Power distribution station house emergency practical training evaluation method and system based on big data acquisition

Abstract

The invention relates to the technical field of practical training evaluation, and discloses a power distribution station house emergency practical training evaluation method and system based on big data acquisition, and the method comprises the steps: constructing a standard emergency emergency operation process in a water immersion scene of a power distribution station house, and decomposing the standard emergency emergency operation process into a plurality of groups of key emergency action units; and carrying out matching analysis on the behavior track data and the key action template, extracting track fragment sub-data containing a key emergency action unit, carrying out practical training comparative analysis, and generating a multi-dimensional practical training evaluation result for practical training feedback. According to the method, a standard emergency rescue operation process is constructed, a key action template is formed, behavior track data of training personnel is collected, track fragment sub-data is extracted for training comparative analysis based on matching analysis of the track data and the key action template, a corresponding evaluation result is obtained, and finally multi-dimensional training evaluation feedback is formed.

Description

Technical Field

[0001] The present invention relates to the technical field of big data collection, in particular to the technical field of training evaluation, and specifically to a method and system for evaluating emergency training of power distribution stations based on big data collection. Background Art

[0002] During the operation of distribution substations, extreme weather or natural disasters such as heavy rainfall and flooding can easily cause waterlogging or even flooding. These emergencies pose a serious threat to the safe operation of distribution equipment and the lives of personnel. Therefore, various emergency response plans and training systems have become core tools for improving the power industry's emergency response capabilities. Emergency response training not only covers the complete rescue process but also assesses trainees' compliance with regulations, timeliness, and effectiveness in realistic simulated scenarios, thereby testing their practical rescue capabilities.

[0003] The existing evaluation methods for emergency training are still mainly based on manual observation and scoring, supplemented by simple video playback for qualitative analysis. They have significant limitations such as strong subjectivity, delayed feedback, and a single analysis dimension, making it difficult to form a quantifiable, structured, and systematic evaluation and feedback mechanism. Some studies have attempted to introduce big data and AI technology to conduct intelligent analysis of training results. For example, patent CN115841403B discloses an AI training method and system based on big data. It screens multi-module test questions from AI teaching big data, sends test question data to the student terminal device, and conducts a comprehensive analysis based on the completion status to output the learning progress evaluation results, thereby forming a personalized teaching and training plan. This type of method shows good adaptability and intelligent recommendation capabilities in teaching and training tasks, but it still focuses on cognitive level assessment based on static answer data, and lacks accurate description of the dynamic action execution process, the timing logic of operation behavior, and the physical interaction process.

[0004] Furthermore, traditional power emergency drill systems often rely on a combination of video recording and on-site supervision for evaluation. This is limited by a lack of structured representation of behavioral trajectories, recognition of key actions, and real-time feedback, making it difficult to quantitatively evaluate key actions and optimize training effectiveness. Currently, there is a lack of a systematic, granular, multi-dimensional, and accurate feedback-based evaluation method for emergency rescue training in substation flooding scenarios. In particular, there is a lack of a mechanism for in-depth mining of trainee behavioral trajectory data and action-level comparison with standard process templates, which limits the upgrading of training systems towards intelligent, standardized, and precise development.

[0005] Therefore, there is an urgent need for a distribution station rescue training evaluation method based on big data collection that integrates multi-dimensional key scoring factors and action trajectory data, so as to realize structured modeling of the training operation process, key action matching extraction and quantitative evaluation analysis, and improve the intelligence, pertinence and feedback effectiveness of the training evaluation. Summary of the Invention

[0006] In view of this, the present invention provides a distribution station emergency rescue training evaluation method and system based on big data collection. By constructing a key action template applied to the emergency response of distribution station flooding, the behavioral trajectories of trainees are collected and matched and analyzed, key action fragments are extracted and evaluated, and the quality of action execution in the training process is quantitatively identified. A feedback mechanism is formed by combining multi-dimensional scoring results, which effectively improves the scientificity and intelligence level of emergency rescue training.

[0007] To achieve the above objectives, the present invention provides a method for evaluating emergency training for power distribution stations based on big data collection, comprising the following steps:

[0008] S1: Construct a standard emergency rescue operation process for the substation flooding scenario, decompose the standard emergency rescue operation process into several groups of key rescue action units, and form a key action template from the key rescue action units;

[0009] S2: Collect the behavioral trajectory data of trainees during the training process;

[0010] S3: Matching and analyzing the behavior trajectory data with the key action templates to extract the trajectory segment sub-data containing the key rescue action units;

[0011] S4: Conducting a practical training comparative analysis on the trajectory segment sub-data and the key rescue action units contained therein to obtain a practical training comparative evaluation result of the trajectory segment sub-data;

[0012] S5: Comprehensively compare and evaluate the training results of all trajectory segment sub-data, and generate multi-dimensional training evaluation results for training feedback.

[0013] As a further improvement method of the present invention:

[0014] Optionally, the standard emergency rescue operation process is a directed graph structure, including nodes, edges and node attributes, including:

[0015] The nodes are operation units in the rescue process, and the operation units describe specific rescue operation behaviors. The edges represent the operation timing logic or conditional jumps between nodes. The node attributes include the tools required to execute the node, the action category, the execution object, the time expectation, and the key attributes of the node.

[0016] The key attributes include the node's hazard sensitivity, path turning degree, and recognition accuracy. The hazard sensitivity indicates the level of safety risk that may be caused if the rescue operation behavior corresponding to the node is executed incorrectly. The path turning degree indicates the turning role of the node in the entire standard emergency rescue operation process. The recognition accuracy indicates whether the rescue operation behavior corresponding to the node can be accurately identified.

[0017] Optionally, based on the key attributes of the nodes, the standard emergency rescue operation process is decomposed into several groups of key rescue action units, including:

[0018] The key attributes of the nodes are normalized as the key scoring vector of the nodes, which includes the hazard sensitivity score, path turning score and recognition accuracy score. Nodes whose hazard sensitivity score, path turning score and recognition accuracy score are higher than the preset hazard sensitivity threshold, path turning threshold and recognition accuracy threshold, respectively, are selected, and the node names and node attributes are extracted as a set of key rescue action units.

[0019] Optionally, the behavior trajectory data is composed of a plurality of continuous behavior sequence points, each behavior sequence point including the trainee's position, action posture, operation object and timestamp information, including:

[0020] The trainee's position, movement posture and operation object are obtained by monitoring sensors, wherein the sensor types include cameras, sound sensors and positioning devices. The positioning device is used to monitor the trainee's position under timestamp information, and the camera is used to monitor images in different scenes. Based on the trainee's position under timestamp information, the scene image of the trainee in the timestamp information and the position is obtained, and the trainee's movement posture and operation object are extracted from the scene image. The trainee's position, the extracted trainee's movement posture, operation object and timestamp information in the scene image are used as behavior sequence points.

[0021] Optionally, matching and analyzing the behavior trajectory data with key action templates includes:

[0022] Preprocessing the behavior sequence points in the behavior trajectory data, wherein the preprocessing method includes removing the behavior sequence points that are completely identical except for the timestamp information, to obtain preprocessed behavior trajectory data;

[0023] The pre-processed behavior trajectory data is processed in a sliding window form to obtain multiple sliding window sequences, where each sliding window sequence consists of multiple behavior sequence points. The timestamp information difference between the last behavior sequence point and the first behavior sequence point in the sliding window sequence is Len, where Len is the length of the sliding window and step is the step size of the sliding window.

[0024] Calculate the matching coefficient between the sliding window sequence and the key rescue action unit in the key action template, and use the key rescue action unit with a matching coefficient higher than a preset coefficient threshold as the matching result of the sliding window sequence. Use the sliding window sequence with the matching result as the trajectory segment sub-data, and use the matched key rescue unit as the inclusion value of the trajectory segment sub-data. The matching coefficient is calculated as follows:

[0025] Sim(S,unit)=w1·Sim1(S,unit)+w2·Sim2((S,unit)+w3·Sim3(S,unit);

[0026] Among them, Sim(S, unit) represents the matching coefficient between the sliding window sequence S and the key rescue action unit unit, Sim1(S, unit) represents the action similarity between the sliding window sequence S and the key rescue action unit unit, Sim2(S, unit) represents the tool matching degree between the sliding window sequence S and the key rescue action unit unit, and Sim3(S, unit) represents the execution object matching degree between the sliding window sequence S and the key rescue action unit unit;

[0027] w1, w2, and w3 all represent matching coefficient weights. Set w1, w2, and w3 to 0.5, 0.3, and 0.2 respectively.

[0028] Optionally, the trajectory segment sub-data is subjected to a practical training comparative analysis with the key rescue action units contained therein to obtain a practical training comparative evaluation result of the trajectory segment sub-data, including:

[0029] The calculation method of the training comparison evaluation results of the trajectory segment sub-data is:

[0030]

[0031] Among them, Score unit_n The trajectory segment sub-data S containing the nth key rescue action unit unit_n unit_n The training comparison evaluation results of [Sim1(S unit_n ,unit_n),Sim2(S unit_n ,unit_n),Sim3((S unit_n , unit_n) are the trajectory segment sub-data S unit_n The action similarity, tool matching and execution object matching between the key rescue action unit unit_n, Represents the trajectory segment sub-data S unit_nThe time matching degree between the key emergency action unit unit_n, exp((·) represents the exponential function with the natural constant as the base, σ represents the time control parameter, set σ to 2, time(S unit_n ) represents the trajectory segment sub-data S unit_n Length of time, time unit_n Indicates the time expectation in the key rescue action unit unit_n, max time Indicates the normalized control parameter, set max time 90 seconds;

[0032] n∈[1,N], where N represents the number of key rescue action units.

[0033] Optionally, the training comparative evaluation results of all trajectory segment sub-data are integrated to generate a training evaluation result, including:

[0034] The mean action similarity, tool matching, execution object matching, and time matching between N key rescue action units and the associated trajectory segment sub-data are calculated as the training evaluation results of the trainees in terms of operation completion, tool usage correctness, object positioning accuracy, and time execution rationality.

[0035] In order to solve the above problems, the present invention provides a distribution substation emergency training and evaluation system based on big data collection, which includes a server and a training process decomposition device. The server includes a trajectory matching module and a training evaluation module:

[0036] The trajectory matching module is used to collect the trainee's behavioral trajectory data during the training process, match and analyze the behavioral trajectory data with the key action template, and extract the trajectory segment sub-data containing the key rescue action unit;

[0037] The training evaluation module is used to perform training comparative analysis on the trajectory segment sub-data and the key rescue action units contained therein, obtain the training comparative evaluation results of the trajectory segment sub-data, and integrate the training comparative evaluation results of all trajectory segment sub-data to generate multi-dimensional training evaluation results for training feedback;

[0038] The training process decomposition device is used to construct a standard emergency rescue operation process in a distribution station flooding scenario, and decompose the standard emergency rescue operation process into several groups of key rescue action units, and the key rescue action units form a key action template;

[0039] In order to realize the above-mentioned distribution station emergency training evaluation method based on big data collection.

[0040] In order to solve the above problem, the present invention provides an electronic device, comprising:

[0041] a memory storing at least one instruction;

[0042] Communication interfaces to enable electronic equipment to communicate; and

[0043] The processor executes the instructions stored in the memory to implement the above-mentioned distribution station emergency training evaluation method based on big data collection.

[0044] In order to solve the above problems, the present invention also provides a computer-readable storage medium, which stores at least one instruction, and the at least one instruction is executed by a processor in an electronic device to implement the above-mentioned distribution station emergency training and evaluation method based on big data collection.

[0045] Compared with the existing technology, the present invention proposes a method and system for evaluating emergency training in power distribution stations based on big data collection. This technology has the following advantages:

[0046] First, this application constructs a standard emergency rescue operation process for flooding scenarios, and combines multi-dimensional indicators such as node hazard sensitivity, path turning degree and recognition accuracy to scientifically decompose a series of key rescue action units. This decomposition not only retains the core links of the task, but also enhances the pertinence and accuracy of subsequent identification and evaluation. The key rescue action units are standardized into templates, covering elements such as action sequence, required tools, execution objects and time expectations, to achieve structured and alignable modeling of rescue behaviors.

[0047] At the same time, a comparative analysis of practical training is conducted between the trajectory fragment sub-data and the key rescue action units, and a comparative analysis is conducted based on the action, tool, object and time dimensions respectively, providing high-credibility support for subsequent multi-dimensional feedback. This step not only has the ability to diagnose action-level differences, but also can clearly point out the specific source of the problem (such as incorrect selection of operation objects). The practical training comparison evaluation results of all trajectory fragment sub-data are attributed and aggregated to generate multi-dimensional practical training scores such as operation completion, correctness of tool use, object positioning accuracy and rationality of time execution, realizing an accurate and explainable practical training feedback mechanism. This feedback can not only be used to evaluate the effect of a single training session, but also support progress tracking of multiple rounds of training and individual ability modeling. BRIEF DESCRIPTION OF THE DRAWINGS

[0048] Figure 1 A flowchart of a method for emergency training and evaluation of a power distribution station based on big data collection is provided in accordance with an embodiment of the present invention.

[0049] Figure 2 A flowchart of a standard emergency rescue process training and evaluation provided by one embodiment of the present invention.

[0050] The purpose, features and advantages of the present invention will be further described with reference to the accompanying drawings and in conjunction with the embodiments. DETAILED DESCRIPTION

[0051] It should be understood that the specific embodiments described herein are only used to explain the present invention and are not intended to limit the present invention.

[0052] The embodiment of the present application provides a method for evaluating the emergency training of distribution substations based on big data collection. The execution subject of the method for evaluating the emergency training of distribution substations based on big data collection includes but is not limited to at least one of the electronic devices such as a server and a terminal that can be configured to execute the method provided by the embodiment of the present application. In other words, the method for evaluating the emergency training of distribution substations based on big data collection can be executed by software or hardware installed on a terminal device or a server device, and the software can be a blockchain platform. The server includes but is not limited to: a single server, a server cluster, a cloud server or a cloud server cluster, etc.

[0053] Reference Figure 1 , embodiment 1 of the present invention is:

[0054] A method for evaluating power distribution station emergency training based on big data collection includes the following steps:

[0055] S1: Construct a standard emergency rescue operation process for the distribution station flooding scenario, decompose the standard emergency rescue operation process into several groups of key rescue action units, and form a key action template from the key rescue action units.

[0056] The standard emergency rescue operation process is a directed graph structure, including nodes, edges and node attributes, including:

[0057] The nodes are operation units in the rescue process, and the operation units describe specific rescue operation behaviors. The edges represent the operation timing logic or conditional jumps between nodes. The node attributes include the tools required to execute the node, the action category, the execution object, the time expectation, and the key attributes of the node.

[0058] The key attributes include the node's hazard sensitivity, path turning degree, and recognition accuracy. The hazard sensitivity indicates the level of safety risk that may be caused if the rescue operation behavior corresponding to the node is executed incorrectly. The path turning degree indicates the turning role of the node in the entire standard emergency rescue operation process. The recognition accuracy indicates whether the rescue operation behavior corresponding to the node can be accurately identified.

[0059] Specifically, the key attributes of the node are:

[0060]

[0061] in, The node node represents the danger sensitivity, path turning degree, and recognition accuracy of the node. The danger sensitivity of the node is evaluated and calculated based on a historical risk assessment method. The historical risk assessment method obtains the historical rescue results of the node, calculates the frequency of occurrence of different safety risk levels after execution errors, and performs weighted calculation to obtain the danger sensitivity. As an embodiment of the present invention, the danger sensitivity of the node node is calculated as follows:

[0062]

[0063] in, Indicates the frequency of occurrence of the mth safety risk level after the rescue operation behavior corresponding to node node is executed incorrectly in the historical rescue results of node node. The safety risk levels are numbered from 1 to M, with M set to 4. The higher the safety risk level, the more serious the safety accident that may be caused.

[0064] Deg out (node) represents the outgoing edge degree of node node in the standard emergency rescue operation process, that is, the number of nodes that node node can jump to, Deg in (node) represents the in-edge degree of node node in the standard emergency rescue operation process, which means the number of nodes that can jump to node node. Both represent the weights for calculating the inbound and outbound edges.

[0065] dim node Indicates the number of sensors required to identify the emergency operation behavior at node node, where sensor types include cameras, sound sensors, and positioning devices. represents the recognition accuracy of the sth sensor deployed at node node, where the recognition accuracy of the sensor is calculated based on the test data;

[0066] As a preferred algorithm of this application, by constructing a standard emergency rescue operation process under the distribution station flooding scenario, integrating the three indicators of node danger sensitivity, path turning degree and recognition accuracy, a method for accurately selecting key rescue action units is proposed. Compared with the traditional method of relying on expert experience or empirical decomposition, this method can quantify the safety risks and behavioral stability that may be caused by nodes based on historical rescue results, node topology characteristics and multi-source sensor recognition data, and effectively improve the coverage and practicality of the key action template library.

[0067] Specifically, the hazard sensitivity index comprehensively considers the safety level and historical frequency of erroneous operations that may be caused by node operation errors, so that high-risk nodes are preferentially selected as key rescue action units; the path turning degree measurement standard determines the decision-making change position of the node in the emergency rescue operation process. Nodes with high turning degrees are often key control points in the operation process; recognition accuracy reflects the stability of the node's corresponding operation in sensor recognition, ensuring the accuracy of subsequent action extraction and evaluation. The three together construct a multi-dimensional evaluation mechanism for key action judgment, effectively avoiding the problem of low-risk or unstable operation fragments being mistakenly selected as key rescue action units.

[0068] Based on the key attributes of the nodes, the standard emergency rescue operation process is decomposed into several groups of key rescue action units, including:

[0069] The key attributes of the nodes are normalized as the key scoring vector of the nodes, which includes the hazard sensitivity score, path turning score and recognition accuracy score. Nodes whose hazard sensitivity score, path turning score and recognition accuracy score are higher than the preset hazard sensitivity threshold, path turning threshold and recognition accuracy threshold, respectively, are selected, and the node names and node attributes are extracted as a set of key rescue action units.

[0070] Specifically, the hazard sensitivity score, path turning score, and recognition accuracy score are normalized results of hazard sensitivity, path turning, and recognition accuracy, respectively. The preset hazard sensitivity threshold, path turning threshold, and recognition accuracy threshold are 0.6, 0.4, and 0.5, respectively.

[0071] Specifically, the key action template consists of several groups of key rescue action units, each group of key rescue action units describes the name of the rescue operation behavior, the action sequence of the rescue operation behavior, the tools required to perform the rescue operation behavior, the execution object of the rescue operation behavior, the time expectation for performing the rescue operation behavior, and the key scoring vector corresponding to the rescue operation behavior.

[0072] S2: Collect behavioral trajectory data of trainees during the training process.

[0073] The behavior trajectory data consists of multiple continuous behavior sequence points, each of which includes the trainee's position, action posture, operation object and timestamp information, including:

[0074] The trainee's position, movement posture and operation object are obtained by monitoring sensors, wherein the sensor types include cameras, sound sensors and positioning devices. The positioning device is used to monitor the trainee's position under timestamp information, and the camera is used to monitor images in different scenes. Based on the trainee's position under timestamp information, the scene image of the trainee in the timestamp information and the position is obtained, and the trainee's movement posture and operation object are extracted from the scene image. The trainee's position, the extracted trainee's movement posture, operation object and timestamp information in the scene image are used as behavior sequence points.

[0075] Specifically, the process of extracting the trainee's action posture and operation object from the scene image is as follows:

[0076] A human posture recognition algorithm is used to extract the trainee's human key point skeleton from the scene image, wherein the human key point skeleton is composed of human key points, and the human key points are connected to form a human key point skeleton. The types of human key points include head, shoulders, elbows, wrists, hips, knees, ankles, etc. The connection mode of the human key points includes head connected to shoulders, shoulders connected to elbows, elbows connected to wrists, hips connected to knees, and knees connected to ankles; the human posture recognition algorithm is the OpenPose algorithm;

[0077] Calculate the length between the interconnected human key points in the human key point skeleton, and calculate the angle of the human key point. The angle calculation method is to extract the coordinates of two human key points connected to the human key point of the angle to be measured, form two vectors, and calculate the vector angle as the angle of the human key point of the angle to be measured. As an embodiment of the present invention, the coordinates of the two human key points connected to the human key point at the coordinate (x1, y1) are (x2, y2) and (x3, y3). The two vectors formed are v2 = (x2-x1, y2-y1) and v3 = (x3-x1, y3-y1). The angle of the human key point at the coordinate (x1, y1) is: Where β(x1,y1) represents the angle of the key point of the human body at the coordinate (x1,y1), arccos(·) is the inverse cosine function, and ||·||2 represents the L2 norm;

[0078] The lengths between the interconnected key points of the human body and the angles between the key points of the human body are used as human skeleton feature vectors, and the human skeleton feature vectors are classified using a posture classification model. The classification results are used as the action postures of the trainees, and the classification categories of the action postures include approaching, grasping, inserting, pulling out, turning a valve, pressing, tightening, and loosening;

[0079] The Mask R-CNN model is used to identify scene objects in scene images and calculate the center coordinates of the scene objects. The scene objects are objects that need to be operated during the power distribution station rescue training, including tools and execution objects. The coordinates of the human body key points and scene objects are the two-dimensional pixel coordinates of the human body key points in the scene image.

[0080] Calculate the distance between the coordinates of the wrist in the human body key point skeleton and the center coordinates of the scene object, select the scene object with the closest distance and less than a preset distance threshold as the training personnel's operation object in the scene image, the preset distance threshold is (the average length of the line between the elbow and the wrist in all scene images / 2), the wrist coordinate (x 1 ,y 1 ) and the center coordinates ((x 2 ,y 2 ) is:

[0081]

[0082] Among them, D((x 1 ,y 1 ),(x 2 ,y 2 )) represents the wrist coordinate (x 1 ,y 1 ) and the center coordinate (x 2 ,y 2 ), dis((x 1 ,y 1 ),(x 2 ,y 2 )) represents the wrist coordinate (x 1 ,y 1 ) and the center coordinate (x 2 ,y 2 ), ε represents the direction control parameter, ε is set to 0.5, ||·||2 represents the L2 norm, Indicates the wrist coordinates at the previous timestamp, ((x 2 ,y 2 )-(x 1 ,y 1 )) represents the direction vector of the line connecting the wrist to the scene object, Indicates the direction of wrist movement, c(x 1 ,y 1 ;x 2 ,y 2 ) indicates the consistency between the wrist movement direction and the direction of the scene object, c(x 1 ,y 1 ;x 2 ,y 2) is greater than 0, indicating that the wrist is moving in the direction of the scene object, and the coordinate distance is shortened. 1 ,y 1 ;x 2 ,y 2 ) is less than 0, indicating that the wrist movement direction is away from the direction of the scene object, and the coordinate distance is lengthened;

[0083] The distance between the wrist coordinates and the center coordinates not only considers the geometric proximity between the wrist and the object, but also considers whether the movement is directed towards the scene object, thereby more accurately judging whether the trainee is actually manipulating the scene object;

[0084] Traditional spatial distance only reflects the geometric proximity between a person and a target object, making it difficult to accurately characterize whether a person is engaging in intentional manipulation. For example, when a person is standing or passing by an object, while close, their hand may not be pointing toward it or performing any related actions. In this case, judging manipulation solely based on Euclidean distance can lead to numerous misjudgments. To address this issue, the interactive distance metric, based on geometric distance, incorporates the cosine of the angle between the hand's motion direction and the line connecting the person and the object, creating a semantically enhanced interactive feature.

[0085] Specifically, if the direction of wrist movement is highly consistent with the direction of the line connecting the target object, that is, the cosine of the angle is close to 1, it means that the wrist movement has a clear trend toward the target object. At this time, even if the actual geometric distance is slightly larger, it can still be inferred that there is potential interaction behavior. On the contrary, if the direction is opposite, that is, the cosine of the angle is negative, it means that the hand is moving away from the target object. Even if the physical distance between the two is close, it should be regarded as a non-interactive state.

[0086] This interaction distance metric has several advantages: significantly improving the accuracy of interaction discrimination, especially in scenes with multiple objects densely packed or mistakenly approaching each other, effectively reducing interference misjudgments; enhancing the ability to capture operational intentions, enabling the system to identify real operational behaviors consisting of approach + orientation, rather than simple contact or proximity; and flexibly extending to continuous frame judgment and trajectory modeling, forming an action-object interaction evolution trajectory under temporal evolution, providing high-quality prerequisites for trajectory segment extraction, action matching, and other links.

[0087] S3: Match and analyze the behavior trajectory data with the key action template to extract the trajectory segment sub-data containing the key rescue action unit.

[0088] Matching and analyzing the behavior trajectory data with key action templates includes:

[0089] Preprocessing the behavior sequence points in the behavior trajectory data, wherein the preprocessing method includes removing the behavior sequence points that are completely identical except for the timestamp information, to obtain preprocessed behavior trajectory data;

[0090] The pre-processed behavior trajectory data is processed in a sliding window form to obtain multiple sliding window sequences, where each sliding window sequence consists of multiple behavior sequence points. The timestamp information difference between the last behavior sequence point and the first behavior sequence point in the sliding window sequence is Len, where Len is the length of the sliding window and step is the step size of the sliding window.

[0091] Specifically, according to the expected maximum time of the key rescue action unit, the timestamp information difference is set to the expected maximum time, and the step is set to 10 seconds;

[0092] Calculate the matching coefficient between the sliding window sequence and the key rescue action unit in the key action template, and use the key rescue action unit with a matching coefficient higher than a preset coefficient threshold as the matching result of the sliding window sequence. Use the sliding window sequence with the matching result as the trajectory segment sub-data, and use the matched key rescue unit as the inclusion value of the trajectory segment sub-data. The matching coefficient is calculated as follows:

[0093] Sim(S,unit)=w1·Sim1(S,unit)+w2·Sim2((S,unit)+w3·Sim3(S,unit);

[0094] Among them, Sim(S, unit) represents the matching coefficient between the sliding window sequence S and the key rescue action unit unit, Sim1(S, unit) represents the action similarity between the sliding window sequence S and the key rescue action unit unit, Sim2(S, unit) represents the tool matching degree between the sliding window sequence S and the key rescue action unit unit, and Sim3(S, unit) represents the execution object matching degree between the sliding window sequence S and the key rescue action unit unit;

[0095] w1, w2, w3 represent the matching coefficient weights, and w1, w2, w3 are set to 0.5, 0.3, and 0.2 respectively.

[0096] Specifically, the action similarity Sim1(S, unit) is calculated as follows: extracting action postures from behavior sequence points in the sliding window sequence S to form an action posture sequence, extracting action sequences from key rescue action units unit, and mapping the action sequences to action postures, using action postures to characterize the action sequences to obtain action sequences based on action postures, and using a dynamic time warping algorithm to calculate the minimum distance between the action posture sequences and the action sequences based on action postures, wherein the sequence value categories of the two sequences are consistent with the classification categories of the action postures, and the classification categories of the action postures include approaching, grabbing, inserting, pulling out, turning the valve, pressing, tightening, and loosening;

[0097] Extract the operation objects in the behavior sequence points in the sliding window sequence S to form an operation object sequence. Calculate the overlap between all operation objects in the operation object sequence and the tools required in the key rescue action unit unit, as well as the overlap with the execution objects in the key rescue action unit unit, as the tool matching degree Sim2(S, unit) and the execution object matching degree Sim3(S, unit).

[0098] The calculation method of the overlap of the required tools is: the number of operation objects in the operation object sequence that are the tools required in the key rescue action unit unit / the total number of tools in the operation object sequence;

[0099] The calculation method of the overlap degree of the execution objects is: the number of operation objects in the key emergency action unit unit in the operation object sequence / the total number of execution objects in the operation object sequence;

[0100] In the process of calculating the matching coefficient, the behavior sequence points at both ends of the sliding window sequence are filtered and adjusted, that is, some behavior sequence points are deleted, and the matching coefficient between the adjusted sliding window sequence and the key rescue action unit is recalculated to obtain the sliding window sequence that also maximizes the matching coefficient.

[0101] S4: Conducting a practical training comparative analysis on the trajectory segment sub-data and the key rescue action units contained therein to obtain a practical training comparative evaluation result of the trajectory segment sub-data.

[0102] The trajectory segment sub-data is subjected to a practical training comparative analysis with the key rescue action units contained therein, and a practical training comparative evaluation result of the trajectory segment sub-data is obtained, including:

[0103] The calculation method of the training comparison evaluation results of the trajectory segment sub-data is:

[0104]

[0105] Among them, Score unit_nThe trajectory segment sub-data S containing the nth key rescue action unit unit_n unit_n The training comparison evaluation results of [Sim1(S unit_n ,unit_n),Sim2(S unit_n ,unit_n),Sim3((S unit_n , unit_n) are the trajectory segment sub-data S unit_n The action similarity, tool matching and execution object matching between the key rescue action unit unit_n, Represents the trajectory segment sub-data S unit_n The time matching degree between the key emergency action unit unit_n, exp((·) represents the exponential function with the natural constant as the base, σ represents the time control parameter, set σ to 2, time(S unit_n ) represents the trajectory segment sub-data S unit_n Length of time, time unit_n Indicates the time expectation in the key rescue action unit unit_n, max time Indicates the normalized control parameter, set max time 90 seconds;

[0106] n∈[1,N], where N represents the number of key rescue action units.

[0107] S5: Comprehensively compare and evaluate the training results of all trajectory segment sub-data, and generate multi-dimensional training evaluation results for training feedback.

[0108] The training and comparative evaluation results of all trajectory segment sub-data are integrated to generate the training evaluation results, including:

[0109] The mean action similarity, tool matching, execution object matching, and time matching between N key rescue action units and the associated trajectory segment sub-data are calculated as the training evaluation results of the trainees in terms of operation completion, tool usage correctness, object positioning accuracy, and time execution rationality.

[0110] As an embodiment of the present invention, by identifying the training comparative evaluation results of the trajectory segment sub-data, the training evaluation results are converted into a training evaluation report. An example of the report is:

[0111]

Practical training task

[0112]

Completion

[0113] [Action Standardization] 87% (There was an error in the action sequence during the safety cordon setting stage)

[0114]

Tool Usage Correctness

[0115] [Operation object positioning accuracy] 81% (operation object recognition offset, failure to identify operation on the water pump and drainage pipe interface)

[0116]

Time execution rationality

[0117]

Overall Rating

[0118] [Suggestion] Improve the ability to identify the consistency of interaction direction in object positioning actions.

[0119] Example 2:

[0120] like Figure 2 The training evaluation flow chart of the standard emergency rescue process shown is as follows: Node1->Node2->Node3->Node4->Node5->Node6->Node7->Node8, wherein Node1 to Node8 are respectively preliminary water level detection and flooding alarm confirmation, power off the main switch, setting a safety cordon, activating the backup power supply, preparation and layout of the drainage pump, starting the drainage pump, electrical equipment safety inspection, completion of the rescue and reporting of the rescue record; the nodes whose key score vectors are higher than the preset hazard sensitivity threshold, path turning threshold and recognition accuracy threshold include Node2 (power off the main switch), Node3 (setting a safety cordon) and Node6 (starting the drainage pump), and the node names and node attributes are extracted as key rescue action units unit1, unit2, unit3;

[0121] An example of the key action template is:

[0122] Key rescue action unit unit 1:

[0123] Name of emergency operation behavior: cut off the main power switch;

[0124] Action sequence: confirm power supply → wear insulating gloves → use insulated pliers → pull down the main power switch;

[0125] Required tools: insulated pliers, safety helmet, insulated gloves;

[0126] Execution object: Main power switch panel;

[0127] Time expectation: 30 seconds;

[0128] Key score vector: (0.95, 0.5, 0.91);

[0129] Key rescue action unit 2:

[0130] Name of emergency operation behavior: Setting up safety cordon;

[0131] Action sequence: Demarcate the warning range → Take out the warning line → Deploy it at the station door → Set up warning signs;

[0132] Required tools: safety cordon, warning signs;

[0133] Implementation object: the peripheral area of the power distribution station;

[0134] Time expectation: 90 seconds;

[0135] Key score vector: (0.71, 0.5, 0.88);

[0136] Key rescue action unit unit 3:

[0137] Name of emergency operation behavior: Start the drainage pump;

[0138] Action sequence: confirm the placement of the water pump → turn on the power switch → monitor the water output status;

[0139] Required tools: drainage pump, rubber pad, detector;

[0140] Implementation object: water pump and drainage pipe interface;

[0141] Time expectation: 60 seconds;

[0142] Key score vector: (0.86, 0.5, 0.92).

[0143] Example 3:

[0144] A distribution substation emergency training and evaluation system based on big data collection, comprising a server and a training process decomposition device, wherein the server comprises a trajectory matching module and a training evaluation module:

[0145] The trajectory matching module is used to collect the trainee's behavioral trajectory data during the training process, match and analyze the behavioral trajectory data with the key action template, and extract the trajectory segment sub-data containing the key rescue action unit;

[0146] The training evaluation module is used to perform training comparative analysis on the trajectory segment sub-data and the key rescue action units contained therein, obtain the training comparative evaluation results of the trajectory segment sub-data, and integrate the training comparative evaluation results of all trajectory segment sub-data to generate multi-dimensional training evaluation results for training feedback;

[0147] The training process decomposition device is used to construct a standard emergency rescue operation process in a distribution station flooding scenario, and decompose the standard emergency rescue operation process into several groups of key rescue action units, and the key rescue action units form a key action template;

[0148] This is to implement a distribution station emergency training and evaluation method based on big data collection in the above embodiment.

[0149] It should be understood that the embodiment is for illustration only and the scope of the patent application is not limited to this structure.

[0150] It should be noted that the serial numbers of the above-mentioned embodiments of the present invention are for descriptive purposes only and do not represent the advantages or disadvantages of the embodiments. In addition, the terms "including", "comprising" or any other variations thereof are intended to cover non-exclusive inclusion, so that a process, device, article or method comprising a series of elements includes not only those elements, but also other elements not explicitly listed, or also includes elements inherent to such process, device, article or method. In the absence of further restrictions, an element defined by the sentence "including a ..." does not exclude the presence of other identical elements in the process, device, article or method comprising the element.

[0151] Through the description of the above embodiments, those skilled in the art can clearly understand that the above-mentioned embodiment methods can be implemented by means of software plus the necessary general hardware platform, and of course can also be implemented by hardware, but in many cases the former is a better embodiment. Based on this understanding, the technical solution of the present invention is essentially or the part that contributes to the prior art can be embodied in the form of a software product, which is stored in a storage medium (such as ROM / RAM, magnetic disk, optical disk) as described above, and includes a number of instructions for enabling a terminal device (which can be a mobile phone, computer, server, or network device, etc.) to execute the methods described in each embodiment of the present invention.

[0152] The above are only preferred embodiments of the present invention and are not intended to limit the patent scope of the present invention. Any equivalent structure or equivalent process transformation made using the contents of the present invention description and drawings, or directly or indirectly applied in other related technical fields, are also included in the patent protection scope of the present invention.

Claims

1. A method for evaluating emergency training in power distribution stations based on big data collection, characterized in that: The method comprises: S1: Construct a standard emergency rescue operation process for the substation flooding scenario, decompose the standard emergency rescue operation process into several groups of key rescue action units, and form a key action template from the key rescue action units; S2: Collect the behavioral trajectory data of trainees during the training process; S3: Matching and analyzing the behavior trajectory data with the key action templates to extract the trajectory segment sub-data containing the key rescue action units; S4: Conducting a practical training comparative analysis on the trajectory segment sub-data and the key rescue action units contained therein to obtain a practical training comparative evaluation result of the trajectory segment sub-data; S5: Comprehensively compare and evaluate the training results of all trajectory segment sub-data, and generate multi-dimensional training evaluation results for training feedback.

2. The method for evaluating power distribution station emergency training based on big data collection according to claim 1, characterized in that: The standard emergency rescue operation process is a directed graph structure, including nodes, edges and node attributes, including: The nodes are operation units in the rescue process, and the operation units describe specific rescue operation behaviors. The edges represent the operation timing logic or conditional jumps between nodes. The node attributes include the tools required to execute the node, the action category, the execution object, the time expectation, and the key attributes of the node. The key attributes include the node's hazard sensitivity, path turning degree, and recognition accuracy. The hazard sensitivity indicates the level of safety risk that may be caused if the rescue operation behavior corresponding to the node is executed incorrectly. The path turning degree indicates the turning role of the node in the entire standard emergency rescue operation process. The recognition accuracy indicates whether the rescue operation behavior corresponding to the node can be accurately identified.

3. A method for evaluating power distribution station emergency training based on big data collection as claimed in claim 2, characterized in that: Based on the key attributes of the nodes, the standard emergency rescue operation process is decomposed into several groups of key rescue action units, including: The key attributes of the nodes are normalized as the key scoring vector of the nodes, which includes the hazard sensitivity score, path turning score and recognition accuracy score. Nodes whose hazard sensitivity score, path turning score and recognition accuracy score are higher than the preset hazard sensitivity threshold, path turning threshold and recognition accuracy threshold, respectively, are selected, and the node names and node attributes are extracted as a set of key rescue action units.

4. The method for evaluating power distribution station emergency training based on big data collection according to claim 1, characterized in that: The behavior trajectory data consists of multiple continuous behavior sequence points, each of which includes the trainee's position, action posture, operation object and timestamp information, including: The trainee's position, movement posture and operation object are obtained by monitoring sensors, wherein the sensor types include cameras, sound sensors and positioning devices. The positioning device is used to monitor the trainee's position under timestamp information, and the camera is used to monitor images in different scenes. Based on the trainee's position under timestamp information, the scene image of the trainee in the timestamp information and the position is obtained, and the trainee's movement posture and operation object are extracted from the scene image. The trainee's position, the extracted trainee's movement posture, operation object and timestamp information in the scene image are used as behavior sequence points.

5. The method for evaluating power distribution station emergency training based on big data collection according to claim 4, characterized in that: Matching and analyzing the behavior trajectory data with key action templates includes: Preprocessing the behavior sequence points in the behavior trajectory data, wherein the preprocessing method includes removing the behavior sequence points that are completely identical except for the timestamp information, to obtain preprocessed behavior trajectory data; The pre-processed behavior trajectory data is processed in a sliding window form to obtain multiple sliding window sequences, where each sliding window sequence consists of multiple behavior sequence points. The timestamp information difference between the last behavior sequence point and the first behavior sequence point in the sliding window sequence is Len, where Len is the length of the sliding window. step is the step size of the sliding window; Calculate the matching coefficient between the sliding window sequence and the key rescue action unit in the key action template, and use the key rescue action unit with a matching coefficient higher than a preset coefficient threshold as the matching result of the sliding window sequence. Use the sliding window sequence with the matching result as the trajectory segment sub-data, and use the matched key rescue unit as the inclusion value of the trajectory segment sub-data. The matching coefficient is calculated as follows: Sim(S,unit)=w1·Sim1(S,unit)+w2·Sim2((S,unit)+w3·Sim3(S,unit); Among them, Sim(S, unit) represents the matching coefficient between the sliding window sequence S and the key rescue action unit unit, Sim1(S, unit) represents the action similarity between the sliding window sequence S and the key rescue action unit unit, Sim2(S, unit) represents the tool matching degree between the sliding window sequence S and the key rescue action unit unit, Sim3(S, unit) represents the execution object matching degree between the sliding window sequence S and the key rescue action unit unit; w1, w2, and w3 all represent matching coefficient weights. Set w1, w2, and w3 to 0.5, 0.3, and 0.2 respectively.

6. A method for evaluating power distribution station emergency training based on big data collection as claimed in claim 5, characterized in that: The trajectory segment sub-data is subjected to a practical training comparative analysis with the key rescue action units contained therein, and a practical training comparative evaluation result of the trajectory segment sub-data is obtained, including: The calculation method of the training comparison evaluation results of the trajectory segment sub-data is: Among them, Score unit_n The trajectory segment sub-data S containing the nth key rescue action unit unit_n unit_n The training comparison evaluation results of [Sim1(S unit_n ,unit_n),Sim2(S unit_n ,unit_n),Sim3((S unit_n , unit_n) are the trajectory segment sub-data S unit_n The action similarity, tool matching and execution object matching between the key rescue action unit unit_n, Represents the trajectory segment sub-data S unit_n The time matching degree between the key emergency action unit unit_n, exp((·) represents the exponential function with the natural constant as the base, σ represents the time control parameter, set σ to 2, time(S unit_n ) represents the trajectory segment sub-data S unit_n Length of time, time unit_n Indicates the time expectation in the key rescue action unit unit_n, max time Indicates the normalized control parameter, set max time 90 seconds; n∈[1,N], where N represents the number of key rescue action units.

7. A method for evaluating power distribution station emergency training based on big data collection as claimed in claim 6, characterized in that: The training and comparative evaluation results of all trajectory segment sub-data are integrated to generate the training evaluation results, including: The mean action similarity, tool matching, execution object matching, and time matching between N key rescue action units and the associated trajectory segment sub-data are calculated as the training evaluation results of the trainees in terms of operation completion, tool usage correctness, object positioning accuracy, and time execution rationality.

8. A distribution station emergency training and evaluation system based on big data collection, characterized by: The distribution station emergency training and evaluation system based on big data collection includes a server and a training process decomposition device. The server includes a trajectory matching module and a training evaluation module: The trajectory matching module is used to collect the trainee's behavioral trajectory data during the training process, match and analyze the behavioral trajectory data with the key action template, and extract the trajectory segment sub-data containing the key rescue action unit; The training evaluation module is used to perform training comparative analysis on the trajectory segment sub-data and the key rescue action units contained therein, obtain the training comparative evaluation results of the trajectory segment sub-data, and integrate the training comparative evaluation results of all trajectory segment sub-data to generate multi-dimensional training evaluation results for training feedback; The training process decomposition device is used to construct a standard emergency rescue operation process in a distribution station flooding scenario, and decompose the standard emergency rescue operation process into several groups of key rescue action units, and the key rescue action units form a key action template; To realize a distribution station emergency training evaluation method based on big data collection as described in any one of claims 1-8.

Images