A monitoring method, device, equipment and storage medium for a transformer substation protection pressure plate

Abstract

The present application discloses a monitoring method, device, equipment and storage medium for a transformer substation protection pressure plate, and relates to the field of electrical digital data processing technology. The present application adopts dual monitoring means of electrical monitoring and visual monitoring to ensure more accurate monitoring results. By monitoring the current throw-in / out state and visual throw-in / out state of each connection piece respectively, and verifying whether the connection piece is in the same state of the current throw-in / out state and the visual throw-in / out state at the same time, it is determined whether the connection piece is abnormal, and the connection piece whose current throw-in / out state does not match the visual throw-in / out state is defined as an abnormal connection piece. Compared with the prior art, the present application does not require manual participation in the monitoring process, thereby eliminating subjective errors and reducing labor costs, and the dual judgment mechanism can further improve the monitoring accuracy.

Description

Technical Field

[0001] The present application relates to the technical field of electrical digital data processing, and in particular to a monitoring method, device, equipment and storage medium for a transformer substation protection pressure plate. Background Art

[0002] Substation protection pressure plates are an important part of the power system. They play a key role in ensuring the safe and stable operation of the power system. Protection pressure plates, also known as protection links, are the bridge and link between the protection device and the external wiring, and are directly related to whether the protection function and action output can function normally. With the promotion and application of centralized monitoring and unmanned operation mode of substations, centralized data information transmission and unified management have become an inevitable trend of development.

[0003] In the power system, since the protection equipment is relatively scattered and there are many pressure plates that need to be protected, when operating in a harsh environment, especially in coastal or industrial areas with poor air quality, such as salt and other corrosive substances in the air, once the air is humid on rainy days, many protection pressure plates will be seriously corroded, and the pressure plate function will easily fail. There are many deficiencies in the inspection of protection pressure plates in many places, and there is no strict and effective inspection system or specific indicators. Therefore, even if there are fatal defects in the protection pressure plate circuit, they cannot be discovered in time.

[0004] There is an inner gasket between the connecting ends of the pressure plate. After relevant experiments, it was found that this pressure plate connection has a direct impact on the resistance between the pressure plates. The tripping of the protection pressure plate is due to the fact that when the protection pressure plate is corroded, the connection between the pressure plates will also be corroded. The degree and distribution of corrosion of the connection will affect the resistance of the contact part. Under normal circumstances, the higher the degree of corrosion of the contact part of the connection and the larger the area covered by corrosion, the greater the contact resistance. When the protection trips after the electrical appliance is started, the circuit cannot be conducted and the tripping coil cannot be energized, causing the connection piece to refuse to move. When inspecting the pressure plate, the relevant inspectors generally only check whether the pressure plate can meet the use requirements at the time when checking its status. No specific inspection is carried out on the connection of the pressure plate and whether the circuit control is intact, which leads to the failure to detect defects in the circuit in time.

[0005] For the above problems, the commonly used monitoring measures are (1) Appearance inspection: The appearance inspection of the protective pressure plate is mainly to check whether it is connected and the corrosion condition. When inspectors are inspecting the pressure plate, they should maintain a sense of responsibility for their work, observe carefully while working, and comprehensively inspect the appearance of the pressure plate. Not only should they check the connecting piece of the pressure plate, but also check whether there is corrosion on the surface of the metal structure such as the port. If corrosion and rust occur, the inspectors should record and report to the relevant departments in a timely manner to ensure that the corrosion problem of the pressure plate device can be solved in time. (2) Multimeter measurement: When the pressure plate is put into operation, use a multimeter to measure the DC voltage between the two ends of the pressure plate opening and the ground to ensure that there is no "one positive and one negative" voltage at both ends of the pressure plate opening before it can be put into operation.

[0006] The above inspection method requires manual inspection of each connected piece one by one, which is time-consuming and laborious, and there are obvious subjective errors if it only relies on observing the appearance. At the same time, errors are prone to occur when measuring with a multimeter: when the staff uses a multimeter to measure the pressure plate of the stabilizing device, the insulation of the DC voltage range of the multimeter is broken down, causing the pressure plate to short-circuit to the ground, and the interference of factors such as distributed capacitance causes the line switch in operation to trip; when the staff uses a multimeter to measure the pressure plate of the main transformer protection screen, the resistance range of the multimeter is selected (the DC voltage range should be selected), causing the pressure plate to short-circuit to the ground through the resistance range of the multimeter, causing the main transformer low 501 switch to trip and large-scale power outages. Summary of the invention

[0007] The main purpose of this application is to provide a monitoring method, device, equipment and storage medium for a substation protection pressure plate, so as to solve the problems in the prior art that the status monitoring of the substation protection pressure plate is time-consuming and labor-intensive, has subjective errors, has a high probability of misoperation, and has inaccurate results.

[0008] In order to achieve the above objectives, this application provides the following technical solutions:

[0009] A monitoring method for a transformer substation protection pressure plate, the transformer substation protection pressure plate comprises a plate surface fixed to an external plane, a plurality of rotating contacts penetrating the plate surface, a plurality of fixed contacts penetrating the plate surface, each fixed contact matches a rotating contact and is spaced a preset distance from the current rotating contact, each rotating contact is rotatably connected to a connecting piece, the current connecting piece is connected or disconnected with the current fixed contact by rotation, a plurality of current detection pieces are also installed on a side of the plate surface away from all the connecting pieces, each current detection piece matches a fixed contact and is electrically connected to the matching fixed contact, the monitoring method comprises:

[0010] Step S1, obtaining current information of each fixed contact through each current detection element at intervals of a preset time period;

[0011] Step S2, defining the current information when the current value reaches a preset threshold as a current input state, and defining the current information when the current value is zero as a current exit state;

[0012] Step S3, defining the current state classification of all connected pieces, wherein the current state classification includes the current input state and the current exit state;

[0013] Step S4, defining the visual state classification of the connected piece, wherein the visual state classification includes a visual engagement state, a visual exit state, and an unclassified state;

[0014] Step S5, acquiring image information of each connected piece by using a target detection algorithm at intervals of the preset time period;

[0015] Step S6, obtaining the visual boundary of each connected piece by using a sliding window algorithm;

[0016] Step S7, classifying all visual boundaries into the visual state classification by a classification algorithm;

[0017] Step S8, traversing the current throw-in / out state and visual throw-in / out state corresponding to each link, and defining the link whose current throw-in / out state does not match the visual throw-in / out state as an abnormal link;

[0018] Step S9, defining the connected pieces classified into the unclassified state as the abnormal connected pieces, and obtaining the position information of all abnormal connected pieces based on the board surface to send to the external receiving end.

[0019] As a further improvement of the present application, step S5, obtaining image information of each contiguous piece by using a target detection algorithm at intervals of the preset time period, includes:

[0020] Step S51, photographing all the connected pieces on the board surface by a preset photographing end based on an external fixed position, and obtaining a visual image based on all the connected pieces each time photographing;

[0021] Step S52, dividing the visual image into a preset number of grids on average, wherein the size of each grid is smaller than the size of each connected piece;

[0022] Step S53, detecting the connected pieces in the visual image based on all grids according to the target detection algorithm and obtaining a plurality of bounding boxes;

[0023] Step S54, respectively obtaining the confidence of each bounding box, and marking the bounding box with the highest confidence among all bounding boxes as a first-order bounding box;

[0024] Step S55, calculating the intersection-and-union ratio of each other bounding box with the first-order bounding box, and selecting a bounding box whose intersection-and-union ratio is greater than or equal to a preset threshold as a second-order bounding box;

[0025] Step S56, obtaining the second-order bounding box with the highest confidence among all the second-order bounding boxes and using it as the third-order bounding box;

[0026] Step S57, defining the third-order bounding box as all connected image information;

[0027] Step S58, evenly dividing all the connected pieces of image information to form a number of image units, each image unit having only one connected piece;

[0028] Step S59, defining each image unit as each contiguous piece of image information.

[0029] As a further improvement of the present application, step S6, respectively obtaining the visual boundary of each connected piece by a sliding window algorithm, includes:

[0030] Step S61, binarizing each piece of image information to obtain a black and white visual image;

[0031] Step S62, dividing the black and white visual image into grids of a preset density;

[0032] Step S63, respectively obtaining the color of each grid, and assigning a value of 0 to a grid with a black color, and assigning a value of 1 to a grid with a white color;

[0033] Step S64, defining each grid and eight adjacent grids as a 3×3 window;

[0034] Step S65, traversing all grids through the 3×3 window with one grid as one step;

[0035] Step S66, based on all traversal results, delete the grids whose values ​​are all 1 and the grids whose values ​​are all 0 in the 3×3 window;

[0036] Step S67, obtaining the retained grid and defining it as the visual boundary of the current connected piece.

[0037] As a further improvement of the present application, step S7, classifying all visual boundaries into the visual state classification by a classification algorithm, includes:

[0038] Step S71: define a set of data to be classified according to all visual boundaries ,in The data set to be classified The A visual boundary, is the number of all visual boundaries;

[0039] Step S72: defining a category set according to the visual state classification ,in, For the category set The Visual state classification, The number of categories for all visual states;

[0040] Step S73, calculate the conditional probability of each visual boundary under each visual state classification according to formula (1):

[0041] (1);

[0042] in, For the Under the visual state classification, the data set to be classified The conditional probability of For the The marginal probability of each visual state classification; For the Under the visual state classification, The conditional probability of a visual boundary;

[0043] Step S74, classify each visual boundary into the visual state classification with the highest conditional probability.

[0044] As a further improvement of the present application, in step S9, the connected pieces classified into the unclassified state are defined as the abnormal connected pieces, and the position information of all abnormal connected pieces based on the board surface is obtained to be sent to the external receiving end, and then, the following is included:

[0045] Step S10, dividing the current switching states of all preset time periods of the current contiguous area into a first training set and a first sample set according to a preset ratio;

[0046] Step S20, dividing the visual projection and retreat states of all preset time periods of the current continuous film into a second training set and a second sample set according to the preset ratio;

[0047] Step S30, inputting the first training set into a preset neural network model and performing training to obtain a first prediction model, and inputting the second training set into the preset neural network model and performing training to obtain a second prediction model;

[0048] Step S40, inputting the first sample set into the first prediction model to obtain a first prediction value, and inputting the second sample set into the second prediction model to obtain a second prediction value;

[0049] Step S50, obtaining a mismatch ratio between the second predicted value and the first predicted value;

[0050] Step S60, defining the mismatch ratio as the failure rate of the current connected piece;

[0051] Step S70, obtaining the failure rates of all connected slices and sending them to an external receiving end.

[0052] As a further improvement of the present application, step S30, inputting the first training set into a preset neural network model and training it to obtain a first prediction model, and inputting the second training set into the preset neural network model and training it to obtain a second prediction model, includes:

[0053] Step S301, defining a topological relationship of the preset neural network model, wherein the topological relationship includes an input layer, a hidden layer, and an output layer that are sequentially signal-connected;

[0054] Step S302, inputting the first training set into the input layer, training the first training set for a first preset number of times by using the LM estimation method, and obtaining a first root mean square error between the first sample set and the current training result based on each training;

[0055] Step S303, obtaining a first minimum value among all first root mean square errors, and obtaining a training result corresponding to the first minimum value as the first prediction model;

[0056] Step S304, inputting the second training set into the input layer, training the second training set for a second preset number of times by using the LM estimation method, and obtaining a second root mean square error between the second sample set and the current training result based on each training;

[0057] Step S305, obtaining the second minimum value among all second root mean square errors, and obtaining the training result corresponding to the second minimum value as the second prediction model.

[0058] As a further improvement of the present application, the preset neural network model is represented by formula (2):

[0059] (2);

[0060] in, The preset neural network model; is the input layer and , is the first input nodes, each input node corresponds to one of the first training sets or one of the second training sets, is the first The input nodes to the hidden layer The weights of the input nodes; is the first The bias of the input node; is the transfer function, and .

[0061] In order to achieve the above objectives, this application also provides the following technical solutions:

[0062] A monitoring device for a substation protection pressure plate, the monitoring device for the substation protection pressure plate is applied to the above-mentioned monitoring method for the substation protection pressure plate, and the monitoring device for the substation protection pressure plate comprises:

[0063] A contact current information acquisition module, used to acquire the current information of each fixed contact through each current detection element at intervals of a preset time period;

[0064] A current input / output state definition module, used to define the current information when the current value reaches a preset threshold as the current input state, and define the current information when the current value is zero as the current output state;

[0065] A current state classification definition module, used to define the current state classification of all connected pieces, wherein the current state classification includes the current input state and the current exit state;

[0066] A visual state classification definition module, used to define the visual state classification of the contiguous piece, wherein the visual state classification includes a visual engagement state, a visual exit state, and an unclassified state;

[0067] A contiguous image information acquisition module, used to acquire image information of each contiguous piece by using a target detection algorithm at intervals of the preset time period;

[0068] A connected visual boundary acquisition module is used to obtain the visual boundary of each connected piece through a sliding window algorithm;

[0069] A contiguous visual boundary classification module, used for classifying all visual boundaries into the visual state classification by a classification algorithm;

[0070] A dual determination module for the throw-in / out status is used to traverse the current throw-in / out status and the visual throw-in / out status corresponding to each connection, and define the connection whose current throw-in / out status does not match the visual throw-in / out status as an abnormal connection;

[0071] The abnormal contiguous piece definition sending module is used to define the contiguous piece classified into the unclassified state as the abnormal contiguous piece, and obtain the position information of all abnormal contiguous pieces based on the board surface to send to the external receiving end.

[0072] In order to achieve the above objectives, this application also provides the following technical solutions:

[0073] An electronic device includes a processor and a memory coupled to the processor, wherein the memory stores program instructions executable by the processor; when the processor executes the program instructions stored in the memory, the monitoring method for the substation protection pressure plate as described above is implemented.

[0074] In order to achieve the above objectives, this application also provides the following technical solutions:

[0075] A storage medium stores program instructions, and when the program instructions are executed by a processor, the monitoring method of the transformer substation protection pressure plate as described above can be implemented.

[0076] The present application obtains the current information of each fixed contact through each current detection component at intervals of preset time periods; defines the current information when the current value reaches a preset threshold value as the current input state, and defines the current information when the current value is zero as the current exit state; defines the current state classification of all connected pieces, and the current state classification includes the current input state and the current exit state; defines the visual state classification of the connected pieces, and the visual state classification includes the visual input state, the visual exit state, and the unclassified state; obtains the image information of each connected piece through the target detection algorithm at intervals of preset time periods; obtains the visual boundary of each connected piece through the sliding window algorithm; classifies all visual boundaries into the visual state classification through the classification algorithm; traverses the current input / output state and visual input / output state corresponding to each connected piece, and defines the connected piece whose current input / output state does not match the visual input / output state as an abnormal connected piece; defines the connected piece classified into the unclassified state as an abnormal connected piece, and obtains the position information of all abnormal connected pieces based on the board surface to send to the external receiving end. This application adopts the dual monitoring means of electrical monitoring and visual monitoring to ensure more accurate monitoring results. By monitoring the current throw-in and throw-out states and visual throw-in and throw-out states of each link respectively, and verifying whether the link is in the same state in the current throw-in and throw-out states and the visual throw-in and throw-out states at the same time (the same link is in the throw-in state in the current throw-in and throw-out state, and is in the throw-in state in the visual throw-in and throw-out state at the same time; or the same link is in the throw-out state in the current throw-in and throw-out state, and is in the throw-out state in the visual throw-in and throw-out state at the same time), it is determined whether the link is abnormal, and the link whose current throw-in and throw-out states do not match the visual throw-in and throw-out states is defined as an abnormal link. Compared with the prior art, this application does not require manual participation in the monitoring process, thereby eliminating subjective errors and reducing labor costs, and the dual judgment mechanism can further improve the monitoring accuracy. BRIEF DESCRIPTION OF THE DRAWINGS

[0077] Figure 1 A schematic diagram of the structure of a substation protection pressure plate according to an embodiment of a monitoring method for a substation protection pressure plate of the present application;

[0078] Figure 2A schematic diagram of the process steps of an embodiment of a monitoring method for a substation protection pressure plate of the present application;

[0079] Figure 3 A functional module diagram of an embodiment of a monitoring device for a protective pressure plate of a substation of the present application;

[0080] Figure 4 This is a schematic diagram of the structure of an embodiment of the electronic device of the present application;

[0081] Figure 5 This is a schematic diagram of the structure of an embodiment of the storage medium of the present application. DETAILED DESCRIPTION

[0082] The following will be combined with the drawings in the embodiments of the present application to clearly and completely describe the technical solutions in the embodiments of the present application. Obviously, the described embodiments are only part of the embodiments of the present application, not all of the embodiments. Based on the embodiments in the present application, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of this application.

[0083] The terms "first", "second" and "third" in this application are only used for descriptive purposes and cannot be understood as indicating or implying relative importance or implicitly indicating the number of indicated technical features. Thus, the features defined as "first", "second" and "third" may explicitly or implicitly include at least one of the features. In the description of this application, the meaning of "multiple" is at least two, such as two, three, etc., unless otherwise clearly and specifically defined. All directional indications in the embodiments of this application, such as up, down, left, right, front, back..., are only used to explain the relative position relationship, movement, etc. between the components in a certain specific posture, as shown in the accompanying drawings. If the specific posture changes, the directional indication also changes accordingly. In addition, the terms "including" and "having" and any of their variations are intended to cover non-exclusive inclusions. For example, a process, method, system, product or device that includes a series of steps or units is not limited to the listed steps or units, but optionally also includes steps or units that are not listed, or optionally also includes other steps or units inherent to these processes, methods, products or devices.

[0084] Reference to "embodiments" herein means that a particular feature, structure, or characteristic described in conjunction with the embodiments may be included in at least one embodiment of the present application. The appearance of the phrase in various locations in the specification does not necessarily refer to the same embodiment, nor is it an independent or alternative embodiment that is mutually exclusive with other embodiments. It is explicitly and implicitly understood by those skilled in the art that the embodiments described herein may be combined with other embodiments.

[0085] This embodiment provides an embodiment of a monitoring method for a transformer substation protection pressure plate. In this embodiment, see Figure 1 , Figure 1 It is a common style of substation protection pressure plate. The substation protection pressure plate includes a plate surface fixed to an external plane, a plurality of rotating contacts penetrating the plate surface, and a plurality of fixed contacts penetrating the plate surface. Each fixed contact matches a rotating contact and is spaced a preset distance from the current rotating contact. Each rotating contact is connected to a connecting piece by rotation. The current connecting piece is connected or disconnected with the current fixed contact by rotation. A plurality of current detection components are also installed on the side of the plate surface away from all the connecting pieces. Each current detection component matches a fixed contact and is electrically connected to the matching fixed contact.

[0086] like Figure 2 As shown, the monitoring method of the substation protection pressure plate of this embodiment includes the following steps:

[0087] Step S1, obtaining current information of each fixed contact through each current detection element at preset time intervals.

[0088] Preferably, the preset time period can be set according to actual needs, such as 1 second, 5 seconds, 10 seconds, 30 seconds, 60 seconds, etc.

[0089] Step S2, defining the current information when the current value reaches a preset threshold as a current input state, and defining the current information when the current value is zero as a current exit state.

[0090] Preferably, the preset threshold value of the current value is generally a value between 0.5A and 20A, and the specific value is determined according to the model of the protection pressure plate.

[0091] Step S3, defining the current state classification of all connected pieces, the current state classification including current input state and current exit state.

[0092] Preferably, the current state classification can be set according to the above-mentioned preset threshold, and an error range can be appropriately added, for example, the current input state interval is a value between 0.5A and 20A plus or minus 0.1A; the current exit state interval is 0A.

[0093] Step S4, defining the visual state classification of the connected pieces, the visual state classification includes visual engagement state, visual exit state, and unclassified state.

[0094] Preferably, see Figure 1 , Figure 1When the connecting piece is in the exit state, there is an obvious and intuitive gap between the end of the connecting piece away from the rotating contact and the fixed contact, and at this time the connecting piece is not in a vertical state, but at an angle of about 30° with the vertical direction. The accuracy of using the gap and angle as visual recognition features is high. The visual engagement state is that the connecting piece is in a vertical state, and the end of the connecting piece away from the rotating contact coincides with the fixed contact; the visual exit state is that the connecting piece is in an inclined state with an angle of 30° to the vertical direction, and there is a gap between the end of the connecting piece away from the rotating contact and the fixed contact, and the complete fixed contact can be seen visually.

[0095] Step S5, obtaining image information of each connected piece by using a target detection algorithm at preset time intervals.

[0096] Preferably, the preset time period in step S5 is set the same as the preset time period in step S1.

[0097] Step S6, obtaining the visual boundary of each connected piece by using a sliding window algorithm.

[0098] Preferably, in order to make the visual boundary of the continuous piece clear enough, the size of the 3×3 window of the sliding window algorithm can be set to the pixel level, for example, 1 pixel as a step, 10 pixels as a step, etc., but the computing power burden of the computer needs to be considered.

[0099] Step S7, classifying all visual boundaries into visual state classifications through a classification algorithm.

[0100] Preferably, this embodiment adopts Naive Bayes to perform classification, and classifies each visual boundary into the visual state classification with the highest conditional probability, so as to prevent subjective errors in manual classification.

[0101] Step S8, traversing the current on / off state and visual on / off state corresponding to each link, and defining the link whose current on / off state does not match the visual on / off state as an abnormal link.

[0102] Preferably, if the current throw-in / out state and the visual throw-in / out state of the same connected piece are both in the throw-in state, they match; similarly, if the current throw-in / out state and the visual throw-in / out state of the same connected piece are both in the throw-out state, they match; otherwise, they do not match.

[0103] Step S9, defining the connected pieces classified into the unclassified state as abnormal connected pieces, and obtaining the position information of all abnormal connected pieces based on the board surface to send to the external receiving end.

[0104] Furthermore, step S5, obtaining image information of each contiguous piece by using a target detection algorithm at preset time intervals, specifically includes the following steps:

[0105] Step S51, photographing all the connected pieces on the board through a preset photographing end based on an external fixed position, and obtaining a visual image based on all the connected pieces each time photographing.

[0106] Preferably, the preset shooting end can be set on the normal line of the board surface, and the normal line passes through the midpoint of the board surface. At the same time, in order to reduce the error caused by the perspective characteristic, the distance between the preset shooting end and the board surface can be set to twice the long side of the board surface. At the same time, a surface light source facing the board surface is set up around the preset shooting end to further reduce the monitoring error caused by the shadow.

[0107] Step S52, dividing the visual image into a preset number of grids on average, wherein the size of each grid is smaller than the size of each connected piece.

[0108] Preferably, the size of the grid may adopt the above-mentioned pixel-level size.

[0109] Step S53, based on all grids, detect the connected pieces in the visual image according to the target detection algorithm and obtain several bounding boxes.

[0110] Preferably, the target detection algorithm of this embodiment mainly includes traditional target detection algorithms VJ, HOG, DPMDetector; deep learning Two-stage target detection algorithms RCNN, SPPNet, FastRCNN, FasterRCNN; target detection Trick algorithms FPN, CascadeRCNN; deep learning one-stage target detection algorithms Yolo, X, SSD, RetinaNet; deep learning Anchor-free target detection algorithms CornerNet, CenterNet, FCOS; Transformer-based target detection algorithm DETR.

[0111] Preferably, this embodiment prefers the Yolo algorithm.

[0112] Step S54, respectively obtain the confidence of each bounding box, and mark the bounding box with the highest confidence among all the bounding boxes as a first-order bounding box.

[0113] Step S55 , calculating the intersection-and-union ratio of each other bounding box with the first-order bounding box, and selecting a bounding box whose intersection-and-union ratio is greater than or equal to a preset threshold as a second-order bounding box.

[0114] Preferably, the intersection-over-union ratio is the intersection of the second-order bounding box and each first-order bounding box, divided by the union of the second-order bounding box and each first-order bounding box, so that the ratio obtained may be an area ratio.

[0115] Preferably, the preset threshold of the intersection-over-union ratio may be set to 0.5.

[0116] Step S56: Obtain the second-order bounding box with the highest confidence among all the second-order bounding boxes and use it as the third-order bounding box.

[0117] Step S57, defining a third-order bounding box as all connected image information.

[0118] Step S58, evenly dividing all the connected pieces of image information to form a number of image units, each image unit having only one connected piece.

[0119] Step S59, defining each image unit as each piece of image information.

[0120] Preferably, each grid is used to predict The coordinates, width and height of the third-order bounding boxes, as well as the confidence of each third-order bounding box, that is, each grid needs to predict value.

[0121] It can be understood that each grid needs to predict indivual ;in, is the offset of the center of the third-order bounding box relative to the grid, is the ratio of the third-order bounding box to the above resized image, is the confidence of the grid, which takes a value of 1 or 0.

[0122] Preferably, the confidence level can be understood as whether there is a target in the current grid and the accuracy of the third-order bounding box.

[0123] For example: suppose there is a target in a resized image, and the width and height of the resized image are but:

[0124] Divide the image into 7×7 (S×S) grids evenly, and there is a grid located at the center of the target. The coordinates of the grid are , let the coordinates of the center of the target be , then according to Calculate the above offset.

[0125] Preferably, in actual detection, if the predicted third-order bounding box and the actual bounding box overlap perfectly, the intersection-and-union ratio is 1. In actual application, the value of the preset ratio can generally be set to 0.5 to determine whether the predicted bounding box is correct, and the accuracy of the bounding box is positively correlated with the intersection-and-union ratio.

[0126] Preferably, the YOLO algorithm also needs to train the third-order bounding box to improve the accuracy of target detection.

[0127] It should be noted that the training model of the third-order bounding box is generally implemented directly through code, for example:

[0128] “class Yolo(object):

[0129] def __init__(self, weights_file, verbose=True):

[0130] self.verbose = verbose

[0131] # detection params

[0132] self.S = 7 # cell size

[0133] self.B = 2 # boxes_per_cell

[0134] self.classes = [*]

[0135] self.C = len(self.classes) # number of classes

[0136] # offset for box center (top left point of each cell)

[0137] self.x_offset= np.transpose(np.reshape(np.array([np.arange(self.S)]*self.S*self.B),

[0138] [self.B, self.S, self.S]), [1, 2, 0])

[0139] self.y_offset = np.transpose(self.x_offset, [1, 0, 2])

[0140] self.threshold = 0.2# confidence scores threshold

[0141] self.iou_threshold = 0.4

[0142] #the maximum number of boxes to be selected by non max suppression

[0143] self.max_output_size = 10".

[0144] Among them, "*" is the target category that needs to be identified, which can be set to continuous English words such as "switch" and "joint".

[0145] Next, the above training model is trained with a preset target training set, and the weights and biases of the training model are adjusted iteratively a certain number of times through the back propagation algorithm to reduce the value of the loss function of the training model.

[0146] Preferably, the loss function is .

[0147] in, For the Grid An indicator function of whether a third-order bounding box is responsible for the target, with a value of 1 or 0; , , , , Corresponding to indivual Predicted value.

[0148] It can be understood that the loss function includes the coordinate value deviation of the third-order bounding box, the confidence deviation, and the prediction probability deviation (or category deviation).

[0149] in, is the third-order bounding box midpoint loss in the coordinate value deviation, is the third-order bounding box width and height loss in the coordinate value deviation, is the confidence deviation, is the deviation of the predicted probability (or class deviation).

[0150] in, is the positioning error penalty. Generally, ; That is the S×S grids mentioned above; is the number of third-order bounding boxes; and For the The estimated values ​​of the horizontal and vertical coordinates of the midpoints of the third-order bounding boxes; and For the An estimate of the width and height of a third-order bounding box; For the The confidence of the third-order bounding box; For the An estimate of the confidence of a third-order bounding box; is the confidence prediction loss, generally ; For the The class probabilities of the third-order bounding boxes; For the The estimated value of the class probability of the third-order bounding box; and In correspond .

[0151] It should be noted that since each grid does not necessarily contain a target, if there is no target in the grid, it will lead to The value of is 0, which makes the gradient span in the subsequent back propagation algorithm too large, so we introduce To control the loss of the third-order bounding box prediction position, and introduce Controls the penalty for non-existent targets within a single grid.

[0152] It should be noted that the symbolic meaning of the above YOLO algorithm is not interchangeable with the symbolic meanings of other positions in this embodiment.

[0153] Furthermore, in step S6, the visual boundary of each connected piece is obtained by a sliding window algorithm, including:

[0154] Step S61, binarizing each piece of image information to obtain a black and white visual image.

[0155] Step S62, dividing the black and white visual image by a grid of preset density.

[0156] Step S63, respectively obtain the color of each grid, and assign a value of 0 to a grid with a black color, and assign a value of 1 to a grid with a white color.

[0157] Step S64, defining each grid and eight adjacent grids as a 3×3 window.

[0158] Step S65, traverse all grids through the 3×3 window with one grid as one step.

[0159] Step S66, based on all traversal results, delete the grids with all values ​​assigned to them as 1 and the grids with all values ​​assigned to them as 0 in the 3×3 window.

[0160] Step S67, obtaining the retained grid and defining it as the visual boundary of the current connected piece.

[0161] Furthermore, in step S7, all visual boundaries are classified into visual state classifications by a classification algorithm, including:

[0162] Step S71: define a set of data to be classified according to all visual boundaries ,in The data set to be classified The A visual boundary, is the number of all visual boundaries.

[0163] Step S72: Define a category set based on visual state classification ,in, For category collection The Visual state classification, The number of categories for all visual states.

[0164] Step S73, calculate the conditional probability of each visual boundary under each visual state classification according to formula (1):

[0165] (1).

[0166] in, For the Under the visual state classification, the data set to be classified The conditional probability of For the The marginal probability of each visual state classification; For the Under the visual state classification, The conditional probability of a visual boundary.

[0167] Step S74, classify each visual boundary into the visual state classification with the highest conditional probability.

[0168] Preferably, this embodiment adopts Naive Bayes to perform classification, and classifies each electrical device into the load level with the highest conditional probability, so as to prevent subjective errors in manual classification.

[0169] Specifically, Naive Bayes is defined as follows:

[0170] ① Let x={a1, a2, a3,…, an} be an item to be classified, and each a is a feature of x.

[0171] ②There is a category set c={y1, y2, y3, ..., y m}.

[0172] ③ Calculate P(y1|x), P(y2|x), ..., P(ym |x).

[0173] ④If P(y k |x)=max{P(y1|x),P(y2|x),...,P(y m |x)}, then x∈y k .

[0174] Then calculate the conditional probabilities in step ③ through the following steps:

[0175] 1. Find a set of items to be classified with known classifications. This set is called a training sample set.

[0176] 2. Statistically obtain the conditional probability estimate of each feature attribute under each category. That is:

[0177] P(a1|y1),P(a2|y1),……,P(a n |y1)

[0178] P(a1|y2),P(a2|y2),……,P(a n |y2);

[0179] …

[0180] P(a1|y m ),P(a2|y m ),……,P(a n |y m );

[0181] 3. Assuming that each feature attribute is conditionally independent, according to the Bayesian principle:

[0182] P(y i |x)=P(x|y i )P(y i ) / p(x).

[0183] Since the denominator is a constant for all categories, we only need to maximize the numerator. And because each feature attribute is conditionally independent, then:

[0184] P(x|y i )P(y i )=P(a1|y i )P(a2|y i )……P(a n |y i )P(y i ).

[0185] It should be noted that the above preferred contents are also explanations of the principles, and the meanings of their symbols are not interchangeable with the meanings of the symbols of other formulas in this embodiment.

[0186] Further, in step S9, the connected pieces classified into the unclassified state are defined as abnormal connected pieces, and the position information of all abnormal connected pieces based on the board surface is obtained to be sent to the external receiving end, and then, it includes:

[0187] Step S10, dividing the current switching states of all preset time periods of the current contiguous area into a first training set and a first sample set according to a preset ratio.

[0188] Step S20, dividing the visual projection and retreat states of all preset time periods of the current continuous film into a second training set and a second sample set according to a preset ratio.

[0189] Preferably, the preset ratio can be set to 8:2, that is, the training set and the sample set are divided into a ratio of 8:2.

[0190] Preferably, in actual application, a certain proportion of validation sets are required to verify the accuracy of the model, that is, verification is required after training is completed, and the sample set prediction is started only after the verification is successful. Usually, the image data is divided into training set, validation set, and sample set in a ratio of 70%:15%:15%, that is, 70% of the data is the training set, 15% of the data is the validation set, and 15% of the data is the sample set.

[0191] Step S30, input the first training set into the preset neural network model and perform training to obtain a first prediction model, and input the second training set into the preset neural network model and perform training to obtain a second prediction model.

[0192] Preferably, the training model training a neural network usually requires providing a large amount of data, namely a data set; the data set is generally divided into three categories, namely the above-mentioned training set (training set), validation set (validation set) and test set (test set).

[0193] Among them, one epoch is the process of training once with all the samples in the training set. The so-called training once refers to one forward pass and one back pass. When the number of samples in an epoch (i.e., training set) is too large, training once may consume too much time, and it is not necessary to use all the data in the training set for each training. In this case, the entire training set needs to be divided into multiple small blocks, that is, divided into multiple batches for training. An epoch consists of one or more batches, where a batch is a part of the training set. Each training process only uses a part of the data, i.e., a batch. The process of training a batch is an iteration.

[0194] Preferably, the neural network training specifically includes a perceptron, which is composed of two layers of neurons. The input layer receives external input signals and transmits them to the output layer. The output layer is MP neurons, and the step function is .

[0195] Preferably, given a training data set, the weights ( =1,2,...,n) and training bias It can be obtained through learning, It can be understood as the weight corresponding to a fixed value of -1,0. .

[0196] It should be noted that the step function here has no interchangeable symbolic meaning with other formulas in the embodiment. This step function is only used for principle explanation and does not participate in the calculation of other formulas.

[0197] Preferably, in this embodiment, the number of neural network training times can be set to 1000 times.

[0198] Preferably, the learning rate from the 1st to the 500th epoch can be set to 0.01, the learning rate from the 501st to the 750th epoch can be set to 0.001, and the learning rate from the 751st to the 1000th epoch can be set to 0.0001.

[0199] It can be understood that the neural network training of this embodiment mainly includes the following ideas:

[0200] ① Initialize the weights and bias items in the network.

[0201] Initializing parameter values ​​(output unit weights, bias terms and hidden unit weights, bias terms are all model parameters) is to activate forward propagation, obtain the output value of each layer element, and then obtain the value of the loss function.

[0202] ②Activate forward propagation to obtain the output value of each layer and the expected value of the loss function of each layer.

[0203] ③According to the loss function, calculate the error term of the output unit and the error term of the hidden unit.

[0204] Calculate various errors, calculate the gradient of the parameters with respect to the loss function, or calculate partial derivatives according to the chain rule of calculus. For partial derivatives of vectors or matrices in a composite function, the partial derivative of the function inside the composite function is always multiplied on the left; for partial derivatives of scalars in a composite function, the partial derivative of the function inside the composite function can be multiplied on the left or on the right.

[0205] ④Update the weights and bias items in the neural network.

[0206] ⑤ Repeat ②~④ until the loss function is less than the preset bias or the number of iterations is used up, and the output parameters at this time are the current optimal parameters.

[0207] Step S40: input the first sample set into the first prediction model to obtain a first prediction value, and input the second sample set into the second prediction model to obtain a second prediction value.

[0208] Step S50, obtaining a mismatch ratio between the second predicted value and the first predicted value.

[0209] Preferably, the mismatch ratio here is for the current contiguous piece, that is, each contiguous piece has a mismatch ratio. After simulation, after the current throw-in and visual throw-out are trained by the neural network respectively, the two throw-in and throw-out states predicted based on the same contiguous piece have a mismatch ratio of 0.58% to 4.21%, and some predicted samples reach 10.07%.

[0210] Step S60, defining the mismatch ratio as the failure rate of the current connected slice.

[0211] Step S70, obtaining the failure rates of all connected slices and sending them to an external receiving end.

[0212] Further, step S30, inputting the first training set into a preset neural network model and training it to obtain a first prediction model, and inputting the second training set into a preset neural network model and training it to obtain a second prediction model, including:

[0213] Step S301, defining the topological relationship of the preset neural network model, the topological relationship including an input layer, a hidden layer, and an output layer that are sequentially signal-connected.

[0214] Preferably, in the actual training process of the visual part, since the continuous pieces may have different colors, the image data needs to be normalized to prevent the large difference in data magnitude during the training process from resulting in a lack of significance. In this embodiment, the visual boundary can be first coordinateized, and the boundary of the image data can be defined as 0 to 1 (the image data can be cut into squares at the same position at this time). At this time, the coordinate values ​​of the visual boundary are all between 0 and 1, and training can be performed based on the coordinate values. The number of coordinates is positively correlated with or equal to the above-mentioned grid size.

[0215] Step S302: input the first training set to the input layer, train the first training set for a first preset number of times using the LM estimation method, and obtain a first root mean square error between the first sample set and the current training result based on each training.

[0216] It is worth noting that the first root mean square error and the second root mean square error mentioned below are only used to distinguish different application scenarios of two root mean square errors: the first sample set and the second sample set, rather than referring to one of the first sample set and the second sample set having two root mean square errors. The same applies to the first minimum value and the second minimum value mentioned below.

[0217] Step S303, obtaining a first minimum value among all first root mean square errors, and obtaining a training result corresponding to the first minimum value as a first prediction model.

[0218] Step S304, input the second training set to the input layer, train the second training set for a second preset number of times using the LM estimation method, and obtain a second root mean square error between the second sample set and the current training result based on each training.

[0219] Preferably, the LM estimation method is specifically the Levenberg-Marquardt method, which is a nonlinear least squares method used to solve nonlinear optimization problems, such as the neural network model in this embodiment. The method uses the gradient information of the objective function to adjust the parameters in each iteration to find the best model parameter estimation. The method uses an adjustment factor called Levenberg-Marquardt, which balances the gradient descent and Gauss-Newton method according to the current iteration step. In each iteration, the method calculates a parameter increment, and then decides whether to accept the parameter increment by comparing the size of the current model residual and the model residual estimated using the new parameter.

[0220] Preferably, the iteration step size of the Levenberg-Marquardt method is given by Representation.

[0221] in, is the iteration step size of the Levenberg-Marquardt method, is the current model residual, is the Jacobian matrix of the current model residual, is the transposed matrix of the Jacobian matrix of the current model residual, is the Levenberg-Marquardt adjustment factor mentioned above, is the identity matrix, is the covariance matrix.

[0222] Preferably, the order of the identity matrix is ​​the same as the order of the Jacobian matrix.

[0223] Preferably, when hour, is a positive number, Prefer the Gauss-Newton method; when hour, Tends to balance gradient descent.

[0224] Step S305, obtaining the second minimum value among all the second root mean square errors, and obtaining the training result corresponding to the second minimum value as the second prediction model.

[0225] Preferably, the root mean square error is: .

[0226] in, is the root mean square error, is the number of the first training set or the number of the second training set, For the The true value of the first training set or The true value of the second training set, For the The first training set or The training results after the second training set training is completed.

[0227] Furthermore, the preset neural network model is represented by formula (2):

[0228] (2).

[0229] in, To preset the neural network model; is the input layer and , The input layer input nodes, each input node corresponds to a first training set or a second training set, The input layer The first input node to the hidden layer The weights of the input nodes; is connected to the hidden layer The bias of the input node; is the transfer function, and .

[0230] Preferably, the number in the brackets of the superscript at the upper right corner of the symbol in formula (2) is the number of layers, for example, superscript (1) is the first layer, i.e., the input layer, superscript (2) is the second layer, i.e., the hidden layer, and superscript (1, 2) is from the first layer to the second layer, i.e., from the input layer to the hidden layer.

[0231] Preferably, the learning rate of the neural network model is the above-mentioned iteration step size. A smaller learning rate will make the update speed of weights and biases slower, which helps to find the optimal solution of the loss function more accurately, but may lead to slower convergence. A larger learning rate will make the update speed of weights and biases faster, which can converge faster, but may miss the optimal solution. The best learning rate can usually be determined by trying different learning rates and observing the decrease of the loss function. If the loss function decreases slowly or is unstable, you can try to increase the learning rate; if the loss function decreases too quickly or does not converge, you can try to reduce the learning rate.

[0232] Preferably, 1000 samples can be collected and the above-mentioned "the learning rate from the 1st to the 500th epoch can be set to 0.01, the learning rate from the 501st to the 750th epoch can be set to 0.001, and the learning rate from the 751st to the 1000th epoch can be set to 0.0001" can be adopted.

[0233] In this embodiment, the current information of each fixed contact is obtained through each current detection component at intervals of preset time periods; the current information of the current value reaching the preset threshold is defined as the current input state, and the current information of the current value of zero is defined as the current exit state; the current state classification of all connected pieces is defined, and the current state classification includes the current input state and the current exit state; the visual state classification of the connected pieces is defined, and the visual state classification includes the visual input state, the visual exit state, and the unclassified state; the image information of each connected piece is obtained through the target detection algorithm at intervals of preset time periods; the visual boundary of each connected piece is obtained through the sliding window algorithm; all visual boundaries are classified into the visual state classification through the classification algorithm; the current input and output state and the visual input and output state corresponding to each connected piece are traversed, and the connected piece whose current input and output state does not match the visual input and output state is defined as an abnormal connected piece; the connected piece classified into the unclassified state is defined as an abnormal connected piece, and the position information of all abnormal connected pieces based on the board surface is obtained and sent to the external receiving end. This embodiment adopts the dual monitoring means of electrical monitoring and visual monitoring to ensure more accurate monitoring results. By monitoring the current throw-in and throw-out states and the visual throw-in and throw-out states of each link respectively, and verifying whether the link is in the same state in the current throw-in and throw-out states and the visual throw-in and throw-out states at the same time (the same link is in the throw-in state in the current throw-in and throw-out state, and is in the throw-in state in the visual throw-in and throw-out state at the same time; or the same link is in the throw-out state in the current throw-in and throw-out state, and is in the throw-out state in the visual throw-in and throw-out state at the same time), it is determined whether the link is abnormal, and the link whose current throw-in and throw-out states do not match the visual throw-in and throw-out states is defined as an abnormal link. Compared with the prior art, this embodiment does not require manual participation in the monitoring process, thereby eliminating subjective errors and reducing labor costs, and the dual judgment mechanism can further improve the monitoring accuracy.

[0234] like Figure 3 As shown, this embodiment provides an embodiment of a monitoring device for a transformer substation protection pressure plate. In this embodiment, the monitoring device is applied to the monitoring method as in the above embodiment, and the monitoring device includes a contact current information acquisition module 1, a current throw-in / out state definition module 2, a current state classification definition module 3, a visual state classification definition module 4, a continuous image information acquisition module 5, a continuous visual boundary acquisition module 6, a continuous visual boundary classification module 7, a throw-in / out state dual judgment module 8, and an abnormal continuous definition sending module 9 which are electrically connected in sequence.

[0235] Among them, the contact current information acquisition module 1 is used to obtain the current information of each fixed contact through each current detection element at intervals of a preset time period; the current throw-in / out state definition module 2 is used to define the current information of the current value reaching the preset threshold as the current throw-in state, and define the current information of the current value zero as the current exit state; the current state classification definition module 3 is used to define the current state classification of all connected pieces, and the current state classification includes the current throw-in state and the current exit state; the visual state classification definition module 4 is used to define the visual state classification of the connected pieces, and the visual state classification includes the visual throw-in state, the visual exit state, and the unclassified state; the connected piece image information acquisition module 5 is used to define the current state classification of all connected pieces. The image information of each piece is obtained by the target detection algorithm at preset time intervals; the piece visual boundary acquisition module 6 is used to obtain the visual boundary of each piece by the sliding window algorithm; the piece visual boundary classification module 7 is used to classify all visual boundaries into visual state classifications by the classification algorithm; the dual determination module for the throw-in / out state 8 is used to traverse the current throw-in / out state and the visual throw-in / out state corresponding to each piece, and define the piece whose current throw-in / out state does not match the visual throw-in / out state as an abnormal piece; the abnormal piece definition sending module 9 is used to define the piece classified into the unclassified state as an abnormal piece, and obtain the position information of all abnormal pieces based on the board surface to send to the external receiving end.

[0236] Furthermore, the contiguous image information acquisition module specifically includes a first contiguous image information acquisition submodule, a second contiguous image information acquisition submodule, a third contiguous image information acquisition submodule, a fourth contiguous image information acquisition submodule, a fifth contiguous image information acquisition submodule, a sixth contiguous image information acquisition submodule, a seventh contiguous image information acquisition submodule, an eighth contiguous image information acquisition submodule, and a ninth contiguous image information acquisition submodule, which are electrically connected in sequence; the first contiguous image information acquisition submodule is electrically connected to the visual state classification definition module, and the ninth contiguous image information acquisition submodule is electrically connected to the contiguous visual boundary acquisition module.

[0237] Among them, the first contiguous image information acquisition submodule is used to shoot all the contiguous pieces on the board through a preset shooting end based on an external fixed position, and each shooting obtains a visual image based on all the contiguous pieces; the second contiguous image information acquisition submodule is used to evenly divide the visual image into a preset number of grids, and the size of each grid is smaller than the size of each contiguous piece; the third contiguous image information acquisition submodule is used to detect the contiguous pieces in the visual image according to the target detection algorithm based on all grids and obtain several bounding boxes; the fourth contiguous image information acquisition submodule is used to respectively obtain the confidence of each bounding box, and mark the bounding box with the highest confidence among all bounding boxes as the first-order bounding box; the fifth contiguous image information acquisition submodule is used to divide the visual image into a preset number of grids, and the size of each grid is smaller than the size of each contiguous piece; the third contiguous image information acquisition submodule is used to detect the contiguous pieces in the visual image according to the target detection algorithm based on all grids and obtain several bounding boxes; the fourth contiguous image information acquisition submodule is used to respectively obtain the confidence of each bounding box, and mark the bounding box with the highest confidence among all bounding boxes as the first-order bounding box; The image information acquisition submodule is used to calculate the intersection and union ratio of each other bounding box with the first-order bounding box, and select the bounding box with the intersection and union ratio greater than or equal to the preset threshold as the second-order bounding box; the sixth contiguous image information acquisition submodule is used to obtain the second-order bounding box with the highest confidence among all second-order bounding boxes and use it as the third-order bounding box; the seventh contiguous image information acquisition submodule is used to define the third-order bounding box as the image information of all contiguous pieces; the eighth contiguous image information acquisition submodule is used to evenly divide the image information of all contiguous pieces to form a number of image units, each image unit has and only has one contiguous piece; the ninth contiguous image information acquisition submodule is used to define each image unit as the image information of each contiguous piece.

[0238] Furthermore, the contiguous visual boundary acquisition module specifically includes a first contiguous visual boundary acquisition submodule, a second contiguous visual boundary acquisition submodule, a third contiguous visual boundary acquisition submodule, a fourth contiguous visual boundary acquisition submodule, a fifth contiguous visual boundary acquisition submodule, a sixth contiguous visual boundary acquisition submodule, and a seventh contiguous visual boundary acquisition submodule, which are electrically connected in sequence; the first contiguous visual boundary acquisition submodule is electrically connected to the ninth contiguous image information acquisition submodule, and the seventh contiguous visual boundary acquisition submodule is electrically connected to the contiguous visual boundary classification module.

[0239] Among them, the first connected visual boundary acquisition submodule is used to binarize the image information of each connected piece respectively to obtain a black and white visual image; the second connected visual boundary acquisition submodule is used to divide the black and white visual image through grids of preset density; the third connected visual boundary acquisition submodule is used to obtain the color of each grid respectively, and assign the black grid to 0 and the white grid to 1; the fourth connected visual boundary acquisition submodule is used to define each grid and the eight adjacent grids as a 3×3 window; the fifth connected visual boundary acquisition submodule is used to traverse all grids through the 3×3 window with one grid as a step; the sixth connected visual boundary acquisition submodule is used to delete the grids with all values ​​1 and all values ​​0 in the 3×3 window based on all traversal results; the seventh connected visual boundary acquisition submodule is used to obtain the retained grids and define them as the visual boundaries of the current connected piece.

[0240] Furthermore, the contiguous visual boundary classification module specifically includes a first contiguous visual boundary classification submodule, a second contiguous visual boundary classification submodule, a third contiguous visual boundary classification submodule, and a fourth contiguous visual boundary classification submodule, which are electrically connected in sequence; the first contiguous visual boundary classification submodule is electrically connected to the seventh contiguous visual boundary acquisition submodule, and the fourth contiguous visual boundary classification submodule is electrically connected to the dual determination module of the investment and withdrawal status.

[0241] The first contiguous visual boundary classification submodule is used to define a set of data to be classified according to all visual boundaries. ,in The data set to be classified The A visual boundary, is the number of all visual boundaries.

[0242] The second contiguous visual boundary classification submodule is used to define a category set according to the visual state classification. ,in, For category collection The Visual state classification, The number of categories for all visual states.

[0243] The third contiguous visual boundary classification submodule is used to calculate the conditional probability of each visual boundary under each visual state classification according to formula (1):

[0244] (1).

[0245] in, For the Under the visual state classification, the data set to be classified The conditional probability of For the The marginal probability of each visual state classification; For the Under the visual state classification, The conditional probability of a visual boundary.

[0246] The fourth contiguous visual boundary classification submodule is used to classify each visual boundary into the visual state classification with the highest conditional probability.

[0247] Furthermore, the monitoring device also includes a current throw-in / out state sample division module, a visual throw-in / out state sample division module, an throw-in / out state sample training module, an throw-in / out state sample prediction module, an throw-in / out state mismatch ratio acquisition module, a fault rate definition module, and a fault rate acquisition and sending module, which are electrically connected in sequence; the current throw-in / out state sample division module is electrically connected to the abnormal connection definition and sending module.

[0248] Among them, the current throw-in / out state sample division module is used to divide the current throw-in / out state of all preset time periods of the current connection into a first training set and a first sample set according to a preset ratio; the visual throw-in / out state sample division module is used to divide the visual throw-in / out state of all preset time periods of the current connection into a second training set and a second sample set according to a preset ratio; the throw-in / out state sample training module is used to input the first training set into a preset neural network model and train it to obtain a first prediction model, and input the second training set into a preset neural network model and train it to obtain a second prediction model; the throw-in / out state sample prediction module is used to input the first sample set into the first prediction model to obtain a first prediction value, and input the second sample set into the second prediction model to obtain a second prediction value; the throw-in / out state mismatch ratio acquisition module is used to obtain the mismatch ratio between the second prediction value and the first prediction value; the failure rate definition module is used to define the mismatch ratio as the failure rate of the current connection; the failure rate acquisition and sending module is used to obtain the failure rates of all connections and send them to an external receiving end.

[0249] Furthermore, the investment and withdrawal state sample training module specifically includes a first investment and withdrawal state sample prediction submodule, a second investment and withdrawal state sample prediction submodule, a third investment and withdrawal state sample prediction submodule, a fourth investment and withdrawal state sample prediction submodule, and a fifth investment and withdrawal state sample prediction submodule, which are electrically connected in sequence; the first investment and withdrawal state sample prediction submodule is electrically connected to the visual investment and withdrawal state sample division module, and the fifth investment and withdrawal state sample prediction submodule is electrically connected to the investment and withdrawal state sample prediction module.

[0250] Among them, the first investment and withdrawal status sample prediction submodule is used to define the topological relationship of the preset neural network model, and the topological relationship includes an input layer, a hidden layer, and an output layer that are sequentially signal-connected; the second investment and withdrawal status sample prediction submodule is used to input the first training set into the input layer, train the first training set for a first preset number of times through the LM estimation method, and obtain the first root mean square error between the first sample set and the current training result based on each training; the third investment and withdrawal status sample prediction submodule is used to obtain the first minimum value among all first root mean square errors, and obtain the training result corresponding to the first minimum value as the first prediction model; the fourth investment and withdrawal status sample prediction submodule is used to input the second training set into the input layer, train the second training set for a second preset number of times through the LM estimation method, and obtain the second root mean square error between the second sample set and the current training result based on each training; the fifth investment and withdrawal status sample prediction submodule is used to obtain the second minimum value among all second root mean square errors, and obtain the training result corresponding to the second minimum value as the second prediction model.

[0251] Furthermore, the first investment and withdrawal status sample prediction submodule is equipped with a preset neural network model represented by formula (2):

[0252] (2).

[0253] in, To preset the neural network model; is the input layer and , The input layer input nodes, each input node corresponds to a first training set or a second training set, The input layer The first input node to the hidden layer The weights of the input nodes; is connected to the hidden layer The bias of the input node; is the transfer function, and .

[0254] It should be noted that this embodiment is an apparatus embodiment based on the above method embodiment. Additional contents such as the preference, extension, limitation, and example illustration of this embodiment can be found in the above method embodiment, and will not be elaborated in this embodiment.

[0255] In this embodiment, the current information of each fixed contact is obtained through each current detection component at intervals of preset time periods; the current information of the current value reaching the preset threshold is defined as the current input state, and the current information of the current value of zero is defined as the current exit state; the current state classification of all connected pieces is defined, and the current state classification includes the current input state and the current exit state; the visual state classification of the connected pieces is defined, and the visual state classification includes the visual input state, the visual exit state, and the unclassified state; the image information of each connected piece is obtained through the target detection algorithm at intervals of preset time periods; the visual boundary of each connected piece is obtained through the sliding window algorithm; all visual boundaries are classified into the visual state classification through the classification algorithm; the current input and output state and the visual input and output state corresponding to each connected piece are traversed, and the connected piece whose current input and output state does not match the visual input and output state is defined as an abnormal connected piece; the connected piece classified into the unclassified state is defined as an abnormal connected piece, and the position information of all abnormal connected pieces based on the board surface is obtained and sent to the external receiving end. This embodiment adopts the dual monitoring means of electrical monitoring and visual monitoring to ensure more accurate monitoring results. By monitoring the current throw-in and throw-out states and the visual throw-in and throw-out states of each link respectively, and verifying whether the link is in the same state in the current throw-in and throw-out states and the visual throw-in and throw-out states at the same time (the same link is in the throw-in state in the current throw-in and throw-out state, and is in the throw-in state in the visual throw-in and throw-out state at the same time; or the same link is in the throw-out state in the current throw-in and throw-out state, and is in the throw-out state in the visual throw-in and throw-out state at the same time), it is determined whether the link is abnormal, and the link whose current throw-in and throw-out states do not match the visual throw-in and throw-out states is defined as an abnormal link. Compared with the prior art, this embodiment does not require manual participation in the monitoring process, thereby eliminating subjective errors and reducing labor costs, and the dual judgment mechanism can further improve the monitoring accuracy.

[0256] like Figure 4 As shown, this embodiment provides an embodiment of an electronic device. In this embodiment, the electronic device 10 includes a processor 101 and a memory 102 coupled to the processor 101 .

[0257] The memory 102 stores program instructions for implementing the monitoring method of the substation protection pressure plate of any of the above embodiments.

[0258] The processor 101 is used to execute program instructions stored in the memory 102 to monitor the protection pressure plate of the substation.

[0259] The processor 101 may also be referred to as a CPU (Central Processing Unit). The processor 101 may be an integrated circuit chip having data processing capabilities. The processor 101 may also be a general-purpose processor, a digital data processor (DSP), an application-specific integrated circuit (ASIC), a field programmable gate array (FPGA) or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components. The general-purpose processor may be a microprocessor or the processor may also be any conventional processor, etc.

[0260] Further, Figure 5 The schematic diagram of the structure of the storage medium of an embodiment of the present application is that the storage medium 11 of the embodiment of the present application stores program instructions 111 that can implement all the above methods, wherein the program instructions 111 can be stored in the above storage medium in the form of a software product, including several instructions for enabling a computer device (which can be a personal computer, server, or network device, etc.) or a processor to execute all or part of the steps of the methods described in each embodiment of the present application. The aforementioned storage medium includes: various media that can store program codes, such as a USB flash drive, a mobile hard disk, a read-only memory (ROM), a random access memory (RAM), a magnetic disk or an optical disk, or terminal devices such as a computer, a server, a mobile phone, and a tablet.

[0261] In the several embodiments provided in the present application, it should be understood that the disclosed systems, devices and methods can be implemented in other ways. For example, the device embodiments described above are only schematic. For example, the division of units is only a logical function division. There may be other division methods in actual implementation. For example, multiple units or components can be combined or integrated into another system, or some features can be ignored or not executed. Another point is that the mutual coupling or direct coupling or communication connection shown or discussed can be an indirect coupling or communication connection through some interfaces, devices or units, which can be electrical, mechanical or other forms.

[0262] In addition, each functional unit in each embodiment of the present application can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit. The above integrated unit can be implemented in the form of hardware or in the form of software functional units. The above is only an implementation method of the present application, and does not limit the patent scope of the present application. Any equivalent structure or equivalent process transformation made using the contents of the specification and drawings of this application, or directly or indirectly used in other related technical fields, is also included in the patent protection scope of the present application.

[0263] The specific implementation methods of the present application are described in detail above, but they are only examples, and the present application is not limited to the specific implementation methods described above. For those skilled in the art, any equivalent modification or substitution of the present application is also within the scope of the present application, and therefore, the equalization, modification, and improvement made without departing from the spirit and principle of the present application should be included in the scope of the present application.

Claims

1. A monitoring method for a transformer substation protection pressure plate, the transformer substation protection pressure plate comprising a plate surface fixed to an external plane, a plurality of rotating contacts penetrating the plate surface, and a plurality of fixed contacts penetrating the plate surface, each fixed contact matches a rotating contact and is spaced a preset distance from the current rotating contact, each rotating contact is rotatably connected to a connecting piece, and the current connecting piece is connected or disconnected with the current fixed contact by rotating, characterized in that: A plurality of current detection components are also installed on one side of the plate away from all the connected pieces, each current detection component matches a fixed contact and is electrically connected to the matching fixed contact. The monitoring method includes: Step S1, obtaining current information of each fixed contact through each current detection element at intervals of a preset time period; Step S2, defining the current information when the current value reaches a preset threshold as a current input state, and defining the current information when the current value is zero as a current exit state; Step S3, defining the current state classification of all connected pieces, wherein the current state classification includes the current input state and the current exit state; Step S4, defining the visual state classification of the connected piece, wherein the visual state classification includes a visual engagement state, a visual exit state, and an unclassified state; Step S5, acquiring image information of each connected piece by using a target detection algorithm at intervals of the preset time period; Step S6, obtaining the visual boundary of each connected piece by using a sliding window algorithm; Step S7, classifying all visual boundaries into the visual state classification by a classification algorithm; Step S8, traversing the current throw-in / out state and visual throw-in / out state corresponding to each link, and defining the link whose current throw-in / out state does not match the visual throw-in / out state as an abnormal link; Step S9, defining the connected pieces classified into the unclassified state as the abnormal connected pieces, and obtaining the position information of all abnormal connected pieces based on the board surface to send to the external receiving end; Step S7, classifying all visual boundaries into the visual state classification by a classification algorithm, including: Step S71: define a set of data to be classified according to all visual boundaries ,in The data set to be classified The A visual boundary, is the number of all visual boundaries; Step S72: defining a category set according to the visual state classification ,in, For the category set The Visual state classification, The number of categories for all visual states; Step S73, calculate the conditional probability of each visual boundary under each visual state classification according to formula (1): （1）； in, For the Under the visual state classification, the data set to be classified The conditional probability of For the The marginal probability of each visual state classification; For the Under the visual state classification, The conditional probability of a visual boundary; Step S74, classify each visual boundary into the visual state classification with the highest conditional probability.

2. The monitoring method of the substation protection pressure plate according to claim 1 is characterized in that: Step S5, obtaining image information of each contiguous piece by using a target detection algorithm at intervals of the preset time period, including: Step S51, photographing all the connected pieces on the board surface by a preset photographing end based on an external fixed position, and obtaining a visual image based on all the connected pieces each time photographing; Step S52, dividing the visual image into a preset number of grids on average, wherein the size of each grid is smaller than the size of each connected piece; Step S53, detecting the connected pieces in the visual image based on all grids according to the target detection algorithm and obtaining a plurality of bounding boxes; Step S54, respectively obtaining the confidence of each bounding box, and marking the bounding box with the highest confidence among all bounding boxes as a first-order bounding box; Step S55, calculating the intersection-and-union ratio of each other bounding box with the first-order bounding box, and selecting a bounding box whose intersection-and-union ratio is greater than or equal to a preset threshold as a second-order bounding box; Step S56, obtaining the second-order bounding box with the highest confidence among all the second-order bounding boxes and using it as the third-order bounding box; Step S57, defining the third-order bounding box as all connected image information; Step S58, evenly dividing all the connected pieces of image information to form a number of image units, each image unit having only one connected piece; Step S59, defining each image unit as each contiguous piece of image information.

3. The monitoring method of the substation protection pressure plate according to claim 1 is characterized in that: Step S6, obtaining the visual boundaries of each connected piece by a sliding window algorithm, including: Step S61, binarizing each piece of image information to obtain a black and white visual image; Step S62, dividing the black and white visual image into grids of a preset density; Step S63, respectively obtaining the color of each grid, and assigning a value of 0 to a grid with a black color, and assigning a value of 1 to a grid with a white color; Step S64, defining each grid and eight adjacent grids as a 3×3 window; Step S65, traversing all grids through the 3×3 window with one grid as one step; Step S66, based on all traversal results, delete the grids whose values ​​are all 1 and the grids whose values ​​are all 0 in the 3×3 window; Step S67, obtaining the retained grid and defining it as the visual boundary of the current connected piece.

4. The monitoring method of the substation protection pressure plate according to claim 1 is characterized in that: Step S9, defining the connected pieces classified into the unclassified state as the abnormal connected pieces, and obtaining the position information of all abnormal connected pieces based on the board surface to send to the external receiving end, and then comprising: Step S10, dividing the current switching states of all preset time periods of the current contiguous area into a first training set and a first sample set according to a preset ratio; Step S20, dividing the visual projection and retreat states of all preset time periods of the current continuous film into a second training set and a second sample set according to the preset ratio; Step S30, inputting the first training set into a preset neural network model and performing training to obtain a first prediction model, and inputting the second training set into the preset neural network model and performing training to obtain a second prediction model; Step S40, inputting the first sample set into the first prediction model to obtain a first prediction value, and inputting the second sample set into the second prediction model to obtain a second prediction value; Step S50, obtaining a mismatch ratio between the second predicted value and the first predicted value; Step S60, defining the mismatch ratio as the failure rate of the current connected piece; Step S70, obtaining the failure rates of all connected slices and sending them to an external receiving end.

5. The monitoring method of the substation protection pressure plate according to claim 4 is characterized in that: Step S30, inputting the first training set into a preset neural network model and training it to obtain a first prediction model, and inputting the second training set into the preset neural network model and training it to obtain a second prediction model, including: Step S301, defining a topological relationship of the preset neural network model, wherein the topological relationship includes an input layer, a hidden layer, and an output layer that are sequentially signal-connected; Step S302, inputting the first training set into the input layer, training the first training set for a first preset number of times by using the LM estimation method, and obtaining a first root mean square error between the first sample set and the current training result based on each training; Step S303, obtaining a first minimum value among all first root mean square errors, and obtaining a training result corresponding to the first minimum value as the first prediction model; Step S304, inputting the second training set into the input layer, training the second training set for a second preset number of times by using the LM estimation method, and obtaining a second root mean square error between the second sample set and the current training result based on each training; Step S305, obtaining the second minimum value among all second root mean square errors, and obtaining the training result corresponding to the second minimum value as the second prediction model.

6. The monitoring method of the substation protection pressure plate according to claim 5 is characterized in that: The preset neural network model is represented by formula (2): （2）； in, The preset neural network model; is the input layer and , is the first input nodes, each input node corresponds to one of the first training sets or one of the second training sets, is the first The input nodes to the hidden layer The weights of the input nodes; is the first The bias of the input node; is the transfer function, and .

7. A monitoring device for a substation protection pressure plate, the monitoring device for a substation protection pressure plate being applied to the monitoring method for a substation protection pressure plate according to any one of claims 1 to 6, characterized in that: The monitoring device of the substation protection pressure plate comprises: A contact current information acquisition module, used to acquire the current information of each fixed contact through each current detection element at intervals of a preset time period; A current input / output state definition module, used to define the current information when the current value reaches a preset threshold as the current input state, and define the current information when the current value is zero as the current output state; A current state classification definition module, used to define the current state classification of all connected pieces, wherein the current state classification includes the current input state and the current exit state; A visual state classification definition module, used to define the visual state classification of the contiguous piece, wherein the visual state classification includes a visual engagement state, a visual exit state, and an unclassified state; A contiguous image information acquisition module, used to acquire image information of each contiguous piece by using a target detection algorithm at intervals of the preset time period; A connected visual boundary acquisition module is used to obtain the visual boundary of each connected piece through a sliding window algorithm; A contiguous visual boundary classification module, used for classifying all visual boundaries into the visual state classification by a classification algorithm; A dual determination module for the throw-in / out status is used to traverse the current throw-in / out status and the visual throw-in / out status corresponding to each connection, and define the connection whose current throw-in / out status does not match the visual throw-in / out status as an abnormal connection; The abnormal contiguous piece definition sending module is used to define the contiguous piece classified into the unclassified state as the abnormal contiguous piece, and obtain the position information of all abnormal contiguous pieces based on the board surface to send to the external receiving end.

8. An electronic device, characterized in that: It includes a processor and a memory coupled to the processor, wherein the memory stores program instructions that can be executed by the processor; when the processor executes the program instructions stored in the memory, it implements the monitoring method for the substation protection pressure plate as described in any one of claims 1 to 6.

9. A storage medium, characterized in that: The storage medium stores program instructions, and when the program instructions are executed by the processor, the method for monitoring the substation protection pressure plate according to any one of claims 1 to 6 can be implemented.

Images