obtain historical fault data of the secondary system, and train a secondary system fault location model based on the historical fault data of the secondary system, wherein the secondary system fault location model is a dual-channel convolutional neural network model; A preprocessing module is used to monitor in real time the secondary system alarm signals generated during the process of substation access to the dispatching control system and the centralized monitoring system, and to preprocess the secondary system alarm signals; The positioning module is used to input the pre-processed secondary system alarm signal into the secondary system fault positioning model to obtain the positioning result.

Images