Power grid alarm information intelligent classification system based on deep learning

Abstract

The invention discloses a power grid alarm information intelligent classification system based on deep learning, which relates to the technical field of power grid alarm information intelligent classification and comprises a power grid alarm monitoring data acquisition module, a power grid alarm evaluation module, an information analysis module, a multi-dimensional feature analysis module, a disposal strategy distribution module and an intelligent classification module. The power grid alarm monitoring data acquisition module is used for acquiring power grid alarm monitoring data and preprocessing the acquired data; the power grid alarm evaluation module evaluates a power grid alarm coefficient based on the preprocessed power grid alarm monitoring data. According to the invention, a data acquisition technology, a power grid alarm evaluation technology, an information analysis technology, a multi-dimensional feature analysis technology, a convolutional neural network algorithm, a bidirectional long-short-term memory network technology and a knowledge graph drawing technology are combined with a modern information technology; and the intelligent degree in the power grid alarm information intelligent classification process based on deep learning is obviously enhanced.

Description

Technical Field

[0001] The present invention relates to the technical field of intelligent classification of power grid alarm information, and in particular to a power grid alarm information intelligent classification system based on deep learning. Background Art

[0002] In today's power system, ensuring safe, stable operation and reliable power supply is crucial, and grid alarm information processing has become a key link. With the continuous expansion of the power grid, the increasing complexity of the grid structure, and the accelerated advancement of new power system construction, the number of alarm messages generated by grid operation status monitoring has exploded. Currently, main grid control faces nearly 10,000 monitoring messages daily, which puts enormous pressure on the work. Especially when anomalies or faults occur, a large number of alarm messages emerge. Traditional intelligent classification of grid alarm information relies on manual identification by dispatchers, which exposes many drawbacks in this situation. On the one hand, event analysis is extremely difficult. Dispatchers cannot accurately extract key content from the massive amount of information in a short period of time, which easily leads to missed monitoring and poses a major threat to the safe operation of the power grid. On the other hand, existing alarm information is relatively discrete, making it difficult to fully and accurately analyze grid alarm information. With the widespread construction of smart substations and the extensive application of intelligent electronic devices, the breadth and magnitude of monitoring information from various devices have increased significantly, further exacerbating the complexity of information processing. Against this background, an efficient and intelligent grid alarm information classification system has emerged. Although existing technologies have made great progress in power grid alarms, there are still some problems that need to be optimized. Existing technologies make it difficult to combine multiple types of data to analyze the power grid, resulting in low accuracy in the intelligent classification of power grid alarm information. Therefore, how to combine multiple types of substation data to analyze abnormal power grid alarms and realize the association between intelligent classification information, precise positioning and disposal strategies of power grid alarms is the problem we need to solve. Summary of the Invention

[0003] To achieve the above objectives, the present invention is implemented through the following technical solutions: a power grid alarm information intelligent classification system based on deep learning, including a power grid alarm monitoring data acquisition module, a power grid alarm evaluation module, an information parsing module, a multi-dimensional feature analysis module, a disposal strategy allocation module and an intelligent classification module, wherein each module is communicatively connected; The power grid alarm monitoring data acquisition module is used to collect power grid alarm monitoring data including power outage monitoring data, section monitoring data, overload monitoring data, environmental monitoring data and positioning data, providing data guarantee for the implementation of subsequent module functions; The grid alarm evaluation module evaluates the grid alarm coefficient based on the pre-processed grid alarm monitoring data, providing a technical basis for analyzing the substation alarm level; The information parsing module analyzes the substation alarm level based on the assessed grid alarm coefficient, and then generates a grid alarm text. It also uses natural language processing technology to identify the grid alarm text, laying the foundation for constructing a multi-dimensional feature vector of the grid alarm. The multidimensional feature analysis module optimizes the recognition results of the power grid alarm text through the convolutional neural network algorithm, and then constructs the power grid alarm multidimensional feature vector; The handling strategy allocation module allocates corresponding power grid alarm handling strategies based on the multidimensional feature vector of the power grid alarm. It uses a bidirectional long short-term memory network to establish a mapping relationship between the multidimensional feature vector of the power grid alarm and the power grid alarm handling strategy, thereby realizing the association between the intelligent classification information of the power grid alarm information and its corresponding handling strategy. The intelligent classification module, based on the mapping relationship between the multi-dimensional feature vectors of power grid alarms and their corresponding power grid alarm handling strategies, combines positioning data to draw a knowledge graph of intelligent classification of power grid alarm information, integrates the latitude and longitude coordinates of each small substation area, and ultimately realizes the association between intelligent classification information of power grid alarms, precise positioning and handling strategies.

[0004] A further improvement of the technical solution of the present invention is that: the power grid alarm monitoring data acquisition module, the power grid alarm monitoring data acquisition process includes: The substation alarm monitoring area is divided into several small substation areas, and each small substation area and its transformer equipment, busbars, feeders and sections are numbered, so that the numbers of each small substation area are associated with the substation equipment numbers, busbar numbers, feeder numbers and section numbers within the area; Deploy different types of data collection equipment within each small substation area to collect power outage monitoring data, section monitoring data, overload monitoring data, environmental monitoring data, and positioning data. The data collection equipment includes a multi-function meter, temperature sensor, humidity sensor, ultrasonic anemometer, imaging spectrometer, laser ceilometer, electric field intensity meter, and GPS receiver. The power outage monitoring data includes bus voltage and feeder current; the section monitoring data includes active power, reactive power and frequency of the section; the overload monitoring data includes load current and operating temperature of substation equipment; the environmental monitoring data includes temperature, humidity, wind speed, reflectivity factor, cloud top height and electric field strength of each small substation area; the positioning data includes the latitude and longitude coordinates of each small substation area; Specifically, multi-function electricity meters are used to collect power outage monitoring data, cross-section monitoring data, and load current of substation equipment. Temperature sensors are used to collect the operating temperature of substation equipment and the temperature of the environment in each small substation area. Humidity sensors, ultrasonic anemometers, imaging spectrometers, laser ceilometers, and electric field intensity meters are used to collect the humidity, wind speed, reflectivity factor, cloud top height, and electric field intensity of each small substation area. GPS receivers are used to collect the longitude and latitude coordinates of each small substation area. The data source of the above-mentioned collected power grid monitoring data comes from the D5000 system. The collected power grid alarm monitoring data is then obtained through the docking of the dispatching data center with the D5000 system. Perform data cleansing on the collected power grid alarm monitoring data, assigns timestamps to the collected power grid alarm monitoring data, and combines with the power grid clock synchronization system to control the timestamp error of the power grid alarm monitoring data within 10ms. By adjusting the timestamp, the collection time of power outage monitoring data, section monitoring data, overload monitoring data, environmental monitoring data, and positioning data can be synchronized. The power outage monitoring data, section monitoring data, overload monitoring data and environmental monitoring data are integrated to generate a power grid alarm monitoring dataset, which is divided into a training set and a test set, with the ratio of the training set to the test set being 7:3.

[0005] A further improvement of the technical solution of the present invention is that: the evaluation process of the power grid alarm coefficient of the power grid alarm evaluation module includes: S1. The grid alarm coefficient includes a power outage alarm coefficient, a section alarm coefficient, an overload alarm coefficient and an environmental alarm coefficient; S2. Extract the power outage monitoring data from the power grid alarm monitoring data set, use the training set data in combination with the multivariate linear regression algorithm, take the power outage monitoring data as input and the power outage alarm coefficient as output, learn the linear relationship between the power outage monitoring data and the power outage alarm coefficient, and train the power grid power outage situation analysis model; Input the test set data into the power grid outage situation analysis model, adjust the parameters of the power grid outage situation analysis model, optimize the power grid outage situation analysis model, and obtain the final power grid outage situation analysis model; S3. Extract section monitoring data from the power grid alarm monitoring data set, use the training set data in combination with a multivariate linear regression algorithm, take the section monitoring data as input and the section alarm coefficient as output, learn the linear relationship between the section monitoring data and the section alarm coefficient, and train a power grid section situation analysis model; Input the test set data into the power grid section situation analysis model, adjust the parameters of the power grid section situation analysis model, optimize the power grid section situation analysis model, and obtain the final power grid section situation analysis model; S4. Combine the power outage monitoring data and the section monitoring data to output the corresponding power outage alarm coefficient and section alarm coefficient.

[0006] A further improvement of the technical solution of the present invention is that the evaluation process of the overload alarm coefficient of the power grid alarm evaluation module includes: The overload monitoring data from the power grid alarm monitoring data set is extracted. Using the training set data and the multivariate linear regression algorithm, the overload monitoring data is used as input and the overload alarm coefficient is used as output. The linear relationship between the overload monitoring data and the overload alarm coefficient is learned to train the power grid overload situation analysis model. The test set data is input into the power grid overload situation analysis model, the intercept term and regression coefficient of the power grid overload situation analysis model are adjusted, the power grid overload situation analysis model is optimized, and the final power grid overload situation analysis model is obtained. Combined with the overload monitoring data, the corresponding overload alarm coefficient is output.

[0007] A further improvement of the technical solution of the present invention is that the evaluation process of the environmental alarm coefficient of the power grid alarm evaluation module includes: The environmental monitoring data from the power grid alarm monitoring data set is extracted. The training set data is combined with the multivariate linear regression algorithm, which takes the environmental monitoring data as input and the environmental alarm coefficient as output. The linear relationship between the environmental monitoring data and the environmental alarm coefficient is learned to train the power grid environmental situation analysis model. The test set data is input into the power grid environment situation analysis model, the intercept term and regression coefficient of the power grid environment situation analysis model are adjusted, the power grid environment situation analysis model is optimized, the final power grid environment situation analysis model is obtained, and the corresponding environmental alarm coefficient is output in combination with the environmental monitoring data.

[0008] A further improvement of the technical solution of the present invention is that the information parsing module generates a power grid alarm text including: According to the power outage alarm coefficient, section alarm coefficient, overload alarm coefficient and environmental alarm coefficient, the power outage alarm level, section alarm level, overload alarm level and environmental alarm level of each small substation area are analyzed. Specifically, when the power outage alarm coefficient is lower than 0.4, it corresponds to the low power outage alarm level; when the power outage alarm coefficient is between 0.4 and 0.7, it corresponds to the medium power outage alarm level; when the power outage alarm coefficient is greater than 0.7, it corresponds to the high power outage alarm level; when the section alarm coefficient is lower than 0.3, it corresponds to the low section alarm level; when the section alarm coefficient is between 0.3 and 0.7, it corresponds to the high power outage alarm level. When the section alarm coefficient is greater than 0.7, it corresponds to the high section alarm level; when the overload alarm coefficient is lower than 0.2, it corresponds to the low overload alarm level; when the overload alarm coefficient is between 0.2 and 0.6, it corresponds to the medium overload alarm level; when the overload alarm coefficient is greater than 0.6, it corresponds to the high overload alarm level; when the environmental alarm coefficient is lower than 0.3, it corresponds to the low environmental alarm level; when the environmental alarm coefficient is between 0.3 and 0.6, it corresponds to the medium environmental alarm level; when the environmental alarm coefficient is greater than 0.6, it corresponds to the high environmental alarm level; The substation alarm level, each small substation area number, substation equipment number, busbar number, feeder number and section number are integrated to generate a power grid alarm text. The substation alarm level includes the power outage alarm level, section alarm level, overload alarm level and environmental alarm level of each small substation area. The power grid alarm text is responsible for classifying and arranging the power grid monitoring fault point numbers corresponding to various substation alarm levels, so that the alarm levels of various substations correspond to their power grid monitoring fault point numbers.

[0009] A further improvement of the technical solution of the present invention is that the process of the information parsing module identifying the power grid alarm text includes: The grid monitoring fault points are set to include each small substation area, substation equipment, busbar, feeder and section. The corresponding grid monitoring fault point number is composed of each small substation area number, substation equipment number, busbar number, feeder number and section number; Using natural language processing technology, the data in the power grid alarm text is labeled according to the substation alarm level entity type, the level value of the substation alarm, the entity type of the power grid monitoring fault point and the number value of the power grid monitoring fault point. Specifically, the power outage alarm level, section alarm level, overload alarm level and environmental alarm level are respectively labeled as the entity type of the substation alarm level, and high, medium and low are respectively labeled as the corresponding level values of the substation alarm; each small substation area, substation equipment, busbar, feeder and section are respectively labeled as the entity type of the power grid monitoring fault point, and each small substation area number, substation equipment number, busbar number, feeder number and section number are respectively labeled as the number value of the power grid monitoring fault point; Select a word segmentation tool and perform word segmentation on the data in the power grid alarm text based on the annotated results. The specific word segmentation tool can be HanLP or Jieba. Through word segmentation, the power grid alarm text is classified according to the substation alarm level entity type, substation alarm level value, power grid monitoring fault point entity type, and power grid monitoring fault point number value, laying the foundation for subsequent feature recognition and extraction operations. Entity recognition and word segmentation recognition are performed on the power grid alarm text after word segmentation processing, and the recognition result of the power grid alarm text is extracted. The extracted substation alarm level is numbered, and the recognition result of the power grid alarm text is integrated into the power grid alarm monitoring data set. The recognition result of the power grid alarm text is the substation alarm level number and its corresponding power grid monitoring fault point number. The substation alarm level number and its corresponding power grid monitoring fault point number are used to construct a multi-dimensional feature vector of the power grid alarm.

[0010] A further improvement of the technical solution of the present invention is that the multidimensional feature analysis module optimizes the recognition result of the power grid alarm text and then constructs the power grid alarm multidimensional feature vector process, which includes: The recognition results of power grid alarm texts are extracted from the power grid alarm monitoring dataset. A convolutional neural network algorithm is used to achieve hierarchical feature learning of the power grid alarm text recognition results through the combination of input layer, convolution layer, pooling layer, fully connected layer and output layer. Using training data, the substation alarm level number and its corresponding grid monitoring fault point number are used as input, and the grid alarm text correlation coefficient is used as output. The nonlinear relationship between the substation alarm level number, the grid monitoring fault point number and the grid alarm text correlation coefficient is learned to train a multidimensional optimization model for grid alarms. The test set data is input into the power grid alarm multidimensional optimization model, the parameters of the power grid alarm multidimensional optimization model are adjusted, the performance of the power grid alarm multidimensional optimization model is optimized, and the power grid alarm multidimensional optimization model is deployed into the system. The corresponding power grid alarm text correlation coefficient is output based on the substation alarm level number and its corresponding power grid monitoring fault point number. The power grid alarm text correlation coefficient describes the correlation between the substation alarm level number and the corresponding power grid monitoring fault point number, thereby optimizing the power grid alarm text recognition results. The substation alarm level number, power grid monitoring fault point number and the corresponding power grid alarm text correlation coefficient are combined to construct a power grid alarm multidimensional feature vector, which is then integrated into the power grid alarm monitoring dataset.

[0011] A further improvement of the technical solution of the present invention is that the process of the handling strategy allocation module allocating corresponding power grid alarm handling strategies and establishing a mapping relationship between the power grid alarm multidimensional feature vector and the power grid alarm handling strategy includes: Extract the grid alarm multidimensional feature vector from the grid alarm monitoring dataset. Match the corresponding substation alarm level and grid monitoring fault point based on the substation alarm level number and grid monitoring fault point number in the grid alarm multidimensional feature vector. Then, assign the corresponding grid alarm handling strategy to the grid monitoring fault point based on the substation alarm level. Specifically, the data source of the assigned grid alarm handling strategy comes from the historical grid alarm handling plan. The grid alarm handling frequency is set to every minute, and the grid alarm handling strategy is then obtained. Using training data, combined with the long short-term memory algorithm, the bidirectional loop algorithm, and the back-propagation algorithm, the multidimensional feature vector of the power grid alarm is used as input, and the power grid alarm handling strategy is used as output. The mapping relationship between the multidimensional feature vector of the power grid alarm and the power grid alarm handling strategy is learned, and a bidirectional long short-term memory network model is trained. The test set data is input into the bidirectional long short-term memory network model, the parameters of the bidirectional long short-term memory network model are adjusted, the performance of the bidirectional long short-term memory network model is optimized, the bidirectional long short-term memory network model is deployed into the system, and the corresponding power grid alarm handling strategy is output based on the multi-dimensional feature vector of the power grid alarm.

[0012] A further improvement of the technical solution of the present invention is that: the process of drawing the intelligent classification knowledge graph of power grid alarm information in the intelligent classification module includes: The positioning data is matched with the regional number of each small substation. Based on the mapping relationship between the grid alarm multidimensional feature vector and the grid alarm handling strategy, the grid alarm handling strategy and the positioning data are integrated into the grid alarm multidimensional feature vector. Integrate the multi-dimensional feature vectors of power grid alarms, power grid alarm handling strategies and positioning data to draw a knowledge graph for intelligent classification of power grid alarm information; The knowledge graph of intelligent classification of power grid alarm information consists of multi-dimensional feature vectors of power grid alarms, power grid alarm handling strategies and positioning data. It aims to monitor the alarm level of substations and accurately identify the specific power grid monitoring fault point number through the mapping relationship between the multi-dimensional feature vectors of power grid alarms and the power grid alarm handling strategies. It then obtains the specific small substation area where the substation alarm exists or the specific transformer equipment, busbar, feeder and section within the small substation area. Combined with the longitude and latitude coordinates of each small substation area, it accurately locates the power grid monitoring fault point corresponding to the power grid alarm. By analyzing multi-type substation data, it realizes intelligent classification and precise positioning of power grid alarm information, thereby improving the efficiency and accuracy of intelligent classification of power grid alarms.

[0013] The beneficial effects of the present invention are: a power grid alarm information intelligent classification system based on deep learning in the present invention, compared with the traditional power grid alarm information intelligent classification system based on deep learning, the data acquisition technology, power grid alarm evaluation technology, information analysis technology, multidimensional feature analysis technology, convolutional neural network algorithm, bidirectional long short-term memory network technology and knowledge graph drawing technology in the system of the present invention are closely integrated with modern information technology, accurately capturing power outage monitoring data, section monitoring data, overload monitoring data, environmental monitoring data and positioning data, and then obtaining power outage alarm coefficient, section alarm coefficient, overload alarm coefficient and environmental alarm coefficient, using the obtained power grid alarm coefficient to analyze the substation alarm level, achieving real-time and comprehensive monitoring of power grid alarm intelligent classification information, and then generating power grid alarm text, combined with natural language processing technology Combined with the convolutional neural network algorithm, a multi-dimensional feature vector of the power grid alarm is constructed, and then the corresponding power grid alarm handling strategy is assigned. A bidirectional long short-term memory network is used to establish a mapping relationship between the multi-dimensional feature vector of the power grid alarm and the power grid alarm handling strategy. Combined with the positioning data, a knowledge graph of intelligent classification of power grid information is generated, which solves the problem that the existing technology is difficult to combine multi-type substation data, analyze abnormal power grid alarm conditions, and realize the association between intelligent classification information of power grid alarms, precise positioning and handling strategies. It ensures that the method in the present invention can refine the dynamic monitoring standards for a power grid alarm information intelligent classification system based on deep learning within a more precise range, so that the monitored data becomes a more accurate indicator under the same conditions. The development and application of this method significantly enhances the level of intelligence in the process of intelligent classification of power grid alarm information based on deep learning. BRIEF DESCRIPTION OF THE DRAWINGS

[0014] In order to more clearly illustrate the embodiments of the present application or the technical solutions in the prior art, the following briefly introduces the drawings required for use in the embodiments. Obviously, the drawings described below are only some embodiments described in the present invention. For ordinary technicians in this field, other drawings can also be obtained based on these drawings.

[0015] Figure 1 This is a block diagram of a deep learning-based intelligent classification system for power grid alarm information in the present invention. DETAILED DESCRIPTION

[0016] To make the objectives, technical solutions, and advantages of the embodiments of the present invention more clear, the technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the accompanying drawings of the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. All other embodiments obtained by ordinary technicians in this field based on the embodiments of the present invention without making any creative efforts shall fall within the scope of protection of the present invention.

[0017] like Figure 1 As shown, the present invention provides a power grid alarm information intelligent classification system based on deep learning, including a power grid alarm monitoring data acquisition module, a power grid alarm evaluation module, an information parsing module, a multi-dimensional feature analysis module, a disposal strategy allocation module and an intelligent classification module, wherein each module is communicatively connected; The power grid alarm monitoring data acquisition module is used to collect power grid alarm monitoring data including power outage monitoring data, section monitoring data, overload monitoring data, environmental monitoring data and positioning data, providing data guarantee for the implementation of subsequent module functions; The power grid alarm assessment module evaluates the power grid alarm coefficient based on pre-processed power grid alarm monitoring data, providing a technical basis for analyzing substation alarm levels; The information analysis module analyzes the substation alarm level based on the assessed grid alarm coefficient, and then generates grid alarm text. It also uses natural language processing technology to identify the grid alarm text, laying the foundation for constructing a multi-dimensional feature vector for the grid alarm. The multi-dimensional feature analysis module uses a convolutional neural network algorithm to optimize the recognition results of power grid alarm text and then construct a multi-dimensional feature vector of the power grid alarm; The handling strategy allocation module allocates corresponding power grid alarm handling strategies based on the multidimensional feature vectors of power grid alarms. It uses a bidirectional long short-term memory network to establish a mapping relationship between the multidimensional feature vectors of power grid alarms and power grid alarm handling strategies, thus realizing the association between the intelligent classification information of power grid alarm information and its corresponding handling strategies. The intelligent classification module, based on the mapping relationship between the multi-dimensional feature vectors of power grid alarms and their corresponding power grid alarm handling strategies, combines positioning data to draw a knowledge graph of intelligent classification of power grid alarm information, integrating the latitude and longitude coordinates of each small substation area, and ultimately realizes the association between intelligent classification information of power grid alarms, precise positioning and handling strategies.

[0018] The power grid alarm monitoring data acquisition module collects power grid alarm monitoring data including: The substation alarm monitoring area is divided into several small substation areas, and each small substation area and its transformer equipment, busbars, feeders and sections are numbered, so that the numbers of each small substation area are associated with the substation equipment numbers, busbar numbers, feeder numbers and section numbers within the area; Deploy different types of data collection equipment within each small substation area to collect power outage monitoring data, cross-section monitoring data, overload monitoring data, environmental monitoring data, and positioning data. The data collection equipment includes multi-function meters, temperature sensors, humidity sensors, ultrasonic anemometers, imaging spectrometers, laser ceilometers, electric field strength meters, and GPS receivers. Power outage monitoring data includes bus voltage and feeder current; section monitoring data includes active power, reactive power, and frequency of the section; overload monitoring data includes load current and operating temperature of substation equipment; environmental monitoring data includes temperature, humidity, wind speed, reflectivity factor, cloud top height, and electric field strength of each small substation area; positioning data includes the longitude and latitude coordinates of each small substation area; Specifically, multi-function electricity meters are used to collect power outage monitoring data, cross-section monitoring data, and load current of substation equipment. Temperature sensors are used to collect the operating temperature of substation equipment and the temperature of the environment in each small substation area. Humidity sensors, ultrasonic anemometers, imaging spectrometers, laser ceilometers, and electric field intensity meters are used to collect the humidity, wind speed, reflectivity factor, cloud top height, and electric field intensity of each small substation area. GPS receivers are used to collect the longitude and latitude coordinates of each small substation area. The data source of the above-mentioned collected power grid monitoring data comes from the D5000 system. The collected power grid alarm monitoring data is then obtained through the docking of the dispatching data center with the D5000 system. Perform data cleansing on the collected power grid alarm monitoring data, assigns timestamps to the collected power grid alarm monitoring data, and combines with the power grid clock synchronization system to control the timestamp error of the power grid alarm monitoring data within 10ms. By adjusting the timestamp, the collection time of power outage monitoring data, section monitoring data, overload monitoring data, environmental monitoring data, and positioning data can be synchronized. The power outage monitoring data, section monitoring data, overload monitoring data and environmental monitoring data are integrated to generate a power grid alarm monitoring dataset, which is divided into a training set and a test set, with the ratio of the training set to the test set being 7:3.

[0019] The grid alarm evaluation module evaluates the grid alarm coefficient and the evaluation process includes: S1. The grid alarm coefficient includes a power outage alarm coefficient, a section alarm coefficient, an overload alarm coefficient and an environmental alarm coefficient; S2. Extract the power outage monitoring data from the power grid alarm monitoring data set, use the training set data in combination with the multivariate linear regression algorithm, take the power outage monitoring data as input and the power outage alarm coefficient as output, learn the linear relationship between the power outage monitoring data and the power outage alarm coefficient, and train the power grid power outage situation analysis model; Input the test set data into the power grid outage situation analysis model, adjust the parameters of the power grid outage situation analysis model, optimize the power grid outage situation analysis model, and obtain the final power grid outage situation analysis model; S3. Extract section monitoring data from the power grid alarm monitoring data set, use the training set data in combination with a multivariate linear regression algorithm, take the section monitoring data as input and the section alarm coefficient as output, learn the linear relationship between the section monitoring data and the section alarm coefficient, and train a power grid section situation analysis model; Input the test set data into the power grid section situation analysis model, adjust the parameters of the power grid section situation analysis model, optimize the power grid section situation analysis model, and obtain the final power grid section situation analysis model; S4. Combine the power outage monitoring data and the section monitoring data to output the corresponding power outage alarm coefficient and section alarm coefficient.

[0020] The grid alarm evaluation module evaluates the overload alarm coefficient in the following ways: The overload monitoring data from the power grid alarm monitoring data set is extracted. Using the training set data and the multivariate linear regression algorithm, the overload monitoring data is used as input and the overload alarm coefficient is used as output. The linear relationship between the overload monitoring data and the overload alarm coefficient is learned to train the power grid overload situation analysis model. The test set data is input into the power grid overload situation analysis model, the intercept term and regression coefficient of the power grid overload situation analysis model are adjusted, the power grid overload situation analysis model is optimized, and the final power grid overload situation analysis model is obtained. Combined with the overload monitoring data, the corresponding overload alarm coefficient is output.

[0021] The evaluation process of the power grid alarm evaluation module and the environmental alarm coefficient includes: The environmental monitoring data from the power grid alarm monitoring data set is extracted. The training set data is combined with the multivariate linear regression algorithm, which takes the environmental monitoring data as input and the environmental alarm coefficient as output. The linear relationship between the environmental monitoring data and the environmental alarm coefficient is learned to train the power grid environmental situation analysis model. The test set data is input into the power grid environment situation analysis model, the intercept term and regression coefficient of the power grid environment situation analysis model are adjusted, the power grid environment situation analysis model is optimized, the final power grid environment situation analysis model is obtained, and the corresponding environmental alarm coefficient is output in combination with the environmental monitoring data.

[0022] Information parsing module, the process of generating power grid alarm text includes: According to the power outage alarm coefficient, section alarm coefficient, overload alarm coefficient and environmental alarm coefficient, the power outage alarm level, section alarm level, overload alarm level and environmental alarm level of each small substation area are analyzed. Specifically, when the power outage alarm coefficient is lower than 0.4, it corresponds to the low power outage alarm level; when the power outage alarm coefficient is between 0.4 and 0.7, it corresponds to the medium power outage alarm level; when the power outage alarm coefficient is greater than 0.7, it corresponds to the high power outage alarm level; when the section alarm coefficient is lower than 0.3, it corresponds to the low section alarm level; when the section alarm coefficient is between 0.3 and 0.7, it corresponds to the high power outage alarm level. When the section alarm coefficient is greater than 0.7, it corresponds to the high section alarm level; when the overload alarm coefficient is lower than 0.2, it corresponds to the low overload alarm level; when the overload alarm coefficient is between 0.2 and 0.6, it corresponds to the medium overload alarm level; when the overload alarm coefficient is greater than 0.6, it corresponds to the high overload alarm level; when the environmental alarm coefficient is lower than 0.3, it corresponds to the low environmental alarm level; when the environmental alarm coefficient is between 0.3 and 0.6, it corresponds to the medium environmental alarm level; when the environmental alarm coefficient is greater than 0.6, it corresponds to the high environmental alarm level; The substation alarm level, each small substation area number, substation equipment number, busbar number, feeder number and section number are integrated to generate a power grid alarm text. The substation alarm level includes the power outage alarm level, section alarm level, overload alarm level and environmental alarm level of each small substation area. The power grid alarm text is responsible for classifying and arranging the power grid monitoring fault point numbers corresponding to various substation alarm levels, so that the alarm levels of various substations correspond to their power grid monitoring fault point numbers.

[0023] The information parsing module identifies the power grid alarm text in the following process: The grid monitoring fault points are set to include each small substation area, substation equipment, busbar, feeder and section. The corresponding grid monitoring fault point number is composed of each small substation area number, substation equipment number, busbar number, feeder number and section number; Using natural language processing technology, the data in the power grid alarm text is labeled according to the substation alarm level entity type, the level value of the substation alarm, the entity type of the power grid monitoring fault point and the number value of the power grid monitoring fault point. Specifically, the power outage alarm level, section alarm level, overload alarm level and environmental alarm level are respectively labeled as the entity type of the substation alarm level, and high, medium and low are respectively labeled as the corresponding level values of the substation alarm; each small substation area, substation equipment, busbar, feeder and section are respectively labeled as the entity type of the power grid monitoring fault point, and each small substation area number, substation equipment number, busbar number, feeder number and section number are respectively labeled as the number value of the power grid monitoring fault point; Select a word segmentation tool and perform word segmentation on the data in the power grid alarm text based on the annotated results. The specific word segmentation tool can be HanLP or Jieba. Through word segmentation, the power grid alarm text is classified according to the substation alarm level entity type, substation alarm level value, power grid monitoring fault point entity type, and power grid monitoring fault point number value, laying the foundation for subsequent feature recognition and extraction operations. Entity recognition and word segmentation recognition are performed on the power grid alarm text after word segmentation processing, and the recognition result of the power grid alarm text is extracted. The extracted substation alarm level is numbered, and the recognition result of the power grid alarm text is integrated into the power grid alarm monitoring data set. The recognition result of the power grid alarm text is the substation alarm level number and its corresponding power grid monitoring fault point number. The substation alarm level number and its corresponding power grid monitoring fault point number are used to construct a multi-dimensional feature vector of the power grid alarm.

[0024] The multi-dimensional feature analysis module optimizes the recognition results of power grid alarm text and constructs the multi-dimensional feature vector of power grid alarm. The process includes: The recognition results of power grid alarm texts are extracted from the power grid alarm monitoring dataset. A convolutional neural network algorithm is used to achieve hierarchical feature learning of the power grid alarm text recognition results through the combination of input layer, convolution layer, pooling layer, fully connected layer and output layer. Using training data, the substation alarm level number and its corresponding grid monitoring fault point number are used as input, and the grid alarm text correlation coefficient is used as output. The nonlinear relationship between the substation alarm level number, the grid monitoring fault point number and the grid alarm text correlation coefficient is learned to train a multidimensional optimization model for grid alarms. The test set data is input into the power grid alarm multidimensional optimization model, the parameters of the power grid alarm multidimensional optimization model are adjusted, the performance of the power grid alarm multidimensional optimization model is optimized, and the power grid alarm multidimensional optimization model is deployed into the system. The corresponding power grid alarm text correlation coefficient is output based on the substation alarm level number and its corresponding power grid monitoring fault point number. The power grid alarm text correlation coefficient describes the correlation between the substation alarm level number and the corresponding power grid monitoring fault point number, thereby optimizing the power grid alarm text recognition results. The substation alarm level number, power grid monitoring fault point number and the corresponding power grid alarm text correlation coefficient are combined to construct a power grid alarm multidimensional feature vector, which is then integrated into the power grid alarm monitoring dataset.

[0025] The processing strategy allocation module allocates corresponding power grid alarm processing strategies and establishes a mapping relationship between the power grid alarm multi-dimensional feature vector and the power grid alarm processing strategy. The process includes: Extract the grid alarm multidimensional feature vector from the grid alarm monitoring dataset. Match the corresponding substation alarm level and grid monitoring fault point based on the substation alarm level number and grid monitoring fault point number in the grid alarm multidimensional feature vector. Then, assign the corresponding grid alarm handling strategy to the grid monitoring fault point based on the substation alarm level. Specifically, the data source of the assigned grid alarm handling strategy comes from the historical grid alarm handling plan. The grid alarm handling frequency is set to every minute, and the grid alarm handling strategy is then obtained. Using training data, combined with the long short-term memory algorithm, the bidirectional loop algorithm, and the back-propagation algorithm, the multidimensional feature vector of the power grid alarm is used as input, and the power grid alarm handling strategy is used as output. The mapping relationship between the multidimensional feature vector of the power grid alarm and the power grid alarm handling strategy is learned, and a bidirectional long short-term memory network model is trained. The test set data is input into the bidirectional long short-term memory network model, the parameters of the bidirectional long short-term memory network model are adjusted, the performance of the bidirectional long short-term memory network model is optimized, the bidirectional long short-term memory network model is deployed into the system, and the corresponding power grid alarm handling strategy is output based on the multi-dimensional feature vector of the power grid alarm.

[0026] The intelligent classification module draws the knowledge graph of intelligent classification of power grid alarm information, including the following steps: The positioning data is matched with the regional number of each small substation. Based on the mapping relationship between the grid alarm multidimensional feature vector and the grid alarm handling strategy, the grid alarm handling strategy and the positioning data are integrated into the grid alarm multidimensional feature vector. Integrate the multi-dimensional feature vectors of power grid alarms, power grid alarm handling strategies and positioning data to draw a knowledge graph for intelligent classification of power grid alarm information; The knowledge graph of intelligent classification of power grid alarm information consists of multi-dimensional feature vectors of power grid alarms, power grid alarm handling strategies and positioning data. It aims to monitor the alarm level of substations and accurately identify the specific power grid monitoring fault point number through the mapping relationship between the multi-dimensional feature vectors of power grid alarms and the power grid alarm handling strategies. It then obtains the specific small substation area where the substation alarm exists or the specific transformer equipment, busbar, feeder and section within the small substation area. Combined with the longitude and latitude coordinates of each small substation area, it accurately locates the power grid monitoring fault point corresponding to the power grid alarm. By analyzing multi-type substation data, it realizes intelligent classification and precise positioning of power grid alarm information, thereby improving the efficiency and accuracy of intelligent classification of power grid alarms.

[0027] First, different types of acquisition equipment are combined to collect power outage monitoring data, section monitoring data, overload monitoring data, environmental monitoring data and positioning data, and the collected data are preprocessed and integrated to generate a power grid alarm monitoring data set; secondly, based on the preprocessed data, the power grid alarm coefficient is analyzed to obtain the substation alarm level, and then the power grid alarm text is generated, and the power grid alarm text is recognized using natural language processing technology; then, the recognition result of the power grid alarm text is optimized through the convolutional neural network algorithm, and a multi-dimensional feature vector of the power grid alarm is constructed; then, based on the multi-dimensional feature vector of the power grid alarm, the corresponding power grid alarm handling strategy is assigned, and a bidirectional long short-term memory network is used to establish a mapping relationship between the multi-dimensional feature vector of the power grid alarm and the power grid alarm handling strategy; finally, based on the mapping relationship between the multi-dimensional feature vector of the power grid alarm and its corresponding power grid alarm handling strategy, combined with the positioning data, a knowledge graph of intelligent classification of power grid alarm information is drawn.

[0028] The above description is merely a specific embodiment of the present application, but the scope of protection of the present application is not limited thereto. Any changes or substitutions that can be easily conceived by a person skilled in the art within the technical scope disclosed in this application should be included in the scope of protection of this application. Therefore, the scope of protection of this application should be based on the scope of protection of the claims.

Claims

1. A power grid alarm information intelligent classification system based on deep learning, including a power grid alarm monitoring data acquisition module, a power grid alarm evaluation module, an information analysis module, a multi-dimensional feature analysis module, a disposal strategy allocation module and an intelligent classification module, wherein: The modules are connected in communication, characterized by: The power grid alarm monitoring data acquisition module is used to collect power grid alarm monitoring data including power outage monitoring data, section monitoring data, overload monitoring data, environmental monitoring data and positioning data; The power grid alarm evaluation module evaluates the power grid alarm coefficient based on the pre-processed power grid alarm monitoring data; The information analysis module analyzes the substation alarm level based on the evaluated power grid alarm coefficient, generates a power grid alarm text, and uses natural language processing technology to identify the power grid alarm text; The multidimensional feature analysis module optimizes the recognition results of the power grid alarm text through the convolutional neural network algorithm, and then constructs the power grid alarm multidimensional feature vector; The processing strategy allocation module allocates corresponding power grid alarm processing strategies based on the power grid alarm multidimensional feature vector, and uses a bidirectional long short-term memory network to establish a mapping relationship between the power grid alarm multidimensional feature vector and the power grid alarm processing strategy; The intelligent classification module draws a knowledge graph of intelligent classification of power grid alarm information based on the mapping relationship between the multi-dimensional feature vector of the power grid alarm and its corresponding power grid alarm handling strategy, combined with the positioning data.

2. The deep learning-based intelligent classification system for power grid alarm information according to claim 1, characterized in that: The power grid alarm monitoring data acquisition module collects power grid alarm monitoring data including: The substation alarm monitoring area is divided into several small substation areas, and each small substation area and its transformer equipment, busbars, feeders and sections are numbered, so that the numbers of each small substation area are associated with the substation equipment numbers, busbar numbers, feeder numbers and section numbers within the area; Deploy different types of data collection equipment within each small substation area to collect power outage monitoring data, section monitoring data, overload monitoring data, environmental monitoring data, and positioning data. The data collection equipment includes a multi-function meter, temperature sensor, humidity sensor, ultrasonic anemometer, imaging spectrometer, laser ceilometer, electric field intensity meter, and GPS receiver. The power outage monitoring data includes bus voltage and feeder current; the section monitoring data includes active power, reactive power and frequency of the section; the overload monitoring data includes load current and operating temperature of substation equipment; the environmental monitoring data includes temperature, humidity, wind speed, reflectivity factor, cloud top height and electric field strength of each small substation area; the positioning data includes the latitude and longitude coordinates of each small substation area; Perform data cleansing on the collected power grid alarm monitoring data, assigns timestamps to the collected power grid alarm monitoring data, and combines with the power grid clock synchronization system to control the timestamp error of the power grid alarm monitoring data within 10ms. By adjusting the timestamp, the collection time of power outage monitoring data, section monitoring data, overload monitoring data, environmental monitoring data, and positioning data can be synchronized. The power outage monitoring data, section monitoring data, overload monitoring data and environmental monitoring data are integrated to generate a power grid alarm monitoring data set, which is then divided into a training set and a test set.

3. The deep learning-based intelligent classification system for power grid alarm information according to claim 2, characterized in that: The grid alarm evaluation module, the grid alarm coefficient evaluation process includes: S1. The grid alarm coefficient includes a power outage alarm coefficient, a section alarm coefficient, an overload alarm coefficient and an environmental alarm coefficient; S2. Extract the power outage monitoring data from the power grid alarm monitoring data set, use the training set data in combination with the multivariate linear regression algorithm, take the power outage monitoring data as input and the power outage alarm coefficient as output, learn the linear relationship between the power outage monitoring data and the power outage alarm coefficient, and train the power grid power outage situation analysis model; Input the test set data into the power grid outage situation analysis model, adjust the parameters of the power grid outage situation analysis model, optimize the power grid outage situation analysis model, and obtain the final power grid outage situation analysis model; S3. Extract section monitoring data from the power grid alarm monitoring data set, use the training set data in combination with a multivariate linear regression algorithm, take the section monitoring data as input and the section alarm coefficient as output, learn the linear relationship between the section monitoring data and the section alarm coefficient, and train a power grid section situation analysis model; Input the test set data into the power grid section situation analysis model, adjust the parameters of the power grid section situation analysis model, optimize the power grid section situation analysis model, and obtain the final power grid section situation analysis model; S4. Combine the power outage monitoring data and the section monitoring data to output the corresponding power outage alarm coefficient and section alarm coefficient.

4. The deep learning-based intelligent classification system for power grid alarm information according to claim 3, characterized in that: The overload alarm coefficient evaluation process of the power grid alarm evaluation module includes: The overload monitoring data from the power grid alarm monitoring data set is extracted. Using the training set data and the multivariate linear regression algorithm, the overload monitoring data is used as input and the overload alarm coefficient is used as output. The linear relationship between the overload monitoring data and the overload alarm coefficient is learned to train the power grid overload situation analysis model. The test set data is input into the power grid overload situation analysis model, the intercept term and regression coefficient of the power grid overload situation analysis model are adjusted, the power grid overload situation analysis model is optimized, and the final power grid overload situation analysis model is obtained. Combined with the overload monitoring data, the corresponding overload alarm coefficient is output.

5. The deep learning-based intelligent classification system for power grid alarm information according to claim 4, characterized in that: The evaluation process of the environmental alarm coefficient of the power grid alarm evaluation module includes: The environmental monitoring data from the power grid alarm monitoring data set is extracted. The training set data is combined with the multivariate linear regression algorithm, which takes the environmental monitoring data as input and the environmental alarm coefficient as output. The linear relationship between the environmental monitoring data and the environmental alarm coefficient is learned to train the power grid environmental situation analysis model. The test set data is input into the power grid environment situation analysis model, the intercept term and regression coefficient of the power grid environment situation analysis model are adjusted, the power grid environment situation analysis model is optimized, the final power grid environment situation analysis model is obtained, and the corresponding environmental alarm coefficient is output in combination with the environmental monitoring data.

6. The deep learning-based intelligent classification system for power grid alarm information according to claim 5, characterized in that: The information parsing module generates the power grid alarm text in the following steps: Analyze the power outage alarm level, section alarm level, overload alarm level, and environmental alarm level of each small substation area based on the power outage alarm coefficient, section alarm coefficient, overload alarm coefficient, and environmental alarm coefficient; The substation alarm level, each small substation area number, substation equipment number, busbar number, feeder number and section number are integrated to generate a power grid alarm text. The substation alarm level includes the power outage alarm level, section alarm level, overload alarm level and environmental alarm level of each small substation area.

7. The deep learning-based intelligent classification system for power grid alarm information according to claim 6, characterized in that: The information parsing module identifies the power grid alarm text in the following steps: The grid monitoring fault points are set to include each small substation area, substation equipment, busbar, feeder and section. The corresponding grid monitoring fault point number is composed of each small substation area number, substation equipment number, busbar number, feeder number and section number; Using natural language processing technology, the data in the power grid alarm text is annotated according to the substation alarm level entity type, substation alarm level value, power grid monitoring fault point entity type and power grid monitoring fault point number value; Select a word segmentation tool and perform word segmentation on the data in the power grid alarm text according to the annotation results of the power grid alarm text; Entity recognition and word segmentation recognition are performed on the power grid alarm text after word segmentation processing, and the recognition result of the power grid alarm text is extracted. The extracted substation alarm level is numbered, and the recognition result of the power grid alarm text is integrated into the power grid alarm monitoring data set. The recognition result of the power grid alarm text is the substation alarm level number and its corresponding power grid monitoring fault point number.

8. The deep learning-based intelligent classification system for power grid alarm information according to claim 7, characterized in that: The multi-dimensional feature analysis module optimizes the recognition result of the power grid alarm text and then constructs the multi-dimensional feature vector of the power grid alarm, including: The recognition results of power grid alarm texts are extracted from the power grid alarm monitoring dataset. A convolutional neural network algorithm is used to achieve hierarchical feature learning of the power grid alarm text recognition results through the combination of input layer, convolution layer, pooling layer, fully connected layer and output layer. Using training data, the substation alarm level number and its corresponding grid monitoring fault point number are used as input, and the grid alarm text correlation coefficient is used as output. The nonlinear relationship between the substation alarm level number, the grid monitoring fault point number and the grid alarm text correlation coefficient is learned to train a multidimensional optimization model for grid alarms. Input the test set data into the power grid alarm multidimensional optimization model, adjust the parameters of the power grid alarm multidimensional optimization model, optimize the performance of the power grid alarm multidimensional optimization model, deploy the power grid alarm multidimensional optimization model into the system, and combine the substation alarm level number and its corresponding power grid monitoring fault point number to output the corresponding power grid alarm text correlation coefficient; The substation alarm level number, power grid monitoring fault point number and the corresponding power grid alarm text correlation coefficient are combined to construct a power grid alarm multidimensional feature vector, which is then integrated into the power grid alarm monitoring dataset.

9. The deep learning-based intelligent classification system for power grid alarm information according to claim 8, characterized in that: The process of the processing strategy allocation module allocating corresponding power grid alarm processing strategies and establishing a mapping relationship between the power grid alarm multi-dimensional feature vector and the power grid alarm processing strategy includes: Extract the grid alarm multidimensional feature vector from the grid alarm monitoring dataset, match the corresponding substation alarm level and its grid monitoring fault point according to the substation alarm level number and grid monitoring fault point number in the grid alarm multidimensional feature vector, and assign the corresponding grid alarm handling strategy to the grid monitoring fault point according to the substation alarm level; Using training data, combined with the long short-term memory algorithm, the bidirectional loop algorithm, and the back-propagation algorithm, the multidimensional feature vector of the power grid alarm is used as input, and the power grid alarm handling strategy is used as output. The mapping relationship between the multidimensional feature vector of the power grid alarm and the power grid alarm handling strategy is learned, and a bidirectional long short-term memory network model is trained. The test set data is input into the bidirectional long short-term memory network model, the parameters of the bidirectional long short-term memory network model are adjusted, the performance of the bidirectional long short-term memory network model is optimized, the bidirectional long short-term memory network model is deployed into the system, and the corresponding power grid alarm handling strategy is output based on the multi-dimensional feature vector of the power grid alarm.

10. The deep learning-based intelligent classification system for power grid alarm information according to claim 9, characterized in that: The intelligent classification module, the process of drawing the knowledge graph of intelligent classification of power grid alarm information includes: The positioning data is matched with the regional number of each small substation. Based on the mapping relationship between the grid alarm multidimensional feature vector and the grid alarm handling strategy, the grid alarm handling strategy and the positioning data are integrated into the grid alarm multidimensional feature vector. Integrate the multi-dimensional feature vectors of power grid alarms, power grid alarm handling strategies and positioning data, and draw a knowledge graph of intelligent classification of power grid alarm information.

Images