Automatic control system and method for power system signal monitoring

Abstract

The present invention discloses an automatic control system and method for power system signal monitoring, which utilizes a multimodal system to process various monitoring data of fault phenomena, locations, states and other information, and provides multiple dimensions for analyzing the state and faults of equipment in the power system. The knowledge graph is introduced to effectively integrate, analyze, reason and query a large amount of knowledge, and provide support for fault diagnosis and operation control of the power system. At the same time, multidimensional output is achieved based on the enhanced knowledge graph and neural network, and multiple related outputs are generated, which improves the representation ability and application scope, and provides better risk prediction and other services for power workers, which can reduce the dependence on the workers' own experience and improve the efficiency of fault analysis, processing and prevention.

Description

Technical Field

[0001] The present invention relates to the technical field of power system informatization, and in particular to an automatic control system and method for power system signal monitoring. Background Art

[0002] The power system has corresponding signal monitoring and control systems in the power generation, transmission, transformation, distribution and power consumption links to measure, adjust, control, protect, communicate and dispatch to ensure the reliability of power supply, ensure the quality of power and improve the economy of power system operation. With the development of technologies such as sensors, Internet of Things, cloud computing, big data, and artificial intelligence, the application of intelligent and automated technologies in power systems has gradually become popular. At the same time, with the expansion of the scale of power grids, the development of power informatization and the construction of intelligent power grids, the monitoring and control of power systems has become a common concern of power departments and users. In order to avoid abnormalities in the operation of the power system, the staff must confirm the current operating data in the system through the monitoring system and analyze and control its operating status in real time. In the existing technology, professional and experienced maintenance personnel are usually required to analyze and determine the cause of the fault when receiving abnormal system operation signals, and restore the operation of the system in a short time.

[0003] The monitoring and maintenance of the power system requires both professional technical field knowledge and maintenance and repair knowledge, as well as various monitoring data reflecting the fault location and status emitted by monitoring equipment (such as sensors) when a fault occurs. Only by effectively combining fault knowledge with monitoring data can we better judge the fault problem and locate the fault location. At the same time, due to its huge scale, massive equipment types and high safety requirements, the power system requires staff to conduct repeated inspections to ensure the safe and stable operation of the system. For example, in the prior art, the patent application with publication number CN109582657A collects historical data of capacitor failures and parameters affecting capacitor failures, and inputs them into a deep neural network for guided training after data processing, thereby predicting capacitor failures. Among them, the above technical solution is still insufficient for the exploration of multidimensional data with complex relationships and the use of large-scale structured knowledge. Summary of the invention

[0004] The present invention focuses on the problems existing in the above-mentioned prior art and provides an automatic control system and method for power system signal monitoring, which can use a multimodal system to monitor various signals in the operation process of power equipment, introduce large-scale structured knowledge through knowledge graphs, and ensure that when abnormalities occur in the monitoring data, the cause of the fault can be quickly and accurately determined, maintenance can be carried out in time, and risk prevention and control can be carried out, thereby improving operating efficiency, reducing the risk of failures, reducing the workload of staff in power work, and reducing costs and increasing efficiency.

[0005] The present invention firstly provides an automatic control system and method for monitoring power system signals.

[0006] The present invention provides an automatic control method for power system signal monitoring, comprising the following steps:

[0007] Step 1: Acquire signals through a monitoring module to obtain multimodal monitoring data, and send the multimodal monitoring data to a processing module; wherein the monitoring module includes a monitoring device, and the monitoring device is used to obtain the multimodal monitoring data of the operation of equipment in the power system;

[0008] Step 2: the processing module preprocesses the multimodal monitoring data, and inputs the preprocessed data into a feature extraction module, the feature extraction module sets a feature extractor for each mode in the multimodal monitoring data, and inputs the multimodal monitoring data into the feature extractor to extract features;

[0009] Step 3: Connect the extracted features of different modes to obtain fused features, and input the fused features into a machine learning model to obtain a device status result;

[0010] Step 4: Convert the device state result into a device state embedding vector;

[0011] Step 5: Create a knowledge graph in the power field;

[0012] Step 6: Processing the electric power field knowledge graph to obtain an enhanced electric power field knowledge graph;

[0013] Step 7: Combining the device state embedding vector, the enhanced power field knowledge graph and the multi-output neural network to obtain a multi-dimensional output result;

[0014] Step 8: Obtain maximum probability fault information, high risk factor fault information and fault prediction information from the multi-dimensional output results, obtain the correlation between the maximum probability fault information and the high risk factor fault information, and recommend maintenance strategies to the staff.

[0015] In step 1, the multimodal monitoring data includes audio monitoring data, video monitoring data, image monitoring data, and sensor feedback numerical monitoring data; for the monitoring of a certain type of power equipment, the signal acquisition frequency of the mode with a small amount of feedback information is increased, and the signal acquisition frequency of the mode with a large amount of feedback information is reduced, and the number of monitoring devices of the certain type of power equipment is inversely proportional to the corresponding signal acquisition frequency.

[0016] The creation of the power field knowledge graph in step 5 is specifically as follows: step 501: collecting and determining the initial setting parameter set of the power system; wherein, it also includes collecting multimodal data sources of indicators, images, videos or audios of different monitoring devices for signal acquisition of power equipment; the data source set of the power field knowledge graph includes structured data, text, images, videos and audio data; step 502: performing knowledge extraction on the above text, images, videos and audio data respectively; step 503: performing knowledge storage, knowledge calculation and visualization to construct the power field knowledge graph.

[0017] Among them, the processing of the electric power field knowledge graph in step 6 to obtain the enhanced electric power field knowledge graph is specifically as follows: step 601: given two selected entities in the electric power field knowledge graph and candidate relationships between the selected two entities, there are multiple paths between the selected two entities connected by the candidate relationships; step 602: after obtaining the multiple paths between the selected two entities, use a convolutional neural network to process the multiple paths to obtain an output vector, input the output vector into a bidirectional long short-term memory module Bi-LSTM, and use the bidirectional long short-term memory module Bi-LSTM to obtain the paths between the selected two entities; The multiple paths are converted into embedding vectors, and the multiple paths of the selected two entities are encoded as low-dimensional embeddings; step 603: a self-attention mechanism is used to capture the semantic relevance between the candidate relationship and each path between the selected two entities, the extracted embedding vectors are predicted through the attention layer and the output layer, the similarity value of each path is calculated, and it is predicted whether the entities should be connected through the candidate relationship; step 604: the attention layer extracts inference evidence from the paths related to the candidate relationship, adds triples with high confidence, and forms the enhanced power field knowledge graph.

[0018] In the step 7, the multi-dimensional output result obtained by combining the device state embedding vector, the enhanced power field knowledge graph and the multi-output neural network is specifically as follows: Step 701: extracting entities and relationships in the enhanced power field knowledge graph, embedding the semantic information of the entities and the relationships into a low-dimensional vector space, and obtaining an entity embedding vector; Step 702: embedding the device state result into a low-dimensional vector space, obtaining a device state embedding vector, and mapping the device state embedding vector to the same vector space as the entity embedding vector; Step 703: inputting the device state embedding vector and the entity embedding vector into a multi-output neural network to obtain an output result.

[0019] In step 8, the maximum probability fault information is the fault cause and location information with the highest result probability, the high risk coefficient fault information is the fault cause and location information with the most serious loss, and the correlation is the degree of correlation between the two; if the maximum probability fault information and the high risk coefficient fault information indicate the same fault, the maximum probability fault information is presented to the staff, and the maintenance strategy of the maximum probability fault information is recommended; if the two are not the same fault and the degree of correlation between the two is low, the maximum probability fault information and the high risk coefficient fault information are presented to the staff in order, and the maintenance strategy is recommended to give priority to the fault corresponding to the maximum probability fault information; if the correlation between the two is high, the recommended maintenance strategy is to simultaneously handle the fault corresponding to the maximum probability fault information and the high risk coefficient fault information; the fault prediction information is the information of the power equipment with the highest probability of fault in advance, and the recommended maintenance strategy is to increase the signal acquisition frequency of the power equipment with the highest probability of fault. Through the above method, the intelligent risk prevention and control capability is improved, which helps the staff to resolve major risks or related risks in a timely manner.

[0020] The present invention also protects an automatic control system for power system signal monitoring, including a monitoring module, a processing module, a feature extraction module, a feature fusion module, an analysis module, and an output module; the monitoring module includes a monitoring device, the monitoring device is used to obtain the multimodal monitoring data of the equipment operation in the power system; the processing module pre-processes the multimodal monitoring data; the feature extraction module sets a feature extractor for each mode in the multimodal monitoring data, and inputs the multimodal monitoring data into the feature extractor to extract features; the feature fusion module connects the extracted features of different modes to obtain fused features, and inputs the fused features into the machine learning model to obtain the equipment status result; the analysis module combines the equipment status embedding vector, the enhanced power field knowledge graph and the multi-output neural network to obtain a multidimensional output result; the output module obtains the maximum probability fault information, high risk factor fault information and fault prediction information from the multidimensional output result, and obtains the correlation between the maximum probability fault information and the high risk factor fault information, and recommends maintenance strategies to the staff. It also includes a security module for protecting the data security between the control device and the power equipment.

[0021] Among them, the above-mentioned monitoring device includes one or more of an audio monitoring device, a video monitoring device, an image capture device, a temperature sensor, a vibration sensor, an optical fiber sensor, a load sensor, a current transformer, a partial discharge sensor, and a humidity monitoring sensor.

[0022] The present invention further provides a computer-readable storage medium having a computer program stored thereon, wherein the computer program, when executed by a processor, implements the steps of an automatic control method for monitoring power system signals.

[0023] The present invention has the following beneficial effects:

[0024] The present invention obtains corresponding monitoring signals through various monitoring devices. The monitoring signals include images, sensor feedback values, videos, audio signals, etc. Different data modes are expressed in different ways, and there will be some information redundancy and information complementarity, which makes up for the limitations of single-mode information. Compared with the prior art, the present invention utilizes the information complementarity characteristics of different data modes to improve the accuracy and effectiveness of data analysis, and provides multiple dimensions for analyzing the status and faults of equipment in the power system, etc., so as to obtain more feature-rich information and improve the performance of the entire model. The present invention introduces knowledge graphs, which enables the model to utilize structured knowledge in large-scale knowledge graphs. Especially in power systems, equipment failures are complex in form, and there is sometimes a causal relationship between the previous and subsequent failures, which can easily cause significant losses. The knowledge graph helps to further perform efficient and accurate diagnosis. At the same time, the present invention realizes multi-dimensional output based on knowledge graphs and neural networks, thereby generating multiple related outputs, improving the representation ability and application scope, providing better risk prediction services for power workers, reducing the dependence on the workers' own experience, and improving the efficiency of fault analysis, processing and prevention. BRIEF DESCRIPTION OF THE DRAWINGS

[0025] In order to more clearly illustrate the technical solutions in the embodiments of the present invention, the prior art and the drawings required for use in the embodiments are briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without creative work.

[0026] Figure 1 It is a flow chart of the automatic control method for power system signal monitoring of the present invention;

[0027] Figure 2 It is a framework diagram of the automatic control method for power system signal monitoring of the present invention;

[0028] Figure 3 It is a module diagram of the automatic control system for power system signal monitoring of the present invention. DETAILED DESCRIPTION

[0029] The present invention is described in detail below in conjunction with specific embodiments. The following embodiments will help those skilled in the art to further understand the present invention, but are not intended to limit the present invention in any form. It should be noted that, for those of ordinary skill in the art, several variations and improvements may be made without departing from the concept of the present invention. These all belong to the protection scope of the present invention.

[0030] my country's power system has gradually developed into the world's largest transmission capacity, the highest transmission voltage level, and a multi-regional AC / DC hybrid power system. Sudden failures and power outages can easily cause serious consequences, even endangering the safety of people and equipment, and causing huge economic losses. The power system must first ensure reliable and uninterrupted power supply during operation. Therefore, real-time monitoring of power equipment is particularly important, such as Figure 1 As shown, the present invention provides an automatic control method for power system signal monitoring, the method comprising the following steps:

[0031] Step 1: Collect signals through the monitoring module to obtain multimodal monitoring data, and send the multimodal monitoring data to the processing module; wherein the monitoring module includes a monitoring device, and the monitoring device is used to obtain the multimodal monitoring data of the equipment operation in the power system.

[0032] Unlike traditional models that focus on a single type of data, multimodal systems can process and understand information in multiple modes, such as sensor data, images, text, audio, video, etc., and use the synergy between different modes to more comprehensively understand the input. The power system needs to monitor many devices and has high monitoring requirements. Different sensors have different working characteristics, detection capabilities and reliability in different environments. The safety, reliability and accuracy of the measurement results of the equipment status monitoring system directly affect the effective monitoring of the equipment and the effective implementation of maintenance and repair strategies. Different sensing technologies are widely used, with mixed results. In the present invention, the monitoring device can obtain sensor information, image information, video information, audio information and other data. For example, for transmission lines, monitoring equipment such as temperature sensors, load sensors, and image capture devices can be used to ensure that the status of the transmission lines can be obtained in all kinds of weather environments.

[0033] In the above step 1, the multimodal monitoring data includes audio monitoring data, video monitoring data, image monitoring data, and sensor feedback numerical monitoring data; for the monitoring of a certain type of power equipment, the signal acquisition frequency of the mode with a small amount of feedback information is increased, and the signal acquisition frequency of the mode with a large amount of feedback information is reduced, and the number of monitoring devices of the certain type of power equipment is inversely proportional to the corresponding signal acquisition frequency.

[0034] The use of multimodal systems for monitoring will obtain a large amount of data. It is necessary to conduct multi-level and multi-space information complementarity and optimal combination processing of various modal collection methods, and strive for energy conservation and emission reduction on the basis of obtaining complete monitoring information.

[0035] For example, when using temperature sensors, load sensors and image monitoring to monitor the icing of power transmission lines, on the one hand, the advantages of traditional electronic sensors such as temperature sensors and load sensors are that the amount of numerical information transmitted back is small and does not occupy a large communication bandwidth, but the disadvantage is that the measurement accuracy is limited. On the other hand, the advantage of image monitoring is high monitoring accuracy, but the disadvantage is that the amount of information generated is large, the required signal transmission performance is also high, and it is difficult to work normally at night and in extreme weather. When applying the above-mentioned multiple different modes, using them in combination according to the amount of information transmitted back is conducive to more efficient monitoring.

[0036] In one embodiment, for monitoring of certain power equipment, the acquisition frequency of modes with a small amount of returned information can be increased, and the acquisition frequency of modes with a large amount of returned information can be reduced. Specifically, when monitoring the icing of transmission lines, a higher sampling frequency is set for the temperature sensor and the load sensor, and a low sampling frequency is set for the image monitoring device. For another example, for certain key equipment, such as circuit breakers, which mainly play a protective role in the power system and are vital to the safe operation of the power system, when a circuit breaker fails, it may cause power grid accidents and cause the scope of the fault to expand, affecting the entire system and causing serious losses. In this regard, sound sensors, vibration sensors, and current signal acquisition devices can be used to obtain multimodal data. For current signals, the features at the time points of the set time period are extracted, and for sound and vibration signals, only the features at the mutation time points are extracted.

[0037] For some power equipment, if multiple monitoring devices are used for signal collection, considering factors such as energy saving and improving processing efficiency, the number of modes can also be set to be inversely proportional to the signal collection frequency. That is to say, the more modes the monitoring device has, the lower the signal collection frequency. Conversely, if the monitoring device has fewer modes, the signal collection frequency will be higher.

[0038] Step 2: The processing module preprocesses the multimodal monitoring data, and inputs the preprocessed data into the feature extraction module. The feature extraction module sets a feature extractor for each mode in the multimodal monitoring data, and inputs the multimodal monitoring data into the feature extractor to extract features. Data of different modes need to be processed in different ways for feature extraction and fusion. For each mode in the multimodality, a feature extractor is set to extract features from the respective input data, thereby converting the multimodal monitoring data into a feature vector that can be understood by the learning model. For example, convolutional neural networks and recurrent neural networks can be used to extract features for image data and audio data.

[0039] Step 3: Connect the extracted features of different modes to obtain fused features, and input the fused features into the machine learning model to obtain the device status results.

[0040] For example, by performing multimodal feature fusion, the extracted features of different modes can be connected into a single high-dimensional feature vector to obtain fused features, and the fused features are input into a classifier to obtain the equipment status results. Specifically, for the vibration signal of the high-voltage circuit breaker, there are certain differences in the sound signals of the circuit breaker in different states. The sound signal sensor and the circuit breaker can be used to collect sound data in a non-contact installation, and then the support vector machine can be used to identify the extracted features to obtain equipment status results such as electromagnet jamming, closing spring fatigue, and increased damping.

[0041] Step 4: Convert the device state result into a device state embedding vector.

[0042] Step 5: Create a knowledge graph in the power field. The knowledge graph is an efficient data structure in which data is stored in the form of nodes and edges. The nodes represent objects and the edges represent the relationship between objects, so that the storage of knowledge is no longer isolated, but interrelated and supportive. Since the knowledge graph only stores important data points, this reduces the retrieval time complexity and reduces the space complexity. The knowledge graph is also a way of expressing and storing knowledge, which can provide rich and structured background knowledge for machine learning, thereby improving the understanding and reasoning ability of the algorithm. The knowledge graph provides a machine-readable way of expressing knowledge, so that artificial intelligence systems can integrate, reason and query knowledge more effectively. Using the knowledge graph, a large amount of data can be integrated and analyzed to provide support for fault diagnosis and operation control of power systems.

[0043] Step 6: Process the power field knowledge graph to obtain an enhanced power field knowledge graph.

[0044] The core components of the knowledge graph include entities, relationships, and attributes. Entities are the basic units in the knowledge graph, relationships describe the various connections between entities, and attributes are specific descriptions of entities. Due to its characteristics, the power sector needs to process and improve the knowledge graph, so the power sector knowledge graph is processed to obtain an enhanced power sector knowledge graph.

[0045] Step 7: Combine the device state embedding vector, the enhanced power field knowledge graph and the multi-output neural network to obtain a multi-dimensional output result. Figure 2 As shown, the use of a multi-output neural network can overcome the limitations of the prior art in dealing with complex problems. The multi-output neural network can extract information from multiple aspects and output corresponding results at the same time.

[0046] Step 8: Obtain maximum probability fault information, high risk factor fault information and fault prediction information from the multi-dimensional output results, obtain the correlation between the maximum probability fault information and the high risk factor fault information, and recommend maintenance strategies to the staff.

[0047] Preferably, the maximum probability fault information is the fault cause and location information with the highest result probability, the high risk coefficient fault information is the fault cause and location information with the most serious loss degree, and the correlation is the degree of correlation between the two; if the maximum probability fault information and the high risk coefficient fault information indicate the same fault, the maximum probability fault information is presented to the staff, and a maintenance strategy for the maximum probability fault information is recommended; if the two are not the same fault and the degree of correlation between the two is low, the maximum probability fault information and the high risk coefficient fault information are presented to the staff in sequence, and a maintenance strategy is recommended to give priority to the fault corresponding to the maximum probability fault information; if the correlation between the two is high, the recommended maintenance strategy is to simultaneously process the faults corresponding to the maximum probability fault information and the high risk coefficient fault information; the fault prediction information is information that predicts in advance the power equipment with the highest probability of fault, and the recommended maintenance strategy is to increase the signal acquisition frequency for the power equipment with the highest probability of fault.

[0048] The core components of the knowledge graph include entities, relationships and attributes. Entities are the basic units in the knowledge graph, relationships describe the various connections between entities, and attributes are specific descriptions of entities. The creation of the power field knowledge graph in step 5 is specifically as follows:

[0049] Step 501: Collect and determine the initial setting parameter set of the power system; which also includes collecting multimodal data sources of indicators, images, videos or audios of different monitoring devices for signal acquisition of power equipment; the data source set of the power field knowledge graph includes the initial setting parameter set and multimodal data sources of indicators, images, videos or audios of different monitoring devices for signal acquisition of power equipment, covering structured data, text, images, videos and audio data. The selection of the data source set directly affects the quality and application scope of the knowledge graph. The power field knowledge graph belongs to the knowledge graph of a specific field. When selecting the data source set, the reliability, relevance, and integrity of the data should be considered, and multiple data sources should be combined to obtain more comprehensive and in-depth knowledge coverage.

[0050] The initial setting parameter set of the power system includes power dictionaries, national standards, industry standards, standards within the power grid company, equipment manufacturers, etc., which mostly exist in the form of structured documents and semi-structured data tables, and the data documents are imported, read and stored. In addition, the data source set for collecting the knowledge graph in the power field also includes a multimodal data source that collects indicators, images, videos or audios of different monitoring devices for signal acquisition for a single power device; the traditional knowledge graph is mainly constructed based on text information and structured data. The present invention further uses data in various forms such as pictures, audio, and video to increase the richness of the data. In the power system, there are many types of sensor devices. For important equipment, multiple sensors can be set to monitor the operating status of the equipment at the same time. For the same power equipment, such as transmission lines, although machine vision can obtain more information, in bad weather conditions such as low visibility, optical fiber sensors are needed to collect information. Even if a sensor monitoring data has a high correlation with the potential fault of the power equipment, it may only contain part of the fault information and cannot fully reflect the problem of the equipment. By using different forms of data sources, complementary information sources can be integrated and summarized to form a composite knowledge graph. Collecting indicators, images, audio, video and other information from different monitoring devices such as sensors and the data between them and faults, and establishing correlations and complementary relationships between them can more accurately characterize the overall status characteristics of the equipment and the relationship between fault problems.

[0051] Step 502: perform knowledge extraction on the above text, image, video and audio data respectively.

[0052] Step 503: Perform knowledge storage, knowledge calculation and visualization to construct the power field knowledge graph. After collecting the data source, data preprocessing is performed. For example, the preprocessing of images is to label the images with text labels corresponding to the entity names. Knowledge extraction is performed on the collected data, including entity extraction, relationship extraction, attribute extraction, etc. For example, the output of relationship extraction is usually a triple. Through knowledge extraction, a relationship triple or an attribute triple is obtained.

[0053] In the power system, after a fault such as a circuit breaker failure to operate or a component group outage occurs, a chain of power outages may occur, with one fault causing subsequent faults. For example, transmission lines are often long and have many branches. If the faults that occur in the transmission line during operation cannot be prevented or located in time, the time to repair the fault point will be extended, causing more serious faults in the distribution system, such as a line short circuit, causing a fire, etc., or there may be multiple fault points in the same cable at different distances from the measuring end, forming a multi-point fault. Therefore, improving the relationship between different components, between different faults, between components and faults, etc., will help improve the diagnostic level of the knowledge graph.

[0054] Preferably, the enhanced electric power knowledge graph obtained by processing the electric power knowledge graph in step 6 is specifically as follows: Step 601: given two selected entities in the electric power knowledge graph and candidate relationships between the two selected entities, there are multiple paths between the two selected entities connected by the candidate relationships. For example, given two entities (short circuit, transmission line), there may be a candidate relationship (short circuit, transmission line, collision) between the two entities, and the reasoning path between the two entities may include but is not limited to, for example, a short circuit caused by insulation damage due to aging of the transmission line, a short circuit caused by deformation and relaxation of the transmission line under the action of strong wind, a short circuit caused by collision due to the transmission line being installed too low and various transport items or metal items, a short circuit caused by biological cross-connection on the transmission line, etc.

[0055] Step 602: After obtaining the multiple paths between the selected two entities, the multiple paths are processed by a convolutional neural network to obtain an output vector, the output vector is input into a bidirectional long short-term memory module Bi-LSTM, the multiple paths are converted into an embedding vector using the bidirectional long short-term memory module Bi-LSTM, and the multiple paths of the selected two entities are encoded into a low-dimensional embedding. Path encoding is performed, and multiple filters with the same window size are slid on the path using a one-dimensional convolution operator to generate features to obtain a vector sequence.

[0056] Step 603: Use the self-attention mechanism to capture the semantic relevance between the candidate relationship and each path between the two selected entities, predict the extracted embedding vector through the attention layer and the output layer, calculate the similarity value of each path, and predict whether the entities should be connected through the candidate relationship. Not all paths between entities represent the relationship between them. In order to measure the importance of the reasoning path to the path combination, the self-attention mechanism is introduced to calculate the similarity value of all paths. Paths that are more relevant to the candidate relationship are given high weights, and paths that are irrelevant to the candidate relationship are given low weights.

[0057] Step 604: The attention layer extracts inference evidence from the paths related to the candidate relationships, adds triples with high confidence, and forms the enhanced power field knowledge graph.

[0058] When obtaining the reasoning results through the knowledge graph, for simple faults, the optimal single triple can often be directly found as the answer to the question. For complex faults, it is necessary to generate the answer to the question by fusing multiple triples through knowledge reasoning. The multi-dimensional output result obtained by combining the device state embedding vector, the enhanced power field knowledge graph and the multi-output neural network in step 7 is specifically:

[0059] Step 701: extract entities and relationships in the enhanced power field knowledge graph, embed semantic information of the entities and the relationships into a low-dimensional vector space, and obtain an entity embedding vector.

[0060] Step 702: Embed the device state result into a low-dimensional vector space to obtain a device state embedding vector, and map the device state embedding vector to the same vector space as the entity embedding vector. Among them, through embedding, the relationships and entities in the knowledge graph can be mapped to a dense low-dimensional space, capturing the semantic information and structural information in the knowledge graph, and the similarity between entities and relationships is represented by the embedded distance and direction, which is beneficial to the calculation and reasoning of the knowledge graph. The embedded vector space can directly use existing machine learning algorithms and neural networks for training and reasoning.

[0061] Step 703: Input the device state embedding vector and the entity embedding vector into a multi-output neural network to obtain an output result.

[0062] In one embodiment, the entity embedding vector and the device state embedding vector can be concatenated to generate a combined feature matrix, the combined embedding vector is obtained through a combination of a convolutional neural network and a feedforward neural network, and then the similarity value between the combined embedding vector and the candidate entity embedding vector is calculated, and the entity with the highest value is used as an output result.

[0063] In step 8, the maximum probability fault information is the fault cause and location information with the highest result probability, the high risk coefficient fault information is the fault cause and location information with the most serious loss, and the correlation is the degree of correlation between the two; if the maximum probability fault information and the high risk coefficient fault information indicate the same fault, the maximum probability fault information is presented to the staff, and a maintenance strategy for the maximum probability fault information is recommended; if the two are not the same fault and the degree of correlation between the two is low, the maximum probability fault information and the high risk coefficient fault information are presented to the staff in sequence, and a maintenance strategy is recommended to give priority to the fault corresponding to the maximum probability fault information; if the correlation between the two is high, the recommended maintenance strategy is to simultaneously handle the faults corresponding to the maximum probability fault information and the high risk coefficient fault information.

[0064] Among them, the acquisition of high-risk coefficient fault information is obtained by comprehensive judgment based on the importance of the power equipment, the possibility of causing chain failures, and the severity of the losses caused. The importance of power equipment is related to the number of modes of its sensors. When multiple modes are used for signal acquisition in a certain power equipment, a higher importance weight is given to the power equipment. At the same time, when the failure of a certain power equipment will cause multiple point failures or cause significant economic losses, a high level of risk weight is given to the power equipment. In this way, high-risk coefficient fault information that is likely to cause serious consequences can be provided to the staff in a timely manner to improve safety assurance. The fault prediction information is the information of the power equipment with the highest probability of failure in advance, and the recommended maintenance strategy is to increase the signal acquisition frequency of the power equipment with the highest probability of failure. When it is predicted that the power equipment at a certain location may fail, the signal acquisition frequency for the power equipment is controlled to be increased. For example, when it is predicted that the transmission line may have an icing problem, the image monitoring frequency is increased to obtain more detection signals and troubleshoot in advance.

[0065] In an exemplary embodiment, Figure 3As shown, an automatic control system for power system signal monitoring is provided, including a monitoring module, a processing module, a feature extraction module, a feature fusion module, an analysis module, and an output module. The monitoring module includes a monitoring device, and the monitoring device is used to obtain the multimodal monitoring data of the equipment operation in the power system. Among them, the monitoring module is composed of a power generation monitoring unit, a transmission monitoring unit, a substation monitoring unit or a distribution monitoring unit. In one embodiment, the power generation monitoring unit may include an electric quantity measurement module and a non-electrical measurement module. The electric quantity measurement module mainly uses electronic and optical fiber transformers to detect voltage and current during the power generation process, and the non-electrical measurement unit mainly uses optical fiber vibration sensors to detect state variables such as generator speed and vibration during the power generation process. The transmission monitoring unit mainly uses various devices such as machine vision, optical fiber sensors, and temperature sensors to monitor the dancing, icing, and tilt of the transmission line. Substation monitoring units mainly include transformer monitoring units, circuit breaker monitoring units, capacitive equipment monitoring units, voltage and current monitoring units, etc. Transformer monitoring units mainly include partial discharge monitoring modules, temperature monitoring modules, mechanical characteristics monitoring modules, grounding performance monitoring modules, etc. Circuit breaker monitoring units mainly include temperature monitoring modules, partial discharge monitoring modules, switch status monitoring modules, etc. The distribution monitoring unit is composed of a large number of small sensors, which can realize the perception, collection and processing of various monitoring information in the network coverage area.

[0066] The processing module preprocesses the multimodal monitoring data. The feature extraction module sets a feature extractor for each mode in the multimodal monitoring data, and inputs the multimodal monitoring data into the feature extractor to extract features. The feature fusion module connects the extracted features of different modes to obtain fused features, and inputs the fused features into the machine learning model to obtain the equipment status results. The analysis module combines the equipment status embedding vector, the enhanced power field knowledge graph and the multi-output neural network to obtain a multidimensional output result. The output module obtains the maximum probability fault information and high risk coefficient fault information from the multidimensional output results, and obtains the correlation between the maximum probability fault information and the high risk coefficient fault information, and recommends maintenance strategies to the staff.

[0067] It also includes a security module to protect the data security between the control device and the power equipment, including data encryption and firewalls.

[0068] Each module in the automatic control system for power system signal monitoring can be implemented by software, hardware and a combination thereof. Preferably, a computer readable storage medium is provided, on which a computer program is stored, and when the computer program is executed by a processor, the steps in the above method embodiments are implemented.

[0069] The above-described embodiments only express several implementation methods of the present application, and the descriptions thereof are relatively specific and detailed, but they cannot be understood as limiting the scope of the present application. It should be pointed out that, for a person of ordinary skill in the art, several variations and improvements can be made without departing from the concept of the present application, and these all belong to the protection scope of the present application. Therefore, the protection scope of the present application shall be subject to the attached claims.

Claims

1. An automatic control method for power system signal monitoring, The following steps are involved: Step 1: Acquire signals through a monitoring module to obtain multimodal monitoring data, and send the multimodal monitoring data to a processing module; wherein the monitoring module includes a monitoring device, and the monitoring device is used to obtain the multimodal monitoring data of the operation of equipment in the power system; For monitoring the first power equipment, a plurality of the monitoring devices are used to collect signals, the signal collection frequency of the mode with a small amount of returned information is increased, and the signal collection frequency of the mode with a large amount of returned information is reduced, and they are used in coordination according to the amount of returned information, and the number of the monitoring devices of the first power equipment is inversely proportional to the corresponding signal collection frequency; Step 2: the processing module preprocesses the multimodal monitoring data, and inputs the preprocessed multimodal monitoring data into a feature extraction module, the feature extraction module sets a feature extractor for each mode in the preprocessed multimodal monitoring data, and inputs the preprocessed multimodal monitoring data into the feature extractor to extract features; Step 3: Connect the extracted features of different modalities to obtain fused features, and input the fused features into a machine learning model to obtain a device status result; Step 4: Convert the device state result into a device state embedding vector; Step 5: Create a knowledge graph in the power field; Step 6: Processing the electric power field knowledge graph to obtain an enhanced electric power field knowledge graph; Step 7: Combining the device state embedding vector, the enhanced power field knowledge graph and the multi-output neural network to obtain a multi-dimensional output result; specifically: Step 701: extracting entities and relationships in the enhanced electric power field knowledge graph, embedding semantic information of the entities and the relationships into a low-dimensional vector space, and obtaining an entity embedding vector; Step 702: embed the device state result into a low-dimensional vector space to obtain a device state embedding vector, and map the device state embedding vector to the same vector space as the entity embedding vector; Step 703: input the device state embedding vector and the entity embedding vector into a multi-output neural network to obtain an output result; Step 8: Obtain maximum probability fault information, high risk factor fault information and fault prediction information from the multi-dimensional output results, obtain the correlation between the maximum probability fault information and the high risk factor fault information, and recommend maintenance strategies to the staff; Among them, the maximum probability fault information is the fault cause and location information with the highest result probability, the high risk coefficient fault information is the fault cause and location information with the most serious loss, and the correlation is the degree of correlation between the two; if the maximum probability fault information and the high risk coefficient fault information indicate the same fault, the maximum probability fault information is presented to the staff, and a maintenance strategy for the maximum probability fault information is recommended; if the two are not the same fault and the degree of correlation between the two is low, the maximum probability fault information and the high risk coefficient fault information are presented to the staff in order, and the recommended maintenance strategy is to give priority to the fault corresponding to the maximum probability fault information; if the correlation between the two is high, the recommended maintenance strategy is to simultaneously process the faults corresponding to the maximum probability fault information and the high risk coefficient fault information; the fault prediction information is information for predicting in advance the power equipment with the highest probability of fault, and the recommended maintenance strategy is to increase the signal acquisition frequency for the power equipment with the highest probability of fault.

2. The automatic control method for power system signal monitoring according to claim 1, Features: The specific steps of creating the knowledge graph in the power field in step 5 are as follows: Step 501: Collect and determine an initial set of parameters for the power system, and collect multi-modal data sources of indicators, images, videos or audios of different monitoring devices for signal collection of power equipment; Step 502: performing knowledge extraction on the text, image, video and audio data collected and determined in step 501 respectively; Step 503: Perform knowledge storage, knowledge calculation and visualization to construct the power field knowledge graph.

3. The automatic control method for power system signal monitoring according to claim 2, Features: The enhanced power field knowledge graph obtained by processing the power field knowledge graph in step 6 is specifically as follows: Step 601: given two selected entities in the electric power field knowledge graph and candidate relationships between the two selected entities, there are multiple paths between the two selected entities connected by the candidate relationships; Step 602: after obtaining the multiple paths between the selected two entities, use a convolutional neural network to process the multiple paths to obtain an output vector, input the output vector into a bidirectional long short-term memory module Bi-LSTM, use the bidirectional long short-term memory module Bi-LSTM to convert the multiple paths into an embedding vector, and encode the multiple paths of the selected two entities into a low-dimensional embedding; Step 603: using a self-attention mechanism to capture the semantic relevance between the candidate relationship and each path between the two selected entities, predicting the extracted embedding vector through the attention layer and the output layer, calculating the similarity value of each path, and predicting whether the entities should be connected through the candidate relationship; Step 604: The attention layer extracts inference evidence from the paths related to the candidate relationships, adds triples with high confidence, and forms the enhanced power field knowledge graph.

4. The automatic control method for power system signal monitoring according to claim 1, Features: In step 1, the multimodal monitoring data includes audio monitoring data, video monitoring data, image monitoring data, and sensor feedback numerical monitoring data.

5. An automatic control system for power system signal monitoring, applying the method according to claim 1, comprising a monitoring module, a processing module, a feature extraction module, a feature fusion module, an analysis module, and an output module; The monitoring module includes a monitoring device, and the monitoring device is used to obtain the multimodal monitoring data of the operation of the equipment in the power system; The processing module pre-processes the multimodal monitoring data; The feature extraction module sets a feature extractor for each mode in the multimodal monitoring data, and inputs the multimodal monitoring data into the feature extractor to extract features; The feature fusion module connects the extracted features of different modalities to obtain fused features, and inputs the fused features into a machine learning model to obtain a device status result; The analysis module combines the device state embedding vector, the enhanced power field knowledge graph and the multi-output neural network to obtain a multi-dimensional output result; The output module obtains maximum probability fault information, high risk factor fault information and fault prediction information from the multi-dimensional output results, and obtains the correlation between the maximum probability fault information and the high risk factor fault information, and recommends maintenance strategies to the staff.

6. The automatic control system for power system signal monitoring according to claim 5, Features: It also includes a security module for protecting data security between the control device and the power equipment.

7. The automatic control system for power system signal monitoring according to claim 5, Features: The monitoring device includes one or more of an audio monitoring device, a video monitoring device, an image capturing device, a temperature sensor, a vibration sensor, an optical fiber light sensor, a load sensor, a current transformer, a partial discharge sensor, and a humidity monitoring sensor.

8. A computer-readable storage medium having a computer program stored thereon, It is characterized in that When the computer program is executed by a processor, the steps of the method according to any one of claims 1 to 4 are implemented.

Images