CN117372723A - Intelligent substation violation operation early warning system - Google Patents
Intelligent substation violation operation early warning system Download PDFInfo
- Publication number
- CN117372723A CN117372723A CN202311002855.3A CN202311002855A CN117372723A CN 117372723 A CN117372723 A CN 117372723A CN 202311002855 A CN202311002855 A CN 202311002855A CN 117372723 A CN117372723 A CN 117372723A
- Authority
- CN
- China
- Prior art keywords
- early warning
- matching
- feature
- information
- data
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Pending
Links
- 230000006399 behavior Effects 0.000 claims abstract description 35
- 238000010801 machine learning Methods 0.000 claims abstract description 32
- 238000000034 method Methods 0.000 claims description 38
- 238000011176 pooling Methods 0.000 claims description 36
- 230000009471 action Effects 0.000 claims description 32
- 238000003066 decision tree Methods 0.000 claims description 31
- 238000012545 processing Methods 0.000 claims description 18
- 238000013138 pruning Methods 0.000 claims description 17
- 230000006870 function Effects 0.000 claims description 16
- 238000000605 extraction Methods 0.000 claims description 12
- 238000012360 testing method Methods 0.000 claims description 12
- 239000013598 vector Substances 0.000 claims description 12
- 230000008569 process Effects 0.000 claims description 11
- 238000013527 convolutional neural network Methods 0.000 claims description 10
- 238000005457 optimization Methods 0.000 claims description 8
- 238000012549 training Methods 0.000 claims description 7
- 238000013145 classification model Methods 0.000 claims description 5
- 239000000284 extract Substances 0.000 claims description 5
- 238000012935 Averaging Methods 0.000 claims description 4
- 238000004140 cleaning Methods 0.000 claims description 4
- 238000013461 design Methods 0.000 claims description 4
- 238000007781 pre-processing Methods 0.000 claims description 4
- 238000011156 evaluation Methods 0.000 claims description 3
- 238000005259 measurement Methods 0.000 claims description 3
- 230000001960 triggered effect Effects 0.000 claims description 2
- 230000033001 locomotion Effects 0.000 claims 3
- 230000003542 behavioural effect Effects 0.000 claims 1
- 238000012544 monitoring process Methods 0.000 abstract description 8
- 238000004519 manufacturing process Methods 0.000 abstract description 5
- 230000002159 abnormal effect Effects 0.000 abstract description 3
- 230000009467 reduction Effects 0.000 abstract description 3
- 238000010223 real-time analysis Methods 0.000 abstract description 2
- 230000000694 effects Effects 0.000 description 12
- 238000010586 diagram Methods 0.000 description 9
- 238000004590 computer program Methods 0.000 description 7
- 230000004913 activation Effects 0.000 description 5
- 238000004458 analytical method Methods 0.000 description 4
- 238000005516 engineering process Methods 0.000 description 4
- 239000011159 matrix material Substances 0.000 description 4
- 238000012986 modification Methods 0.000 description 4
- 230000004048 modification Effects 0.000 description 4
- 238000003860 storage Methods 0.000 description 3
- 230000004075 alteration Effects 0.000 description 2
- 230000008901 benefit Effects 0.000 description 2
- 238000005520 cutting process Methods 0.000 description 2
- 238000007405 data analysis Methods 0.000 description 2
- 230000007547 defect Effects 0.000 description 2
- 238000009826 distribution Methods 0.000 description 2
- 230000007613 environmental effect Effects 0.000 description 2
- 238000012423 maintenance Methods 0.000 description 2
- 230000009466 transformation Effects 0.000 description 2
- 238000012795 verification Methods 0.000 description 2
- 230000009286 beneficial effect Effects 0.000 description 1
- 230000005540 biological transmission Effects 0.000 description 1
- 230000008859 change Effects 0.000 description 1
- 238000006243 chemical reaction Methods 0.000 description 1
- 238000010276 construction Methods 0.000 description 1
- 238000013480 data collection Methods 0.000 description 1
- 238000007418 data mining Methods 0.000 description 1
- 230000006872 improvement Effects 0.000 description 1
- 230000003287 optical effect Effects 0.000 description 1
- 238000002360 preparation method Methods 0.000 description 1
- 230000000717 retained effect Effects 0.000 description 1
- 238000005096 rolling process Methods 0.000 description 1
- 238000012216 screening Methods 0.000 description 1
- 238000010187 selection method Methods 0.000 description 1
- 238000004088 simulation Methods 0.000 description 1
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V10/00—Arrangements for image or video recognition or understanding
- G06V10/70—Arrangements for image or video recognition or understanding using pattern recognition or machine learning
- G06V10/74—Image or video pattern matching; Proximity measures in feature spaces
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/04—Architecture, e.g. interconnection topology
- G06N3/0464—Convolutional networks [CNN, ConvNet]
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/08—Learning methods
- G06N3/082—Learning methods modifying the architecture, e.g. adding, deleting or silencing nodes or connections
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N5/00—Computing arrangements using knowledge-based models
- G06N5/01—Dynamic search techniques; Heuristics; Dynamic trees; Branch-and-bound
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V10/00—Arrangements for image or video recognition or understanding
- G06V10/70—Arrangements for image or video recognition or understanding using pattern recognition or machine learning
- G06V10/764—Arrangements for image or video recognition or understanding using pattern recognition or machine learning using classification, e.g. of video objects
- G06V10/765—Arrangements for image or video recognition or understanding using pattern recognition or machine learning using classification, e.g. of video objects using rules for classification or partitioning the feature space
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V10/00—Arrangements for image or video recognition or understanding
- G06V10/70—Arrangements for image or video recognition or understanding using pattern recognition or machine learning
- G06V10/77—Processing image or video features in feature spaces; using data integration or data reduction, e.g. principal component analysis [PCA] or independent component analysis [ICA] or self-organising maps [SOM]; Blind source separation
- G06V10/774—Generating sets of training patterns; Bootstrap methods, e.g. bagging or boosting
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V10/00—Arrangements for image or video recognition or understanding
- G06V10/70—Arrangements for image or video recognition or understanding using pattern recognition or machine learning
- G06V10/82—Arrangements for image or video recognition or understanding using pattern recognition or machine learning using neural networks
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V40/00—Recognition of biometric, human-related or animal-related patterns in image or video data
- G06V40/20—Movements or behaviour, e.g. gesture recognition
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- Evolutionary Computation (AREA)
- Artificial Intelligence (AREA)
- General Health & Medical Sciences (AREA)
- Software Systems (AREA)
- Computing Systems (AREA)
- Health & Medical Sciences (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Multimedia (AREA)
- Databases & Information Systems (AREA)
- Medical Informatics (AREA)
- Mathematical Physics (AREA)
- General Engineering & Computer Science (AREA)
- Data Mining & Analysis (AREA)
- Computational Linguistics (AREA)
- Biophysics (AREA)
- Biomedical Technology (AREA)
- Life Sciences & Earth Sciences (AREA)
- Molecular Biology (AREA)
- Human Computer Interaction (AREA)
- Social Psychology (AREA)
- Psychiatry (AREA)
- Image Analysis (AREA)
Abstract
The invention discloses an intelligent substation illegal operation early warning system, which relates to the technical field of image recognition and comprises real-time image information acquisition, image recognition, a rule matching algorithm and a machine learning model of a key operation area so as to determine whether the behavior of an operator accords with a safety rule. The real-time analysis of the operation of operators is realized by covering the key operation area by the wireless monitoring equipment. And meanwhile, a matching rule and a machine learning model are used, and the machine model is trained by using historical data, so that the early warning judging efficiency is improved. The operation state of the equipment in the transformer substation is monitored and analyzed, equipment faults and abnormal conditions are found in time, production line shutdown or production efficiency reduction is avoided, human factor influence is reduced, and the operation efficiency and reliability of the transformer substation are improved.
Description
Technical Field
The invention relates to the technical field of intelligent monitoring, in particular to an intelligent substation illegal operation early warning system.
Background
The transformer substation is an important component of a power system, and normal operation and safety and stability of the transformer substation are critical to power grid operation. The transformer substation has a large number of electrical equipment and high voltage power sources, so that operation and maintenance and construction work of the transformer substation need to comply with strict safety specifications and standards. If the illegal operation occurs, serious safety accidents can be caused, huge losses are caused to personnel, equipment and environment, and even the normal operation of the power grid can be seriously influenced.
An intelligent substation illegal operation early warning system is an intelligent system for monitoring, early warning and managing illegal operation behaviors possibly existing in a substation by utilizing computer vision and image recognition technology. The intelligent recognition system can intelligently recognize operators, equipment, scenes and the like in the transformer substation, discover and early warn potential safety hazards and illegal behaviors possibly existing, and timely take measures to avoid accidents.
Patent CN115115474a discloses a method and a system for analyzing violation data of electric power operation, different target early warning models are constructed according to different data analysis requirements, parameter data required to be input by each target early warning model is collected when a task node is reached by setting a timing task mode, each target early warning model is operated to obtain an operation result, finally the operation result is extracted to be the violation data with early warning, and the violation data and the operation result are visually displayed. However, the matching algorithm is still simply used, and the problem of accurate early warning of illegal behaviors in a complex scene of a transformer substation cannot be solved. The system uses an offline system, namely, the existing violation data is analyzed, and the uploading picture cannot be analyzed in real time to timely stop the early warning of the violation.
However, in the process of implementing the technical scheme of the invention in the embodiment of the application, the inventor of the application finds that at least the following technical problems exist in the above technology: a traditional substation illegal operation early warning system monitors an operation area by using a wired monitoring network, and a single matching algorithm used by a server is low in intelligence and low in matching speed, is easy to report in a missing mode or misinformation mode, and cannot early warn illegal operations in time.
Disclosure of Invention
According to the intelligent substation violation operation early warning system, the technical problem that the early warning system cannot accurately early warn violation operation in real time in the prior art is solved, and the effect of accurately identifying staff violation operation and early warning in a complex environment of a substation is achieved.
The system comprises an information acquisition module, an image recognition module, an early warning information matching module, an early warning notification module and a system optimization module: the information acquisition module is used for acquiring substation data information and uploading the substation data information to the server in real time, wherein the substation data information comprises key operation area image information, equipment and position sensor measurement parameters; the image recognition module is used for extracting behavior characteristics of the image data personnel uploaded to the server, including gestures, operation actions and tool use characteristics; the early warning information matching module is used for designing a corresponding early warning database according to the safety specification of the transformer substation, establishing a machine learning model based on the current image data, and simultaneously adopting a rule matching algorithm and the machine learning model to predict and identify the image identification module; the early warning notification module is used for sending out an alarm according to the recognition result of the early warning information matching module; and the system optimization module is used for recording and classifying the early warning information and optimizing the early warning system.
Furthermore, in the image recognition module, the feature extraction is performed on the input image data through a CNN convolutional neural network algorithm, and the specific steps are as follows: carrying out convolution operation on the image data through a convolution layer, extracting local features in input data, and forming a feature map; downsampling the feature map through a pooling layer, reducing the size of the feature map, and reserving the action features of the image; after being processed by a convolution layer and a pooling layer, the feature map is unfolded into one-dimensional vectors, then the one-dimensional vectors are sent to a full-connection layer for processing, and the output is normalized by using a softmax function, so that the prediction probability of each class is obtained.
Further, the convolution operation is to perform convolution operation on the original image and the convolution kernel, and extract the feature map C after convolution i,j The convolution formula is as follows:
where i and j represent the position of the center pixel of the current convolution window in the input feature map,
I i+m-1,j+n-1 representing the value, K, of the (i+m-1, j+n-1) th pixel in the input image I m,n The values of the (M, N) th element in the convolution kernel K are represented, where I is the original image, K is the convolution kernel, C is the feature map after convolution, and M and N are the magnitudes of the convolution kernels.
Further, the pooling layer has a maximum pooling and an average pooling, and the formula is as follows:
maximum pooling:
f(x，y)＝max m，n∈Rx，y (I m，n )，
wherein R is x，y Represents a pooled region centered on the (x, y) th pixel in the feature map, (I) m，n ) The value of the (m, n) th pixel in the feature map is represented. The effect of maximum pooling is to select the largest pixel value from the pooled region as the output characteristic value of the region, so that the most significant characteristic information can be reserved.
Averaging and pooling:
wherein R is x，y Representing the number of pixels of the pooling area, i.e. the area of the pooling area, (I) m，n ) The value of the (m, n) th pixel in the feature map is represented. The effect of the average pooling is to average the pixel values in the pooled region, and the average value of the pooled region is obtained as the output characteristic value of the region, so that the characteristic map can be smoothed, and some secondary characteristic information can be reserved.
Further, the information matching module establishes an early warning database according to the security operation specification of the transformer substation, and is used for describing and matching the image information characteristics to be identified, and the specific steps are as follows: a machine learning method is adopted, and a machine learning classification model is established through substation forward image data; the matching engine performs feature extraction and comparison on the input data to find out the matched part of the input data from the established database so as to realize the matching and execution of the established early warning database; the method of matching the machine learning model and the rules is combined, and meanwhile, the behaviors are classified and matched; in the process of behavior identification and rule matching, a corresponding threshold is set, and when the probability value of the behavior exceeds the set threshold, the behavior is judged to be illegal, and early warning is triggered.
Further, the machine learning classification model is a decision tree algorithm model, and the training of the model comprises the following steps: data preprocessing: preprocessing a given data set, including data cleaning, missing value processing, feature encoding operations, so as to be used by a decision tree algorithm; attribute selection: selecting an optimal attribute from all available attributes as a split attribute of the current node so as to divide the data set into different subsets; evaluating the importance of the attribute using the information gain or the information gain ratio; generating a decision tree: dividing the data set according to the selected attribute to generate a subtree, and recursively executing attribute selection and division operation in the subtree until all samples in the data set belong to the same category or stop conditions are met; pruning of decision trees: to prevent overfitting, pruning of the generated decision tree is required; the method is generally divided into a pre-pruning method and a post-pruning method, wherein the pre-pruning method is to prune in the process of generating a decision tree, and the post-pruning method is to prune after the decision tree is generated, so as to obtain an optimized decision tree model; model evaluation: and testing the produced decision tree by using a test set, and calculating indexes of accuracy, recall rate and regression rate of the model to enable the model to meet design requirements.
Further, the optimal attribute selection method comprises the following steps: when the decision node selects the optimal dividing point, dividing standard of each feature needs to be calculated, wherein the dividing standard comprises information gain and a base index;
the information gain represents the degree of contribution of a feature to classification, and is represented by the following formula:
wherein Gain (D, A) is the Gain of the information sought, D represents the data set, A represents the feature, V represents each value of the feature A, D v Representing a subset of feature A Values v, v ε Values (A) representing a set of feature A Values, ent (D v ) Information entropy representing data set D, where y represents a set of categories of samples in the data set, p k Representing the probability that the sample belongs to category K.
Further, the keni index gini_index is calculated based on the concept of keni purity, and represents the degree of influence of a feature on classification, and is expressed by the following formula:
wherein Gini (D) represents the genii purity, p, of dataset D k Representing the probability that the sample belongs to category K, A represents the feature, v represents each value of feature A, D v Representing a subset of feature a Values v, and Values (a) representing a collection of feature a Values.
The matching and execution of rules in the established early warning rule base are further realized through a matching engine, and the specific realization steps are as follows: defining a database: abstracting the service specification into a database form, wherein the database form comprises names, descriptions, triggering conditions and executing actions of the specification; building a matching engine: building a matching engine by using a matching engine framework, compiling a database into executable rule objects, and storing the executable rule objects in the rule engine; incoming fact data: transmitting fact data which need to be matched, including characteristic information of image recognition and key parameters of sensor recognition into a matching engine; feature matching: the matching engine performs feature matching according to the input fact data and the constructed database, and finds out rules meeting the conditions; the actions are performed: the matching engine performs the action of matching into the database and outputs the result.
Further, the method for combining the machine learning model and the feature matching simultaneously classifies and matches the behaviors, and specifically comprises the following steps: classifying the image feature extraction, and marking different actions to enable the different actions to enter different module matching; for action features marked as low risk coefficients, matching the action features by a matching engine only, and outputting a matching result; for action features marked as complex actions, actions with high risk coefficients and action features not marked in the database, after the matching engine is matched, the machine learning model is used for matching, and a result is output.
The invention has the following beneficial effects:
(1) The intelligent substation illegal operation early warning system carries out matching classification on the identified characteristic data through a decision tree algorithm, realizes quick identification on a power distribution scene, has certain predictive capability analysis capability on safety parameters, and reduces the occurrence probability of danger.
(2) The intelligent substation illegal operation early warning system matches the image characteristics through a mode combining a matching engine and a machine learning model, so that the matching speed is high in accuracy and intelligent level.
(3) The intelligent substation violation operation early warning system performs feature extraction on input image data through a CNN convolutional neural network algorithm, so that feature extraction efficiency and system operation speed are improved.
Of course, it is not necessary for any one product to practice the invention to achieve all of the advantages set forth above at the same time.
Drawings
FIG. 1 is a workflow diagram of each module of the intelligent substation violation operation early warning system;
FIG. 2 is a flow chart of feature extraction from image information by the convolution algorithm of the present invention;
FIG. 3 is a flow chart of feature matching using both a matching engine and a machine learning model in accordance with the present invention.
Detailed Description
According to the intelligent substation illegal operation early warning system, the problems that a traditional illegal operation early warning system cannot comprehensively monitor a key operation area, the extraction efficiency of monitoring picture features is low, the intelligent degree of using a single early warning rule system is low, and missing report and false report are easy to occur are solved.
In order to solve the problem of crosstalk, the technical solution in the embodiment of the present application is as follows:
the system comprises an operation information acquisition module, an image recognition module, an early warning rule matching module, an early warning notification module, an early warning information record archiving and system optimization module. As shown in fig. 1, fig. 1 is a working flow chart of each module of the intelligent substation violation operation early warning system. The method comprises the steps that information of an operation area is collected in real time through an information collecting module, the information of the operation area is transmitted to an image recognition module, after feature recognition is carried out by the image recognition module, recognized action features are transmitted to an early warning information matching module to be matched by a matching engine and a machine learning model, an early warning notification module notifies personnel to change an electric power system according to a matching recognition result so as to ensure environmental safety, meanwhile, a system optimization module can record and classify the early warning information, and the early warning matching model is optimized according to an early warning effect.
And the information acquisition module is used for: acquiring image information of key operation areas of the transformer substation through the whole coverage of the key operation areas by wireless monitoring equipment, and ensuring real-time recording of operation personnel and uploading the operation personnel to a server for processing and analysis; by utilizing the internet of things technology, sensors are installed at key equipment and tower positions in a transformer substation, equipment states, environmental voltage, current and temperature parameter changes are sensed in real time, and the sensor is transmitted to a data center through a wireless network for real-time analysis and processing;
an image recognition module: the image recognition system is used for analyzing and processing the image information based on an image recognition technology to acquire characteristic information;
the early warning rule matching module: designing a corresponding early warning database according to the safety specification of the transformer substation, adopting a rule matching algorithm and a machine learning model, and carrying out real-time matching on the current operation characteristic behavior based on the characteristic information to judge whether the current operation characteristic behavior meets the rule requirement;
the early warning notification module: according to the matching result, sending early warning information to the violation operators in sound, vibration and light modes to remind the violation operators to pay attention to safe operation; for serious illegal behaviors, the system starts emergency measures for cutting off the power supply and isolating equipment so as to reduce the safety risk;
and the early warning information record archiving and system optimizing module: and recording and classifying the early warning information, and optimizing the early warning system by analyzing the early warning data, so that the efficiency and the accuracy of the early warning system are improved.
In order to better understand the above technical solutions, the following description will refer to the drawings and specific embodiments.
Example 1
The image acquisition module acquires the image information of the key operation area of the transformer substation by using the wireless monitoring equipment to cover the key operation area, so that the operation of operators is ensured to be recorded in real time, and the operation is uploaded to the server for processing and analysis.
For the image data uploaded to the server, the image recognition module performs feature extraction aiming at key information in the video picture, and extracts the behavior gesture, operation action and tool use feature of personnel for subsequent behavior recognition and early warning. And the image recognition module performs feature extraction and dimension reduction processing on the input data by using a CNN convolutional neural network algorithm.
The specific steps of image recognition and feature extraction are shown in fig. 2, and specifically include the following steps:
data preparation: each image is scaled to the same size and converted to a digital matrix.
Convolution layer: in CNN, the main role of the convolution layer is to extract features of the image by sliding the convolution kernel (also called a filter). The convolution kernel can be regarded as a small weight matrix that slides over the image, convolving each region of the image to obtain a feature map (also called a convolution map). Each feature map represents a particular local feature in the imageSuch as edges, corner points, textures, etc. Feature map C is obtained through convolution kernel i,j The formula is as follows:
i and j represent the position of the center pixel of the current convolution window in the input feature map, I i+m-1,j+n-1 Representing the value, K, of the (i+m-1, j+n-1) th pixel in the input image I m,n The values of the (M, N) th element in the convolution kernel K are represented, wherein I is an original image, K is the convolution kernel, C is a characteristic diagram after convolution, M and N are the sizes of the convolution kernels, and I and j are the positions of the convolution operation.
Activation function: the convolved feature map requires nonlinear conversion by a ReLU function activation function.
The function of the activation function is to convert the output of the convolution operation into a non-linear form.
Wherein z is i Is the i-th element in the output vector and k is the length of the output vector. The softmax function will each output z i The transformation is a real number between 0 and 1, which can be seen as outputting a probability estimate belonging to class i. Thus, the sum of all output values is 1, which allows the softmax function to be used to construct the probability distribution.
Pooling layer: the pooling layer may reduce the size of the feature map, reduce the computational effort, and may prevent overfitting. The pooling mode used has the maximum pooling and average pooling, and the formula is as follows:
maximum pooling:
wherein R is x，y Represents a pooled region centered on the (x, y) th pixel in the feature map, (I) m，n ) Representing the (m, n) th in the characteristic diagramA value of a pixel. The effect of maximum pooling is to select the largest pixel value from the pooled region as the output characteristic value of the region, so that the most significant characteristic information can be reserved.
Averaging and pooling:
wherein R is x，y Representing the number of pixels of the pooling area, i.e. the area of the pooling area, (I) m，n ) The value of the (m, n) th pixel in the feature map is represented. The effect of the average pooling is to average the pixel values in the pooled region, and the average value of the pooled region is obtained as the output characteristic value of the region, so that the characteristic map can be smoothed to a certain extent, and some secondary characteristic information can be reserved.
In this embodiment, the CNN algorithm needs to obtain high-level abstract features to match the model, so that a multi-layer convolution operation is required, including multiple convolution layers, activation functions, and pooling layers. Each convolution layer may extract features of different levels, and multi-layer convolution may extract more abstract and advanced features.
After the operations of rolling and pooling for many times, the feature map can be converted into one-dimensional vectors, and classification or regression can be performed through the full connection layer. The formula of the full link layer is as follows:
f(Wx+b)，
where W is the weight matrix, x is the input vector, b is the bias term, and f is the activation function. The function of the full connection layer is to spread the output characteristic diagram of the previous convolution layer or pooling layer into a one-dimensional vector, multiply the one-dimensional vector with a weight matrix, perform linear transformation and finally output a new one-dimensional characteristic vector.
Conventional matching modules typically use a rule matching algorithm, which refers to searching a set of data for a data item that meets a rule according to a certain rule. The principle of this algorithm is generally to express a rule using a regular expression or some logic statement, and then find the data item matching the rule by matching the data items one by one. The rule matching algorithm has the advantages of simple implementation, easy understanding and maintenance, and suitability for some simple scenes. The method has the defects that the method can only deal with the rule which is defined in advance, complex data relations cannot be processed adaptively, the rule matching algorithm has higher requirements on the accuracy and the integrity of the rule, and has higher requirements on the establishment of a matching database, and once the rule is changed, the algorithm needs to readjust the rule.
In order to solve some of the drawbacks of rule matching algorithms, machine learning techniques are employed to analyze and process the data to a higher degree of intelligence. Compared with a rule matching algorithm, the machine learning has stronger intelligence and self-adaption, and can automatically learn modes and rules from data, thereby realizing more accurate and efficient data processing. The establishment of the early warning rule module comprises the following specific steps:
establishing a corresponding database according to the safety operation specification of the transformer substation, and describing the mode and the action characteristic to be identified; in the process of behavior recognition and rule matching, corresponding thresholds are set for different behavior and sensor recognition parameters; when the probability value of the behavior or the sensor parameter exceeds a certain threshold value, judging the behavior as illegal behavior, and triggering early warning;
establishing a machine learning classification model, training by using a large amount of historical data, wherein the machine learning model is a decision tree algorithm model, and comprises the following specific steps of:
collecting a sufficient amount of image data including offending and normal operations, and in each case sensor identification parameters; the data can be manually marked or marked by using an unsupervised learning method;
initializing model parameters: before training the model, initializing parameters of the model; the method adopts a random initialization mode, so that the initial weight of the model has certain randomness, and the model is better adapted to data;
attribute selection: selecting an optimal attribute from all available attributes as a split attribute of the current node so as to divide the data set into different subsets; evaluating the importance of the attribute using the information gain or the information gain ratio;
the importance of the features is evaluated by calculating the information gain, and the features with high information gain are selected as the dividing basis.
Wherein D represents a dataset, A represents a feature, V represents each value of feature A, D v Representing a subset of feature A Values v, v ε Values (A) representing a set of feature A Values, ent (D v ) Information entropy representing data set D, where y represents a set of categories of samples in the data set, p k Representing the probability that the sample belongs to category K.
The Gini index is calculated based on the concept of Gini unrepeace, and represents the degree of influence of a feature on classification, expressed by the following formula:
wherein Gini (D) represents the genii purity, p, of dataset D k Representing the probability that the sample belongs to category K, A represents the feature, v represents each value of feature A, D v Representing a subset of feature a Values v, and Values (a) representing a collection of feature a Values.
Generating a decision tree: dividing the data set according to the selected attribute to generate a subtree, and recursively executing attribute selection and division operation in the subtree until all samples in the data set belong to the same category or stop conditions are met;
pruning of decision trees: to prevent overfitting, pruning of the generated decision tree is required; the method is generally divided into a pre-pruning method and a post-pruning method, wherein the pre-pruning method is to prune in the process of generating a decision tree, and the post-pruning method is to prune after the decision tree is generated, so as to optimize the produced decision tree model;
model evaluation: and testing the generated decision tree by using a test set for the optimized decision tree model, and calculating indexes of accuracy, recall rate and regression rate of the model to enable the model to meet design requirements.
When the accuracy rate, recall rate and precision rate of the model meet the design requirements, the matching engine is used for matching and executing the rules in the rule base, and the matching engine is used for extracting and comparing the characteristics of input data to find out the part matched with the rules in the rule base. And combining machine learning and rule matching methods for different types of violation operations, and simultaneously performing feature matching on input behavior features and sensor measurement parameters.
Predicting and identifying various behaviors and states in the transformer substation, verifying and identifying the identity of operators, classifying different types of operations, and identifying and early warning various abnormal conditions in the transformer substation; setting a corresponding threshold value in the process of behavior identification and rule matching; when the probability value of the behavior exceeds a certain threshold, judging that the behavior is illegal, and triggering early warning; as shown in fig. 3, the method of combining machine learning and rule matching simultaneously classifies and rule matching the behaviors, thereby improving the early warning effect, matching the action features marked as low risk factors only by a matching engine, outputting a matching result,
for action features marked as complex actions, actions with high risk coefficients and unlabeled action features in the database, after the matching engine is matched, the machine learning model is used for matching.
For the behavior of the illegal operation identified by the early warning rule matching module, sending early warning information to the illegal operator in a sound, vibration and light mode to remind the illegal operator of the safe operation; and the system can start automatic emergency measures of cutting off the power supply and isolating equipment so as to reduce the safety risk.
Recording and classifying early warning information: for each piece of early warning information, a corresponding recording and archiving system is required to be established, and the source, time, content and grade information of the early warning information are recorded;
corresponding processing flows are required to be designed aiming at the early warning information of different grades and types, so that the early warning information can be timely and effectively transmitted and processed;
for the release of the early warning information, various modes can be adopted, the release mode of the early warning information needs to be determined according to actual conditions, and the early warning information can be timely and accurately transmitted to related personnel.
According to the actual effect of early warning release, parameters of an early warning algorithm are modified and a model structure is adjusted for the insufficient part, and the early warning mode is optimized according to the actual situation by combining the actual early warning effect, so that the transmission effect of early warning information is improved. The specific operation steps are as follows:
and (3) data collection: firstly, data generated by an early warning system needs to be collected, wherein the data comprise early warning information, early warning time, early warning positions, early warning reasons and the like. The data may be collected by means of a data acquisition system, a sensor device, etc.
Data cleaning: the collected data is cleaned and deduplicated, invalid and duplicate data is excluded, and useful data is retained. The purpose of data cleaning is to improve the accuracy and reliability of the data and avoid negative impact on subsequent analysis.
And (3) data processing: and processing the cleaned data, including data analysis, classification, screening and other operations. The data is processed and analyzed using data mining, machine learning techniques, and potential problems and laws are identified.
Optimizing and early warning system: and optimizing the early warning system according to the data processing result. The accuracy and timeliness of the early warning can be improved by adjusting the early warning rule, optimizing the early warning range, adding the early warning means and the like. The optimization early warning system needs to comprehensively consider various factors including accuracy, timeliness, frequency and the like of early warning information, and the parameters of a matching model are modified according to the actual effect of the recognition matching system, for example, the parameters of a machine learning model are modified under the situations of missed judgment and misjudgment, misjudgment and missed judgment behaviors are accurately marked, the machine is enabled to recognize, and under the situation of mismatching, whether a matching rule is wrong or a model is wrong is judged, and the recognition result of the system is corrected.
Testing and verifying: and testing and verifying the optimized early warning system, and verifying the accuracy and the effectiveness of the optimized early warning system. The performance and effect of the system can be evaluated by verification through modes such as simulation test and actual scene test.
Continuous optimization: and continuously optimizing and improving the early warning system according to the test and verification results. For the found problems and defects, timely adjustment and improvement are needed to ensure continuous and stable operation of the early warning system.
In summary, in the intelligent substation illegal operation early warning system, early warning judging efficiency is improved through a machine learning model. Monitoring and analyzing the running state of equipment in the transformer substation, finding out faults and abnormal conditions of the equipment in time, avoiding production line shutdown or production efficiency reduction, reducing human factor influence and improving the running efficiency and reliability of the transformer substation
It will be appreciated by those skilled in the art that embodiments of the present invention may be provided as a method, system, or computer program product. Accordingly, the present invention may take the form of an entirely hardware embodiment, an entirely software embodiment or an embodiment combining software and hardware aspects. Furthermore, the present invention may take the form of a computer program product embodied on one or more computer-usable storage media (including, but not limited to, disk storage, CD-ROM, optical storage, and the like) having computer-usable program code embodied therein.
The present invention is described with reference to flowchart illustrations and/or block diagrams of methods, apparatus (systems) and computer program products according to embodiments of the invention. It will be understood that each flow and/or block of the flowchart illustrations and/or block diagrams, and combinations of flows and/or blocks in the flowchart illustrations and/or block diagrams, can be implemented by computer program instructions. These computer program instructions may be provided to a processor of a general purpose computer, special purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, create means for implementing the functions specified in the flowchart flow or flows and/or block diagram block or blocks.
These computer program instructions may also be stored in a computer-readable memory that can direct a computer or other programmable data processing apparatus to function in a particular manner, such that the instructions stored in the computer-readable memory produce an article of manufacture including instruction means which implement the function specified in the flowchart flow or flows and/or block diagram block or blocks.
These computer program instructions may also be loaded onto a computer or other programmable data processing apparatus to cause a series of operational steps to be performed on the computer or other programmable apparatus to produce a computer implemented process such that the instructions which execute on the computer or other programmable apparatus provide steps for implementing the functions specified in the flowchart flow or flows and/or block diagram block or blocks.
While preferred embodiments of the present invention have been described, additional variations and modifications in those embodiments may occur to those skilled in the art once they learn of the basic inventive concepts. It is therefore intended that the following claims be interpreted as including the preferred embodiments and all such alterations and modifications as fall within the scope of the invention.
It will be apparent to those skilled in the art that various modifications and variations can be made to the present invention without departing from the spirit or scope of the invention. Thus, it is intended that the present invention also include such modifications and alterations insofar as they come within the scope of the appended claims or the equivalents thereof.
Claims (10)
1. The intelligent substation violation operation early warning system is characterized by comprising an information acquisition module, an image recognition module, an early warning information matching module, an early warning notification module and a system optimization module:
the information acquisition module is used for acquiring substation data information and uploading the substation data information to the server in real time, wherein the substation data information comprises key operation area image information, equipment and position sensor measurement parameters;
the image recognition module is used for extracting behavior characteristics of the personnel image data uploaded to the server, including gestures, operation actions and tool use characteristics;
the early warning information matching module is used for designing a corresponding early warning database according to the safety specification of the transformer substation, establishing a machine learning model based on the forward image data, and simultaneously adopting a rule matching algorithm and the machine learning model to predict and identify the image identification module;
the early warning notification module is used for sending out an alarm according to the recognition result of the early warning information matching module;
the system optimization module is used for recording and classifying the early warning information and optimizing the early warning system.
2. The intelligent substation violation operation early warning system according to claim 1, wherein the image recognition module extracts the behavior characteristics of the personnel on the input image data through a CNN convolutional neural network algorithm, and specifically comprises the following steps:
carrying out convolution operation on the image data of each frame in the image data through a convolution layer, extracting local features in the input image data, and forming a feature map;
inputting the feature map into a pooling layer for downsampling, reducing the size of the feature map, and reserving the action features of the image;
after being processed by a convolution layer and a pooling layer, the feature map is unfolded into a one-dimensional vector, then the one-dimensional vector is sent to a full-connection layer for processing, and the output is normalized by using a softmax function, so that the one-dimensional vector feature is obtained.
3. The intelligent substation violation operation early warning system according to claim 2, wherein the convolution operation is to perform a convolution operation on the image data and the convolution kernel, and thenExtracting a convolved feature map C i,j The convolution formula is as follows:
where I and j represent the position of the center pixel of the current convolution window in the input feature map, I i+m-1,j+n-1 Representing the value, K, of the (i+m-1, j+n-1) th pixel in the input image data I m,n The values of the (M, N) th element in the convolution kernel K are represented, where I is the input image data, K is the convolution kernel, C is the feature map after convolution, and M and N are the sizes of the convolution kernels.
4. The intelligent substation violation operation early warning system of claim 2, wherein the pooling layer comprises a maximum pooling and an average pooling, and the formula is as follows:
maximum pooling:
f(x，y)＝max m，n∈Rx，y (I m，n )，
wherein R is x，y Represents a pooled region centered on the (x, y) th pixel in the feature map, (I) m，n ) A value representing the (m, n) th pixel in the feature map; the maximum pooling is used for selecting the maximum pixel value from the pooled region as the output characteristic value of the region, so that the most remarkable characteristic information can be reserved;
averaging and pooling:
wherein R is x，y Representing the number of pixels of the pooling area, i.e. the area of the pooling area, (I) m，n ) A value representing the (m, n) th pixel in the feature map; the average pooling is used for averaging pixel values in the pooled region, and the average value of the pooled region is obtained as an output characteristic value of the region, so that the characteristic map can be smoothed, and secondary characteristic information can be reserved.
5. The intelligent substation violation operation early warning system according to claim 1, wherein the early warning information matching module is used for establishing an early warning database according to a substation safety operation specification and is used for describing and matching the image information characteristics to be identified, and specifically comprises the following steps:
a machine learning method is adopted, and a machine learning classification model is established through training of substation forward image data;
the matching engine performs feature extraction and comparison on the extracted image information features, and finds out the matched part from the established early warning database to realize the matching and execution of the established early warning database;
the method of matching the machine learning model and the rules is combined, and meanwhile, behavior characteristics are classified and matched; in the process of behavior identification and rule matching, a corresponding threshold is set, and when the probability value of the behavior exceeds the set threshold, the behavior is judged to be illegal, and early warning is triggered.
6. The intelligent substation violation operation early warning system according to claim 5, wherein the machine learning classification model is a decision tree algorithm model, the decision tree algorithm divides the samples into different categories through a tree structure, and the decision tree algorithm model is trained by the following steps:
data preprocessing: preprocessing a given historical acquisition substation image data set, including an image training data set and an image testing data set, including data cleaning, missing value processing and feature coding operation;
attribute selection: selecting an optimal attribute from all available attributes as a split attribute of a current node so as to divide an image training data set into different subsets, and evaluating the importance of the attribute by using information gain or an information gain ratio;
generating a decision tree: dividing the image training data set according to the selected attributes to generate subtrees, and recursively executing attribute selection and division operations in the subtrees until all samples in the data set belong to the same category or stop conditions are met;
pruning of decision trees: pre-pruning is carried out in the process of generating the decision tree, and post-pruning is carried out after the generation of the decision tree is completed, so that an optimized latest model of the decision tree is obtained, wherein the pre-pruning is carried out in the process of generating the decision tree, and the post-pruning is carried out after the generation of the decision tree is completed, so that an optimized decision tree algorithm model is obtained;
model evaluation: and testing the latest model of the decision tree by using an image test data set, and calculating indexes of accuracy, recall rate and regression rate of the model to enable the model to meet design requirements.
7. The intelligent substation violation operation early warning system according to claim 6, wherein the optimal attribute selecting method is as follows:
calculating a division standard of each feature when the decision node selects the optimal division point, wherein the division standard comprises information gain and a base index;
the information gain represents the degree of contribution of a feature to classification, and is represented by the following formula:
wherein Gain (D, A) represents the Gain of the information, D represents the data set, A represents the feature, V represents each value of the feature A, D v Representing a subset of feature A Values v, v ε Values (A) representing a set of feature A Values, ent (D v ) Information entropy representing data set D, where y represents a set of categories of samples in the data set, p k Representing the probability that the sample belongs to class k.
8. The intelligent substation violation operation early warning system of claim 7 wherein the base index is calculated based on base unrefreshing and represents the degree of influence of a behavioral characteristic on classification by the following formula:
wherein Gini (D) represents the genii purity, p, of dataset D k' Representing the probability that the sample belongs to class k', gini_index represents the keni index, a represents the feature, v represents each value of feature a, D v Representing a subset of feature a Values v, and Values (a) representing a collection of feature a Values.
9. The intelligent substation violation operation early warning system according to claim 5, wherein the matching and execution of rules in the established early warning rule database are realized by a matching engine, and the specific implementation steps are as follows:
defining a database: abstract the business specification into the form of an early warning rule database, including names, descriptions, triggering conditions and executing actions of the specification;
building a matching engine: building a matching engine by using a matching engine framework, compiling a database into executable rule objects, and storing the executable rule objects in the rule engine;
incoming fact data: transmitting fact data which need to be matched, including characteristic information of image recognition and key parameters of sensor recognition into a matching engine;
feature matching: the matching engine performs feature matching according to the input fact data and the constructed early warning rule database, and finds out the conditions met in the early warning rule database;
the actions are performed: and the matching engine executes the actions in the matched early warning rule database and outputs the result.
10. The intelligent substation violation work early warning system of claim 5, wherein the method of combining machine learning model and rule matching simultaneously classifies and matches behavior features, specifically as follows:
classifying the image motion feature extraction, and marking different motions to enable the motions to enter different module matching;
for action features marked as low risk coefficients, matching the action features by a matching engine only, and outputting a matching result;
for action features marked as complex actions, actions with high risk coefficients and unlabeled action features in the database, after the matching engine is matched, the machine learning model is used for matching.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202311002855.3A CN117372723A (en) | 2023-08-10 | 2023-08-10 | Intelligent substation violation operation early warning system |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202311002855.3A CN117372723A (en) | 2023-08-10 | 2023-08-10 | Intelligent substation violation operation early warning system |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN117372723A true CN117372723A (en) | 2024-01-09 |
Family
ID=89402946
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202311002855.3A Pending CN117372723A (en) | 2023-08-10 | 2023-08-10 | Intelligent substation violation operation early warning system |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN117372723A (en) |
Cited By (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN118505166A (en) * | 2024-07-16 | 2024-08-16 | 国网安徽省电力有限公司信息通信分公司 | Power transmission line unplanned operation checking method and system based on image analysis |
-
2023
- 2023-08-10 CN CN202311002855.3A patent/CN117372723A/en active Pending
Cited By (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN118505166A (en) * | 2024-07-16 | 2024-08-16 | 国网安徽省电力有限公司信息通信分公司 | Power transmission line unplanned operation checking method and system based on image analysis |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN110097037B (en) | Intelligent monitoring method and device, storage medium and electronic equipment | |
| CN110738127A (en) | Helmet identification method based on unsupervised deep learning neural network algorithm | |
| CN113673459B (en) | Video-based production and construction site safety inspection method, system and equipment | |
| CN117421684B (en) | Abnormal data monitoring and analyzing method based on data mining and neural network | |
| CN113283282B (en) | Weak supervision time sequence action detection method based on time domain semantic features | |
| CN112434178B (en) | Image classification method, device, electronic equipment and storage medium | |
| CN113642474A (en) | Hazardous area personnel monitoring method based on YOLOV5 | |
| CN117372723A (en) | Intelligent substation violation operation early warning system | |
| CN116861331A (en) | Expert model decision-fused data identification method and system | |
| CN115358155A (en) | Power big data abnormity early warning method, device, equipment and readable storage medium | |
| CN116824517B (en) | Substation operation and maintenance safety control system based on visualization | |
| CN116457802A (en) | Automatic real-time detection, prediction and prevention of rare faults in industrial systems using unlabeled sensor data | |
| CN117035419B (en) | Intelligent management system and method for enterprise project implementation | |
| CN116756688A (en) | Public opinion risk discovery method based on multi-mode fusion algorithm | |
| KR102366787B1 (en) | Real-time sliding window based anomaly detection system for multivariate data generated by manufacturing equipment | |
| CN118097567B (en) | Work material machine elemental analysis acquisition system and method based on construction site video | |
| CN118038021A (en) | Transformer substation operation site foreign matter intrusion detection method based on improvement yolov4 | |
| CN117729043A (en) | Network security early warning method and system based on big data | |
| Xiao et al. | Monitoring the Abnormal Human Behaviors in Substations based on Probabilistic Behaviours Prediction and YOLO-V5 | |
| CN117423157A (en) | Mine abnormal video action understanding method combining migration learning and regional invasion | |
| CN114169623A (en) | Power equipment fault analysis method and device, electronic equipment and storage medium | |
| An et al. | Slope Collapse Detection Based on Image Processing | |
| CN114401398B (en) | Intelligent video operation and maintenance management system | |
| CN115063643B (en) | Automatic labeling method for power scene image data | |
| CN116630909B (en) | Unmanned intelligent monitoring system and method based on unmanned aerial vehicle |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination |