Power operation environment intelligent monitoring method and system based on deep learning

Abstract

The invention relates to the technical field of falling early warning, and discloses a power operation environment intelligent monitoring method and system based on deep learning, and the method comprises the following steps: collecting comprehensive information; the comprehensive information is coded into first graph structure data, and the comprehensive information of each node is coded into first sequence data; inputting the first sequence data into the arc prediction model, and outputting generated arc information; collecting maintenance personnel influence information and tool influence information; the tool influence information is coded into second graph structure data, and the maintenance personnel influence information is coded into third sequence data; inputting the second graph structure data and the third sequence data into a falling prediction model, and outputting probability representing falling of each tool; and when it is judged that the tool falls off, early warning is carried out. According to the method, the generation of the electric arc is predicted, and the falling condition of each tool is predicted according to the possible influence of the generated electric arc on the maintenance personnel and the tools, so that better early warning is carried out.

Description

Technical Field

[0001] The present invention relates to the field of fall warning technology, and more specifically, to a method and system for intelligent monitoring of an electric power operation environment based on deep learning. Background Art

[0002] In some hospitals, data centers, and financial institutions with harsh environments and complex terrain, the continuous supply of electricity is crucial and cannot be easily interrupted. In these places, when repairing or overhauling the power equipment on the tower, live line operations need to be performed at high altitudes to ensure uninterrupted power supply.

[0003] When performing live-line operations at high altitudes, arc flash incidents may occur in high-voltage power equipment. The energy released instantly will not only cause damage to the equipment, but will also affect maintenance personnel and even endanger their personal safety.

[0004] Due to the complex terrain, harsh environment, sudden changes in the environment and sudden increase in wind speed, the tool placement position may deviate, thereby increasing the risk of falling objects from high altitudes.

[0005] In high-voltage power environments, maintenance personnel have limited operational flexibility and their protective equipment is usually heavy. If maintenance personnel are frightened by arc flash or arc burns, their movements may become unstable, which may cause surrounding tools or components to loosen or fall, increasing the risk of falling objects.

[0006] The high temperature of the arc causes the tool to deform and become damaged. The change in shape may lead to some operational errors by maintenance personnel, thereby increasing the risk of falling objects from heights.

[0007] Since the maintenance process requires multiple maintenance personnel to cooperate with each other at high altitudes and on the ground, they are unable to be informed of possible falls in a timely manner, and safety accidents may occur when objects fall from high altitudes.

[0008] The existing technology can judge the possible tool drop situation based on the maintenance personnel and the current status of the tool, but does not take into account the impact of the risk of arcing during the maintenance process on the maintenance personnel and the tools, making it difficult to accurately predict the possible tool drop situation and the early warning effect is poor. Summary of the invention

[0009] The present invention provides a method and system for intelligent monitoring of an electric power operation environment based on deep learning, which solves the technical problems in the related art that the risk of a tool falling is difficult to predict due to the generation of an electric arc and the early warning effect is poor.

[0010] The present invention provides a method for intelligent monitoring of power operation environment based on deep learning, comprising the following steps:

[0011] Step 100, collecting comprehensive information including the current time point and a total of n consecutive time points before, the comprehensive information including route information, environment information and tool information;

[0012] Line information includes: voltage, current, and temperature of the maintenance line;

[0013] Environmental information includes: wind speed, wind direction, temperature, humidity, precipitation, light intensity, and electromagnetic field intensity;

[0014] Tool information includes: melting point, thermal decomposition temperature and thermal conductivity of each tool, location, usage status, and distance between the tool and the line;

[0015] Step 200, encoding the comprehensive information into No. 1 graph structure data, the No. 1 graph structure data includes nodes and edges connecting the nodes, and a node represents a line or environment or tool;

[0016] The relationship between the environment and the line is: the environment affects the line, and there are edges between nodes;

[0017] The relationship between tools and lines is: tools will be in contact with lines, and there will be edges between nodes;

[0018] The relationship between environment and tools is: the environment affects the use of tools, and there are edges between nodes;

[0019] The node characteristics of the node representing the environment are equal to the environment characteristics, the node characteristics of the node representing the line are equal to the line characteristics, and the node characteristics of the node representing the tool are equal to the tool characteristics;

[0020] Environmental features, line features, and tool features are obtained through feature engineering;

[0021] Encode the comprehensive information of each node into No. 1 sequence data, the No. 1 sequence data includes n sequence units, and the t-th sequence unit represents No. 1 graph structure data generated by the comprehensive information of all nodes at the t-th moment;

[0022] Step 300, inputting the first sequence data into an arc prediction model, the arc prediction model comprising a first hidden layer, a second hidden layer and a first output layer, wherein the first hidden layer is used to input the first sequence data, output the first hidden feature to the second hidden layer, the second hidden layer outputs the second hidden layer to the first output layer, and the first output layer outputs information indicating the generated arc;

[0023] Step 400, collecting maintenance personnel impact information and tool impact information including the current time point and a total of n consecutive time points before, the maintenance personnel impact information includes: maintenance personnel information, environmental information, and generated arc information, and the tool impact information includes: tool information, environmental information, and generated arc information;

[0024] Maintenance personnel information includes: maintenance personnel’s posture and heart rate;

[0025] Step 500, encode the tool impact information into No. 2 graph structure data, the No. 2 graph structure data includes nodes and edges connecting nodes, and one node represents a tool or environment or arc; encode the maintenance personnel impact information into No. 3 sequence data, the No. 3 sequence data includes n sequence units, and the t-th sequence unit represents the maintenance personnel impact information at the t-th moment;

[0026] The relationship between environment and tools is: the environment affects the use of tools, and there are edges between nodes;

[0027] The association between tools and arcs is: arcs affect the use of tools, and there are edges between nodes;

[0028] The relationship between the environment and arcs is: the environment affects the generation of arcs, and there are edges between nodes;

[0029] The node characteristics of the node representing the environment are equal to the environment characteristics, the node characteristics of the node representing the arc are equal to the arc characteristics, and the node characteristics of the node representing the tool are equal to the tool characteristics;

[0030] Environmental features, arc features, and tool features are obtained through feature engineering;

[0031] The tool impact information of each node is encoded into second sequence data, the second sequence data includes n sequence units, and the t-th sequence unit represents the tool impact information of a node at the t-th moment;

[0032] Step 600, inputting the No. 2 graph structure data and the No. 3 sequence data into a drop prediction model, the drop prediction model comprising a third hidden layer, a fourth hidden layer, a fifth hidden layer, a first feature fusion layer, a second output layer, and a third output layer, wherein the third hidden layer is used to input the No. 3 sequence data, output the third hidden state to the first feature fusion layer and the second output layer, the second output layer outputs an indication of the influence of the electric arc on the maintenance personnel, the fourth hidden layer is used to input the No. 2 graph structure data, output the fourth hidden state to the first feature fusion layer, the first feature fusion layer outputs the first fusion state to the fifth hidden layer, the fifth hidden layer outputs the fifth hidden state to the third output layer, and the third output layer outputs an indication of the probability of each tool falling;

[0033] Step 700, when it is determined that a tool drop may occur, an early warning is issued to all maintenance personnel.

[0034] In a preferred embodiment, the loss function during arc prediction model training is as follows:

[0035] ;

[0036] in represents the first loss value, represents the total number of training samples, Indicates The actual arc temperature, light intensity and duration of the samples, The output of the model is The temperature, light intensity and duration of the arc for each sample.

[0037] In a preferred embodiment, the loss function during training of the drop prediction model is as follows:

[0038] ;

[0039] in represents the second loss value, represents the total number of training samples, It is The true classification label of a training sample has a value of 0 or 1, where 1 means the tool represented by the node has fallen, and 0 means the tool represented by the node has not fallen. is the first predicted by the drop prediction model The classification labels of training samples, Represents the logarithmic function with the natural constant e as the base.

[0040] ;

[0041] ;

[0042] in represents the second loss value, represents the third loss value, represents the total loss value, represents the total number of training samples, Indicates The actual posture and heart rate of maintenance personnel of the sample, The output of the model is The posture and heart rate of each sample.

[0043] In a preferred embodiment, when the probability of a tool falling is greater than a threshold, it is determined that the tool will fall, and the default value of the threshold is 0.5.

[0044] The beneficial effect of the present invention is that the present invention predicts the generation of an arc, and then predicts the falling condition of each tool according to the possible impact of the generated arc on maintenance personnel and tools, so as to provide better early warning. BRIEF DESCRIPTION OF THE DRAWINGS

[0045] Figure 1It is a flow chart of the intelligent monitoring method of the power operation environment based on deep learning of the present invention. DETAILED DESCRIPTION

[0046] The subject matter described herein will now be discussed with reference to example implementations. It should be understood that the discussion of these implementations is only to enable those skilled in the art to better understand and implement the subject matter described herein, and the functions and arrangements of the elements discussed may be changed without departing from the scope of protection of the present specification. Various examples may omit, replace, or add various processes or components as needed. In addition, the features described in some examples may also be combined in other examples.

[0047] At least one embodiment of the present invention discloses a method for intelligent monitoring of power operation environment based on deep learning, such as Figure 1 As shown, including:

[0048] Step 100, collecting comprehensive information including the current time point and a total of n consecutive time points before, the comprehensive information including route information, environment information and tool information;

[0049] Line information includes: voltage, current, and temperature of the maintenance line;

[0050] Environmental information includes: wind speed, wind direction, temperature, humidity, precipitation, light intensity, and electromagnetic field intensity;

[0051] Tool information includes: melting point, thermal decomposition temperature and thermal conductivity of each tool, location, usage status, and distance between the tool and the line;

[0052] In one embodiment of the present invention, the tools include: an insulated screwdriver, an insulated pliers, a multimeter, a voltage detector, a phase sequence meter, a wrench, a cable stripper, and the like.

[0053] In one embodiment of the present invention, the use state of the tool includes on and off.

[0054] Step 200, encoding the comprehensive information into No. 1 graph structure data, the No. 1 graph structure data includes nodes and edges connecting the nodes, and a node represents a line or environment or tool;

[0055] The relationship between the environment and the line is: the environment affects the line, and there are edges between nodes;

[0056] The relationship between tools and lines is: tools will be in contact with lines, and there will be edges between nodes;

[0057] The relationship between environment and tools is: the environment affects the use of tools, and there are edges between nodes;

[0058] The node characteristics of the node representing the environment are equal to the environment characteristics, the node characteristics of the node representing the line are equal to the line characteristics, and the node characteristics of the node representing the tool are equal to the tool characteristics;

[0059] Environmental features, line features, and tool features are obtained through feature engineering;

[0060] Encode the comprehensive information of each node into No. 1 sequence data, the No. 1 sequence data includes n sequence units, and the t-th sequence unit represents No. 1 graph structure data generated by the comprehensive information of all nodes at the t-th moment;

[0061] Step 300, inputting the first sequence data into an arc prediction model, the arc prediction model comprising a first hidden layer, a second hidden layer and a first output layer, wherein the first hidden layer is used to input the first sequence data, output the first hidden feature to the second hidden layer, the second hidden layer outputs the second hidden layer to the first output layer, and the first output layer outputs information indicating the generated arc;

[0062] In one embodiment of the present invention, the calculation formula of the first hidden layer is as follows:

[0063] ;

[0064] in The first hidden state of the node of the vth node of the tth sequence unit of the first sequence data, represents the aggregation coefficient of the vth node of the tth sequence unit of the first sequence data, represents the set of nodes that have edges with the vth node of the tth sequence unit of the first sequence data, represents the activation function, represents the graph weight coefficient;

[0065] ;

[0066] ;

[0067] ;

[0068] in and They represent the node features of the vth and uth nodes of the tth sequence unit of the first sequence data, respectively. and They represent the linear transformation features of the vth and uth nodes of the tth sequence unit of the first sequence data, respectively. represents the aggregation weight coefficient, represents the splicing weight coefficient, T represents transposition, represents an exponential function with a natural constant as the base, LeakyRelu represents a modified linear unit function, Represents the set of nodes that have edges with the vth node.

[0069] In one embodiment of the present invention, the calculation formula of the second hidden layer is as follows:

[0070] ;

[0071] ;

[0072] ;

[0073] ;

[0074] in, , , represents the first, second and third sequence weight parameters, , , represents the first, second and third sequence bias parameters, represents the dot product, , and Represent the first, second, and third intermediate states respectively, where , Represents the second hidden state of the vth node of the tth sequence unit of the first sequence data, represents the concatenation of the second hidden states of all nodes of the t-th sequence unit, N represents the set of all nodes, and They represent the t-th and t-1-th second hidden states respectively, n≥t≥1, n represents the total number of sequence units of the first sequence data, and when t=1 , tanh is the hyperbolic tangent function, Represents a sigmoid function.

[0075] In one embodiment of the present invention, the calculation formula of the first output layer is as follows:

[0076] ;

[0077] in represents a vector whose values ​​represent the predicted arc temperature, light intensity and duration, respectively. represents the t-th second hidden state, represents the concatenation of the 1st to nth second hidden states, n represents the total number of sequence units of the first sequence data, is the first weight parameter, is the first bias parameter, Represents a sigmoid function.

[0078] Step 400, collecting maintenance personnel impact information and tool impact information including the current time point and a total of n consecutive time points before, the maintenance personnel impact information includes: maintenance personnel information, environmental information, and generated arc information, and the tool impact information includes: tool information, environmental information, and generated arc information;

[0079] Maintenance personnel information includes: maintenance personnel’s posture and heart rate;

[0080] In one embodiment of the present invention, the posture of the maintenance personnel includes all possible postures of the maintenance personnel, specifically including: standing, squatting, bending over, etc.

[0081] In one embodiment of the present invention, the generated arc information is the result of the output of the arc prediction model, including: the temperature, light intensity and duration of the arc.

[0082] Step 500, encode the tool impact information into No. 2 graph structure data, the No. 2 graph structure data includes nodes and edges connecting nodes, and one node represents a tool or environment or arc; encode the maintenance personnel impact information into No. 3 sequence data, the No. 3 sequence data includes n sequence units, and the t-th sequence unit represents the maintenance personnel impact information at the t-th moment;

[0083] The relationship between environment and tools is: the environment affects the use of tools, and there are edges between nodes;

[0084] The association between tools and arcs is: arcs affect the use of tools, and there are edges between nodes;

[0085] The relationship between the environment and arcs is: the environment affects the generation of arcs, and there are edges between nodes;

[0086] The node characteristics of the node representing the environment are equal to the environment characteristics, the node characteristics of the node representing the arc are equal to the arc characteristics, and the node characteristics of the node representing the tool are equal to the tool characteristics;

[0087] Environmental features, arc features, and tool features are obtained through feature engineering;

[0088] The tool impact information of each node is encoded into second sequence data, the second sequence data includes n sequence units, and the t-th sequence unit represents the tool impact information of a node at the t-th moment;

[0089] Step 600, inputting the No. 2 graph structure data and the No. 3 sequence data into a drop prediction model, the drop prediction model comprising a third hidden layer, a fourth hidden layer, a fifth hidden layer, a first feature fusion layer, a second output layer, and a third output layer, wherein the third hidden layer is used to input the No. 3 sequence data, output the third hidden state to the first feature fusion layer and the second output layer, the second output layer outputs an indication of the influence of the electric arc on the maintenance personnel, the fourth hidden layer is used to input the No. 2 graph structure data, output the fourth hidden state to the first feature fusion layer, the first feature fusion layer outputs the first fusion state to the fifth hidden layer, the fifth hidden layer outputs the fifth hidden state to the third output layer, and the third output layer outputs an indication of the probability of each tool falling;

[0090] In one embodiment of the present invention, the calculation formula of the third hidden layer is as follows:

[0091] ;

[0092] in represents the tth third hidden state, represents the t-1th third hidden state, represents the tth sequence unit of the No. 3 sequence data, and are the first and second sequence weight parameters, is the sequence bias parameter, and tanh is the hyperbolic tangent function.

[0093] In one embodiment of the present invention, the calculation formula of the second output layer is as follows:

[0094] ;

[0095] in represents a vector, the values ​​of which represent the posture and heart rate of the maintenance personnel respectively. represents the tth third hidden state, represents the concatenation of the 1st to nth third hidden states, n represents the total number of sequence units of the second sequence data, is the second weight parameter, is the second bias parameter, Represents a sigmoid function.

[0096] In one embodiment of the present invention, the calculation formula of the fourth hidden layer is as follows:

[0097] The long short-term memory network includes a forget gate, an input gate, a cell state, and an output gate. The tth output of the forget gate of the vth node of the second graph structure data for:

[0098] ;

[0099] in, is the weight matrix of the forget gate, is the bias term of the forget gate, It is the t-1th fourth hidden state of the vth node of the second graph structure data and The splicing, is the sigmoid function; Indicates the tth data item of the second sequence data corresponding to the vth node of the second graph structure data;

[0100] The tth output of the input gate of the vth node of the graph structure data No. 2 The tth candidate cell state of the vth node of the second graph structure data The calculation method is:

[0101] ;

[0102] ;

[0103] in, is the weight matrix of the input gate, is the bias term of the input gate, is the hyperbolic tangent activation function, is the weight matrix for calculating candidate cell states, is the bias term of the candidate cell state;

[0104] The tth cell state of the vth node of the graph structure data No. 2 The t-1th cell state of the vth node of the second graph structure data and Weighted synthesis, the calculation method is:

[0105] ;

[0106] in, Represents element-by-element multiplication operation;

[0107] The tth activation value of the output gate of the vth node of the graph structure data No. 2 The tth fourth hidden state of the vth node of the second graph structure data The calculation method is:

[0108] ;

[0109] ;

[0110] in, is the weight matrix of the output gate, is the bias term of the output gate

[0111] In one embodiment of the present invention, the calculation formula of the first feature fusion layer is as follows:

[0112] ;

[0113] represents the first fusion state of the vth node, represents the nth fourth hidden state of the vth node of the graph structure data No. 2, represents the nth third hidden state, represents the concatenation function, represents the sum weight matrix, represents the summation bias parameter, Represents a sigmoid function.

[0114] In one embodiment of the present invention, the calculation formula of the fifth hidden layer is as follows:

[0115] ;

[0116] in represents the fifth hidden state of the vth node, represents the graph weight parameter, , , , Respectively represent the first fusion state of the vth and uth nodes, is the set of nodes that have edges with node v of the second graph structure data. express The total number of nodes in ;

[0117] ;

[0118] in represents the aggregation coefficient between the vth and uth nodes, tanh represents the hyperbolic tangent function, T represents the transpose, Indicates adjustable parameters; The default value is 1 / N, where N is the dimension of the node features.

[0119] In one embodiment of the present invention, the calculation formula of the third output layer is as follows:

[0120] ;

[0121] in represents a vector, the i-th component of the vector represents the drop probability of the i-th tool, represents the tth fifth hidden state, represents the concatenation of the fifth hidden states of all nodes, M represents the set of all nodes, is the third weight parameter, is the third bias parameter, Represents a sigmoid function.

[0122] In one embodiment of the present invention, when the probability of a tool falling is greater than a threshold, it is determined that the tool will fall, and the default value of the threshold is 0.5.

[0123] Step 700, when it is determined that a tool drop may occur, an early warning is issued to all maintenance personnel.

[0124] In one embodiment of the present invention, the early warning method is to use each person's communication device to issue a voice alarm and a reminder of tool drop.

[0125] In one embodiment of the present invention, the loss function during arc prediction model training is as follows:

[0126] ;

[0127] in represents the first loss value, represents the total number of training samples, Indicates The actual arc temperature, light intensity and duration of the samples, The output of the model is The temperature, light intensity and duration of the arc for each sample.

[0128] In one embodiment of the present invention, the loss function during training of the drop prediction model is as follows:

[0129] ;

[0130] in represents the second loss value, represents the total number of training samples, It is The true classification label of a training sample has a value of 0 or 1, where 1 means the tool represented by the node has fallen, and 0 means the tool represented by the node has not fallen. is the first predicted by the drop prediction model The classification labels of training samples, Represents the logarithmic function with the natural constant e as the base.

[0131] ;

[0132] ;

[0133] in represents the second loss value, represents the third loss value, represents the total loss value, represents the total number of training samples, Indicates The actual posture and heart rate of maintenance personnel of the sample, The output of the model is The posture and heart rate of each sample.

[0134] The above describes an embodiment of the present invention, but this embodiment is not limited to the above-mentioned specific implementation mode. The above-mentioned specific implementation mode is merely illustrative and not restrictive. Under the guidance of this embodiment, ordinary technicians in this field can also make more forms of equivalent embodiments, all of which are within the protection of this embodiment.

Claims

1. A method for intelligent monitoring of power operation environment based on deep learning, characterized in that: The following steps are involved: Step 100, collecting comprehensive information including the current time point and a total of n consecutive time points before, the comprehensive information including route information, environment information and tool information; Step 200, encoding the comprehensive information into No. 1 graph structure data, the No. 1 graph structure data includes nodes and edges connecting the nodes, one node represents a line or environment or tool, and encoding the comprehensive information of each node into No. 1 sequence data; Step 300, inputting the first sequence data into the arc prediction model, and outputting information indicating the generated arc; Step 400, collecting maintenance personnel impact information and tool impact information including the current time point and a total of n consecutive time points before, the maintenance personnel impact information includes: maintenance personnel information, environmental information, and generated arc information, and the tool impact information includes: tool information, environmental information, and generated arc information; Step 500, encode the tool influence information into No. 2 graph structure data, the No. 2 graph structure data includes nodes and edges connecting nodes, and one node represents a tool or environment or arc; encode the tool influence information of each node into No. 2 sequence data; encode the maintenance personnel influence information into No. 3 sequence data; Step 600, inputting the second graph structure data and the third sequence data into the drop prediction model, and outputting the probability of each tool being dropped; Step 700, when it is determined that a tool drop may occur, an early warning is issued to all maintenance personnel.

2. According to claim 1, a method for intelligent monitoring of power operation environment based on deep learning is characterized in that: Line information includes: voltage, current, and temperature of the maintenance line; Environmental information includes: wind speed, wind direction, temperature, humidity, precipitation, light intensity, and electromagnetic field intensity; Tool information includes: melting point, thermal decomposition temperature and thermal conductivity of each tool, location, usage status, and distance between the tool and the line; Maintenance personnel information includes: maintenance personnel’s posture and heart rate; The arc information generated includes: arc temperature, light intensity and duration.

3. The method for intelligent monitoring of power operation environment based on deep learning according to claim 1 is characterized in that: The No. 1 sequence data in step 200 includes n sequence units, and the t-th sequence unit represents the No. 1 graph structure data generated by the comprehensive information of all nodes at the t-th moment. The relationship between the environment and the line is: the environment affects the line, and there are edges between nodes; The relationship between tools and lines is: tools will be in contact with lines, and there will be edges between nodes; The relationship between environment and tools is: the environment affects the use of tools, and there are edges between nodes; The node characteristics of the node representing the environment are equal to the environment characteristics, the node characteristics of the node representing the line are equal to the line characteristics, and the node characteristics of the node representing the tool are equal to the tool characteristics; Environmental features, line features, and tool features are obtained through feature engineering.

4. The method for intelligent monitoring of power operation environment based on deep learning according to claim 1, characterized in that: The second sequence data in step 500 includes n sequence units, and the tth sequence unit represents the tool influence information of a node at the tth moment; the third sequence data includes n sequence units, and the tth sequence unit represents the maintenance personnel influence information at the tth moment; The relationship between environment and tools is: the environment affects the use of tools, and there are edges between nodes; The association between tools and arcs is: arcs affect the use of tools, and there are edges between nodes; The relationship between the environment and arcs is: the environment affects the generation of arcs, and there are edges between nodes; The node characteristics of the node representing the environment are equal to the environment characteristics, the node characteristics of the node representing the arc are equal to the arc characteristics, and the node characteristics of the node representing the tool are equal to the tool characteristics; Environmental features, arc features, and tool features are obtained through feature engineering.

5. The method for intelligent monitoring of power operation environment based on deep learning according to claim 1 is characterized in that: The arc prediction model includes a first hidden layer, a second hidden layer and a first output layer, wherein the first hidden layer is used to input a first sequence data, output a first hidden feature to the second hidden layer, the second hidden layer outputs the second hidden layer to the first output layer, and the first output layer outputs information representing the generated arc.

6. The method for intelligent monitoring of power operation environment based on deep learning according to claim 1, characterized in that: The drop prediction model includes a third hidden layer, a fourth hidden layer, a fifth hidden layer, a first feature fusion layer, a second output layer, and a third output layer, wherein the third hidden layer is used to input sequence data No. 3, output the third hidden state to the first feature fusion layer and the second output layer, the second output layer outputs an indication of the impact of the electric arc on maintenance personnel, the fourth hidden layer is used to input graph structure data No. 2, output the fourth hidden state to the first feature fusion layer, the first feature fusion layer outputs the first fusion state to the fifth hidden layer, the fifth hidden layer outputs the fifth hidden state to the third output layer, and the third output layer outputs an indication of the probability of each tool falling.

7. The method for intelligent monitoring of power operation environment based on deep learning according to claim 1, characterized in that: The loss function when training the arc prediction model is as follows: ； in represents the first loss value, represents the total number of training samples, Indicates The actual arc temperature, light intensity and duration of the samples, The output of the model is The temperature, light intensity and duration of the arc for each sample.

8. The method for intelligent monitoring of power operation environment based on deep learning according to claim 1, characterized in that: The loss function when training the drop prediction model is as follows: ； in represents the second loss value, represents the total number of training samples, It is The true classification label of a training sample has a value of 0 or 1, where 1 means the tool represented by the node has fallen, and 0 means the tool represented by the node has not fallen. is the first predicted by the drop prediction model The classification labels of training samples, It represents the logarithmic function with the natural constant e as the base; ； ； in represents the second loss value, represents the third loss value, represents the total loss value, represents the total number of training samples, Indicates The actual posture and heart rate of maintenance personnel of the sample, The output of the model is The posture and heart rate of each sample.

9. The method for intelligent monitoring of power operation environment based on deep learning according to claim 1, characterized in that: When the probability of a tool falling is greater than a threshold, it is determined that the tool will fall. The default value of this threshold is 0.

5.

10. An intelligent monitoring system for power operation environment based on deep learning, characterized in that: It is used to execute a deep learning-based intelligent monitoring method for an electric power operation environment as described in any one of claims 1-9.

Images