Fire early warning method, device, medium and system for transformer substation

Abstract

The invention provides a fire early warning method, device, medium and system for a transformer substation, and the method comprises the steps: continuously adjusting the weight and threshold of each layer in a neural network model through a neural network technology and a stochastic gradient descent algorithm, obtaining a final neural network model, calculating the fireless, open fire and smoldering fire probability and the fire hazard degree, and carrying out the early warning of the fire hazard. Multi-sensor detection parameters and changes thereof serve as a fire judgment data basis, according to optical time domain reflection and a backward Raman scattering temperature effect of an optical fiber, the defects that a BP algorithm is generally low in learning efficiency, low in convergence speed and prone to sinking into a local minimum state are avoided, and finally, according to an output result and a comparison result of a final neural network model, a fire judgment result is obtained. According to the method, whether the early warning information is generated or not is determined, the fire early warning accuracy is improved, and the problem that the early warning accuracy is low due to the fact that a fire early warning method of a transformer substation in an existing scheme can only collect a single monitoring parameter but abnormity of the single monitoring parameter cannot fully indicate fire occurrence is solved.

Description

Technical Field

[0001] The present application relates to the technical field of fire warning for substations, and in particular to a fire warning method, device, medium and system for substations. Background Art

[0002] Unmanned substations can not only monitor and manage remote power systems, but also improve the intelligence and automation of power systems. Unmanned substations play a vital role in modern power transmission and distribution systems. However, unmanned substations have many safety risks such as accidents and fires, one of the most serious risks of which is fire accidents. Unmanned substations are different from traditional substations. Unmanned substations lack personnel inspection links. Once a fire accident occurs, it is likely to have a serious impact on the substation. The traditional fire protection setting method of substations cannot detect fire hazards early, and there is no fast and efficient technical means for fire confirmation. For large substations, manpower cannot fully control the fire protection status in the field. The incidence of fire directly affects the normal operation of the entire substation. Since most substations are unmanned substations, it is inconvenient to confirm whether the detector alarm information is accurate and whether false alarms occur using traditional technical means, and it is even more difficult to say whether the fire extinguishing is effective. Therefore, how to effectively prevent and detect fire accidents in unmanned substations has become one of the important issues in the safety protection of unmanned substations.

[0003] However, the traditional single-sensor fire detectors are widely used in engineering. In particular, the general substation fire detection alarm device is mainly based on ordinary smoke and temperature detectors that only detect single parameters such as temperature or smoke. However, with the expansion of the application of automatic fire detection systems, the traditional method of using smoke concentration and temperature to identify smoke particles can no longer meet the application requirements. Because of the existence of non-fire particles, the particle characteristics are very unstable, and the collected signals have a certain degree of randomness, which also makes it impossible for any fixed detection algorithm to effectively identify smoke particles. Therefore, it is impossible to identify only from its original signal. Since the scattered light intensity of different particles after being stimulated is affected by external noise and system noise, it has randomness and uncertainty. The sensor detects the development of fire too single, and different types and different fire conditions monitored by different sensors are often misjudged. Misjudgment must be avoided, and information misjudgment is fatal. Since the fire detection of traditional single sensors is to collect single fire parameter information at the detection site and adopt a simple threshold algorithm to judge the occurrence of fire. This detector based on a single physical quantity of fire information is difficult to accurately reflect the overall characteristics of fires with different manifestations. In addition, the non-uniformity of the sensitivity of the detector to the fire characteristic signal, the randomness and uncertainty of the fire signal, the detection of a single parameter is prone to false alarms, missed alarms and late alarms of fire warnings. Fires in reality are diverse and have great randomness and instability. The abnormality of a single monitoring parameter cannot fully explain the occurrence of a fire. Only when multiple parameters have a certain degree of abnormality can it be said that a fire may occur. Fire phenomena in the early stages and false fire phenomena are often mixed together, and the fire parameters collected by the sensor are signals that are unknown or uncertain in advance. They not only change with time, but other environmental changes and electronic noise may also cause them to change. Moreover, this change is often basically similar to the characteristics of fire parameter changes. When the installation location and environmental characteristics of the detector cannot be determined in advance, no matter what kind of fixed algorithm is used, it cannot fully meet the requirements. Therefore, fire detection is a nonlinear structure problem, which is difficult to accurately describe using classical mathematical modeling methods.

[0004] Bad weather can easily cause sensor aging and failure, rain and snow may blur the camera, and communication anomalies in extreme weather will also bring great difficulties to data collection and transmission. At present, most data centers only use one fire characteristic parameter to judge the fire situation for fire detection, and the on-duty personnel in the monitoring room need to activate the fire extinguishing switch to start the fire extinguishing action. The disadvantage of this method is that it is impossible to carry out preliminary fire extinguishing and contain the development of the fire in the first time. When the monitoring room is unmanaged, unattended, and unattended, the fire host is always in a closed state, and the fire fighting equipment cannot play an effective fire fighting function. The traditional intelligent auxiliary control system is limited by the influence of technology and concepts. The online monitoring system, fire alarm system, security system, and fire fighting system of primary equipment are independent of each other, and the communication protocols are not unified, forming an island, and it is impossible to achieve linkage and coordination between equipment. Not only does it increase equipment costs and debugging costs, but it is also difficult to meet operation and maintenance requirements in the later stage.

[0005] The existing scheme will alarm when physical quantities such as smoke and temperature reach the specified state. When the alarm is issued, the fire has generally occurred or the fire is large, which has caused harm and losses, which is not conducive to initial control; the equipment has a single state when it is pre-alarming, which is divided into alarm and fault states. It does not have real-time data monitoring functions, cannot perform data analysis, and cannot judge trend changes. Due to the lack of intermediate data, it does not have the function of hidden danger discovery and troubleshooting; when information is transmitted externally, it only has the upload function, not the download control function, and the amount of information is single, which is not conducive to unified monitoring and management, and cannot be remotely controlled; the logical relationship relied on during fire extinguishing linkage is the status information of the equipment, and the on-site situation cannot be verified twice; maintenance and management are generally recorded in traditional paper documents, and the management methods are backward, which does not meet the requirements of rapid response and the development of the big data era. The substation smart fire protection system integrated with the power Internet of Things also includes a 4-layer framework: perception layer, network layer, platform layer, and application layer, which is the specific practice of the power Internet of Things. But it is not isolated from the solution and equipment selection, but the fire protection integrated with the power Internet of Things is part of the power Internet of Things.

[0006] That is, the fire warning method of the substation in the existing solution can only collect a single monitoring parameter, but the abnormality of a single monitoring parameter cannot fully explain the occurrence of a fire, resulting in low accuracy of the warning. Summary of the invention

[0007] The main purpose of the present application is to provide a fire warning method, device, medium and system for a substation, so as to at least solve the problem that the fire warning method for a substation in the existing solution can only collect a single monitoring parameter, but the abnormality of a single monitoring parameter cannot fully explain the occurrence of a fire, thereby resulting in low accuracy of the warning.

[0008] In order to achieve the above object, according to one aspect of the present application, a fire early warning method for a substation is provided, the method comprising:

[0009] Acquire fire hazard data collected by at least three fire sensors, and perform normalization processing on the fire hazard data to obtain normalized fire hazard data, wherein the fire sensors include point smoke fire detectors, linear smoke fire detectors, temperature fire detectors, flame detectors, smoke sensors, microwave sensors, and temperature sensors;

[0010] Using the smoke particle intelligent recognition algorithm, the electrical fire hazards of the line are identified through edge computing, and the normalized fire hazard data is fused based on multimodal data to identify the fire hazards of the substation and obtain multimodal fusion data;

[0011] Taking the multimodal fusion data as the input of the neural network model, the weights and thresholds of each layer in the neural network model are continuously adjusted through the neural network technology and the stochastic gradient descent algorithm to obtain the final neural network model, and the output of the final neural network model includes the probability of no fire, the probability of open fire, the probability of smoldering fire and the degree of fire hazard;

[0012] According to the optical time-domain reflection and backward Raman scattering temperature effects of the optical fiber, the current real-time temperature measurement value is compared with the preset alarm parameter value to obtain a comparison result. According to the output result and comparison result of the final neural network model, it is determined whether to generate an early warning information.

[0013] Optionally, normalizing the fire hazard data to obtain normalized fire hazard data includes:

[0014] according to Determining the normalized fire hazard data;

[0015] Among them, f(x) is the normalized fire hazard data, x0 is the minimum value of the range in the fire hazard data, a is the normalization coefficient, and x is the characteristic parameter measurement value in the fire hazard data.

[0016] Optionally, in the process of using the multimodal fusion data as the input of the neural network model and continuously adjusting the weights and thresholds of each layer in the neural network model through the neural network technology and the stochastic gradient descent algorithm, the method further includes:

[0017] The neural network model is symmetrically connected using a probabilistic graphical model to obtain the parameters corresponding to the minimum energy function of the neural network model, and the probability distribution of the probabilistic graphical model is calculated by the energy function to obtain a probability value to optimize the parameters of the probabilistic graphical model;

[0018] At the same time, the weights are randomly initialized, and the initial values ​​of the weights are optimized and pre-trained to optimize the initial values ​​of the weights.

[0019] Optionally, in the process of fusing the normalized fire hazard data based on the multimodal data to identify the fire hazard of the substation and obtain the multimodal fusion data, the method further includes:

[0020] When the fusion method of the multimodal data is data-level fusion, the data of the structured wide-area measurement system and the data acquisition and monitoring system are fused based on the data-level fusion, and the visible light and infrared image data of the unstructured substation power equipment are fused at the same time.

[0021] Optionally, in the process of fusing the normalized fire hazard data based on the multimodal data to identify the fire hazard of the substation and obtain the multimodal fusion data, the method further includes:

[0022] When the fusion mode of the multimodal data is feature-level fusion, the hidden fault is diagnosed based on the fusion of the current amplitude distortion, the fault energy value and the switch quantity information characteristics, and the feature extraction of the multimodal data is performed according to the data characteristics, physical meaning and fusion target;

[0023] Convolutional neural networks are used to fuse homogeneous data and heterogeneous data with different features, and to fuse unstructured images and structured stress data to extract image data texture, dynamic features, structured current data wavelet features and information entropy features.

[0024] Optionally, in the process of fusing the normalized fire hazard data based on the multimodal data to identify the fire hazard of the substation and obtain the multimodal fusion data, the method further includes:

[0025] In the case where the fusion mode of the multimodal data is decision-level fusion, the modal data of the same group or different modal data are fused with different decision results to perform multimodal data fusion;

[0026] The substation is determined as a node, and the transmission line connecting the substation is determined as an edge. Each point directly transmits to the gateway center, and the attributes of the node and the edge are set as the network topology structure.

[0027] Optionally, determining whether to generate warning information according to the output result of the final neural network model and the comparison result includes:

[0028] Acquire a comparison mapping relationship, wherein the comparison mapping relationship represents a mapping relationship between an output result of the final neural network model, the comparison result, and a judgment result, wherein the judgment result represents whether to generate the warning information;

[0029] Whether to generate the warning information is determined according to the comparison mapping relationship, the output result of the final neural network model and the comparison result.

[0030] According to another aspect of the present application, a fire warning device for a substation is provided, the device comprising:

[0031] an acquisition unit, configured to acquire fire hazard data collected by at least three fire sensors, and normalize the fire hazard data to obtain normalized fire hazard data, wherein the fire sensors include point-type smoke detectors, linear smoke detectors, temperature detectors, flame detectors, smoke sensors, microwave sensors, and temperature sensors;

[0032] The first processing unit is used to use a smoke particle intelligent recognition algorithm to identify electrical fire hazards of the line through edge computing, and to fuse the normalized fire hazard data based on multimodal data to identify fire hazards of the substation and obtain multimodal fusion data;

[0033] A second processing unit is used to use the multimodal fusion data as an input of the neural network model, and continuously adjust the weights and thresholds of each layer in the neural network model through the neural network technology and the stochastic gradient descent algorithm to obtain a final neural network model, wherein the output of the final neural network model includes a probability of no fire, a probability of open fire, a probability of smoldering fire, and a degree of fire hazard;

[0034] The third processing unit is used to compare the current real-time temperature measurement value with the preset alarm parameter value based on the optical time domain reflection and backward Raman scattering temperature effect of the optical fiber to obtain a comparison result, and determine whether to generate an early warning message based on the output result of the final neural network model and the comparison result.

[0035] According to another aspect of the present application, a computer-readable storage medium is provided, wherein the computer-readable storage medium includes a stored program, wherein when the program is executed, the device where the computer-readable storage medium is located is controlled to execute any one of the methods described.

[0036] According to another aspect of the present application, a fire warning system for a substation is provided, the system comprising: one or more processors, a memory, and one or more programs, wherein the one or more programs are stored in the memory and are configured to be executed by the one or more processors, and the one or more programs include methods for executing any one of the methods described.

[0037] By applying the technical solution of the present application and using at least three fire sensors, subsequent judgments can be made more consistent with actual working conditions. The normalized fire hazard data is fused based on multimodal data to identify fire hazards in substations and obtain multimodal fusion data. The multimodal fusion data is used as the input of the neural network model. The weights and thresholds of each layer in the neural network model are continuously adjusted through neural network technology and stochastic gradient descent algorithm to obtain the final neural network model. The probability of no fire, open fire, smoldering fire and the degree of fire hazard are calculated. The multi-sensor detection parameters and their changes are used as the basis of fire judgment data. According to the optical time domain reflection and back Raman scattering temperature effect of the optical fiber, the disadvantages of the BP algorithm, such as low learning efficiency, slow convergence speed and easy to fall into the local minimum state, are avoided. Finally, according to the output result and comparison result of the final neural network model, it is determined whether to generate warning information, thereby improving the accuracy of fire warning, thereby solving the problem that the fire warning method of the substation in the existing solution can only collect a single monitoring parameter, but the abnormality of a single monitoring parameter cannot fully explain the occurrence of a fire, resulting in low accuracy of the warning. BRIEF DESCRIPTION OF THE DRAWINGS

[0038] The drawings constituting part of the present application are used to provide a further understanding of the present application. The illustrative embodiments and descriptions of the present application are used to explain the present application and do not constitute an improper limitation on the present application. In the drawings:

[0039] Figure 1 A schematic flow chart of a fire warning method for a substation provided according to an embodiment of the present application is shown;

[0040] Figure 2 A structural block diagram of a fire warning device for a substation provided according to an embodiment of the present application is shown. DETAILED DESCRIPTION

[0041] It should be noted that, in the absence of conflict, the embodiments and features in the embodiments of the present application can be combined with each other. The present application will be described in detail below with reference to the accompanying drawings and in combination with the embodiments.

[0042] In order to enable those skilled in the art to better understand the solution of the present application, the technical solution in the embodiments of the present application will be clearly and completely described below in conjunction with the drawings in the embodiments of the present application. Obviously, the described embodiments are only part of the embodiments of the present application, not all of the embodiments. Based on the embodiments in the present application, all other embodiments obtained by ordinary technicians in this field without creative work should fall within the scope of protection of the present application.

[0043] It should be noted that the terms "first", "second", etc. in the specification and claims of the present application and the above-mentioned drawings are used to distinguish similar objects, and are not necessarily used to describe a specific order or sequential order. It should be understood that the data used in this way can be interchanged where appropriate, so that the embodiments of the present application described here. In addition, the terms "including" and "having" and any of their variations are intended to cover non-exclusive inclusions, for example, a process, method, system, product or device comprising a series of steps or units is not necessarily limited to those steps or units clearly listed, but may include other steps or units that are not clearly listed or inherent to these processes, methods, products or devices.

[0044] For the convenience of description, some nouns or terms involved in the embodiments of the present application are explained below:

[0045] A wide-area measurement system is a system used for long-distance measurement and monitoring, and is commonly used in the fields of geographic information systems, meteorology, and environmental monitoring. The system usually consists of multiple measurement points, which can be distributed in different geographical locations, for collecting and transmitting data. A wide-area measurement system usually includes sensors, data acquisition equipment, data transmission equipment, and data processing systems. Sensors are used to collect data on various environmental parameters, such as temperature, humidity, and air pressure. The data acquisition equipment is responsible for processing and storing the data collected by the sensors, and then transmitting the data to the data processing system through the data transmission equipment for analysis and display. The wide-area measurement system can monitor and analyze the changes in various environmental parameters in real time, helping users to understand the environmental conditions in a timely manner and make corresponding adjustments. This plays an important role in protecting the environment, preventing natural disasters, and improving production efficiency.

[0046] A data acquisition and monitoring system is a system used to collect, store, analyze and monitor data. It can help users monitor data changes and trends in real time so that they can make corresponding decisions in a timely manner. A data acquisition and monitoring system usually includes functional modules such as data acquisition, data storage, data processing and data visualization. The data acquisition module is responsible for collecting data from various data sources, such as sensors, equipment, databases, etc.; the data storage module is responsible for storing the collected data in a database or data warehouse; the data processing module is responsible for cleaning, converting and analyzing the data to discover the associations and patterns between the data; the data visualization module is responsible for presenting the processed data in the form of charts, reports, etc., to help users understand the data more intuitively. Data acquisition and monitoring systems are widely used in industrial production, the Internet of Things, finance, medical care and other fields, helping users to monitor production processes, equipment status, market changes and other information in real time, and improve work efficiency and decision-making effects. At the same time, data acquisition and monitoring systems can also help users discover data anomalies, predict future trends, take countermeasures in advance, and reduce risks and losses.

[0047] RBM (Restricted Boltzmann Machine) is a probabilistic graph model, which belongs to the category of undirected graph models. It consists of a set of visible units and a set of hidden units, and can be used to learn the probability distribution of data. The characteristics of RBM are that it can be used for unsupervised learning and feature learning, and can capture complex relationships between data. In the field of deep learning, RBM is often used as a generative model or as part of a pre-trained deep neural network.

[0048] The smoke particle intelligent recognition algorithm is an algorithm that uses computer vision and image processing technology to automatically identify and classify smoke particles. By analyzing the shape, color, size and other characteristics of smoke particles, the algorithm can accurately identify different types of smoke particles and help monitor and prevent disasters such as fires. This algorithm has important application value in industrial production, environmental monitoring, fire safety and other fields.

[0049] The security monitoring EasyCVR platform is a video surveillance management platform that can help users centrally manage video recordings from multiple surveillance cameras, monitor in real time and view surveillance images remotely. It also provides video playback, event detection and alarm functions to help users manage and protect their safety more effectively.

[0050] As introduced in the background technology, the existing scheme will alarm when physical quantities such as smoke and temperature reach a specified state. When the alarm is sounded, the fire has usually occurred or the fire is large, causing harm and losses, which is not conducive to initial control; when the equipment is in a single state during pre-alarm, it is divided into alarm and fault states, and does not have real-time data monitoring functions, cannot perform data analysis, and cannot judge trend changes. Due to the lack of intermediate data, it does not have the function of hidden danger discovery and troubleshooting; when information is transmitted externally, it only has the upload function, not the download control function, and the amount of information is single, which is not conducive to unified monitoring and management, and cannot be remotely controlled; the logical relationship relied on during fire extinguishing linkage is the status information of the equipment, and the on-site situation cannot be verified twice; maintenance and management are generally recorded in traditional paper documents, and the management methods are backward, which does not meet the requirements of rapid response and the development of the big data era. The substation smart fire protection system integrated with the power Internet of Things also includes a 4-layer framework: perception layer, network layer, platform layer, and application layer, which is the specific practice of the power Internet of Things. However, it is not selected based on solutions and equipment alone, but is integrated into the fire protection of the electric power Internet of Things and is a part of the electric power Internet of Things. In order to solve the problem that the fire warning method of the substation in the existing solution can only collect a single monitoring parameter, and the abnormality of a single monitoring parameter cannot fully explain the occurrence of a fire, thereby resulting in low accuracy of the warning, the embodiments of the present application provide a fire warning method, device, medium and system for a substation.

[0051] The technical solutions in the embodiments of the present invention will be described clearly and completely below in conjunction with the accompanying drawings in the embodiments of the present invention.

[0052] In this embodiment, a fire warning method for a substation running on a mobile terminal, a computer terminal or a similar computing device is provided. It should be noted that the steps shown in the flowchart of the accompanying drawings can be executed in a computer system such as a set of computer executable instructions, and although a logical order is shown in the flowchart, in some cases, the steps shown or described can be executed in an order different from that shown here.

[0053] Figure 1 FIG. 1 is a flow chart of a fire warning method for a substation provided according to an embodiment of the present application. Figure 1 As shown, the method comprises the following steps:

[0054] Step S101, obtaining fire hazard data collected by at least three fire sensors, and normalizing the fire hazard data to obtain normalized fire hazard data, wherein the fire sensors include point smoke detectors, linear smoke detectors, temperature detectors, flame detectors, smoke sensors, microwave sensors, and temperature sensors;

[0055] The step S101 of normalizing the fire hazard data to obtain normalized fire hazard data includes:

[0056] according to Determine the above normalized fire hazard data;

[0057] Among them, f(x) is the normalized fire hazard data mentioned above, x0 is the minimum value of the range in the above fire hazard data, a is the normalization coefficient, and x is the characteristic parameter measurement value in the above fire hazard data.

[0058] Specifically, preprocessing is performed to remove noise and abnormal data, and the normalization function is used to normalize the collected data, which facilitates subsequent data processing. After normalization, the temperature variation range accounts for about 66% of the total range, the separation is greatly improved, and the response sensitivity is also improved accordingly.

[0059] Step S102, using a smoke particle intelligent recognition algorithm to identify electrical fire hazards of lines through edge computing, and fusing the above normalized fire hazard data based on multimodal data to identify fire hazards of substations and obtain multimodal fusion data;

[0060] Specifically, a fire hazard identification fusion extraction module is used, and a smoke particle intelligent identification algorithm is used to identify electrical fire hazards in the line through edge computing. The smoke particle intelligent identification algorithm used can achieve various smoke particle identification tasks with high accuracy. And it has better generalization ability than other intelligent pattern recognition algorithms such as BP network. The neural network method is used to automatically identify smoke particles and non-smoke particles with high sensitivity and extremely low false alarm rate, strong anti-interference ability, and good adaptive ability.

[0061] Based on multimodal data, multimodal data is fused, and electrical fire hazards of lines are identified through edge computing. Adaptive weighted fusion is used in combination with intelligent discrimination technology to fuse multimodal data based on multimodal data, realizing multi-parameter full-time, full-domain and full-electricity intelligent perception and precise prevention and control of electrical fires. This detection system, which adopts multi-source fire scene data fusion and adopts multi-source comprehensive detection technology, fuses multimodal data to identify fire hazards in substations, which can avoid missed reports and false alarms when a single detector is disturbed, thereby making a more accurate judgment on the fire situation. The detection system can effectively improve the accuracy and real-time performance of fire detection by using technologies such as multi-node data acquisition, pattern recognition and neural networks and substation fire detection algorithms. The detector can adopt forward and backward, dual-wavelength laser scattering technology, and greatly improve the product's ability to resist false alarms through artificial neural network algorithm processing. The present invention uses pattern recognition technology to extract features from the collected data, integrates data from at least three types of multi-sensor data, and extracts feature information related to fire; enhances the sensitivity and reliability of the alarm output of the entire system, so that the system achieves the purpose of early warning. The experimental results show that it has high accuracy and practicality, and can provide an important reference for the safety of unmanned substations.

[0062] The fire alarm intelligent generation module is adopted, and multimodal data is used as the input of the neural network pattern recognizer. The weights and thresholds of each layer are continuously adjusted through the continuous learning of the neural network and the momentum stochastic gradient descent algorithm combined with the adaptive learning rate and the additional momentum method. The momentum method reduces the sensitivity of the network to the local details of the error surface and effectively suppresses the network from falling into the local minimum; the adaptive learning rate is conducive to shortening the learning time. The combination of the adaptive learning rate and the additional momentum method has a low false alarm rate, improves the learning speed and increases the reliability of the algorithm, avoiding the defects of the BP algorithm. The momentum stochastic gradient descent algorithm is better and faster than the general stochastic gradient descent algorithm. This method helps to accelerate the gradient descent of the vector in the correct direction, so that it converges faster. It can effectively reduce the impact of noise in the training data on the model and improve the performance of the final model on the test data. A neural network that can represent the change law of each detection parameter during the fire is obtained. The self-learning, adaptive and self-organizing characteristics of the neural network make it particularly suitable for processing uncertain signals; in addition, the pattern recognizer used in pattern recognition has the characteristics of simple training and fast classification speed.

[0063] The smoke particle recognition algorithm is a visual recognition algorithm based on deep learning technology, which is mainly used to automatically detect and identify the behavior of smoke and flames in video surveillance scenes. Based on deep learning neural network technology, this technology can dynamically identify the state transition process of smoke and flames from presence to absence, from small to large, from large to small, and from light smoke to thick smoke.

[0064] Data collection and preparation: First, a large amount of video data with smoke and flames needs to be collected and labeled. The labeled data contains the location and time information of the smoke and flames, which is used to train the algorithm model.

[0065] Deep learning model construction: Use deep learning technology, usually convolutional neural network (CNN) as the basic model, to build a model for fireworks recognition. Through multiple layers of convolution and pooling layers, feature information in the video frame is extracted, and then classified through the fully connected layer to determine whether there is fireworks behavior.

[0066] Model training and optimization: Use labeled data to train the fireworks recognition model, and use the back-propagation algorithm to continuously optimize the model parameters so that it can identify fireworks more accurately.

[0067] Fireworks recognition application: Apply the trained firework recognition AI model to the video surveillance system (security monitoring EasyCVR platform) to detect in real time whether there is smoke or flame in the video stream and output corresponding alarm information.

[0068] Identify electrical fire hazards in lines through edge computing; the equipment is manufactured using SMT technology, has strong environmental adaptability and anti-electromagnetic interference performance, and can be connected to the Internet through a monitoring platform; the product has high integration, stable operation, safety and reliability, advanced algorithms, and accurate identification. Edge computing is to bring computing resources closer to the edge of the user end, IoT cloud, and data center, which can provide lower latency and better bandwidth, and reduce data transmission overhead. According to analysis, the edge computing architecture mainly consists of three parts: network edge devices, data storage and processing units, and edge computing applications.

[0069] Data normalization is an important step in data preprocessing. Many methods can be called data normalization. After the original data is processed by data normalization, the values ​​of each variable (feature) are between [0, 1].

[0070] The application of data fusion technology in fire detection and alarm has greatly increased the sensitivity of fire detection, significantly improved the recognition rate of fire detection, and significantly reduced the false alarm rate of fire, thereby achieving early alarm of fire.

[0071] In one embodiment of the present application, in the process of fusing the normalized fire hazard data based on multimodal data to identify the fire hazard of the substation and obtain multimodal fusion data, the method further includes:

[0072] When the fusion method of the above-mentioned multimodal data is data-level fusion, the data of the structured wide-area measurement system and the data acquisition and monitoring system are fused based on the above-mentioned data-level fusion, and the visible light and infrared image data of the unstructured substation power equipment are fused at the same time.

[0073] Specifically, fusion at the data level is a common feature of data-level fusion. Feature extraction and target perception are performed on the data after data-level fusion. Based on data-level fusion, structured wide area measurement system (WAMS) and data acquisition and monitoring system data are fused. At the same time, unstructured visible light and infrared image data of substation power equipment are fused to achieve high-resolution image presentation of substation power equipment. The information loss of data-level fusion is low.

[0074] In one embodiment of the present application, in the process of fusing the normalized fire hazard data based on multimodal data to identify the fire hazard of the substation and obtain multimodal fusion data, the method further includes:

[0075] When the fusion method of the above multimodal data is feature-level fusion, the hidden fault is diagnosed based on the current amplitude distortion, fault energy value and switch quantity information feature fusion, and the multimodal data is feature extracted according to the data characteristics, physical meaning and fusion target;

[0076] Convolutional neural networks are used to fuse homogeneous data and heterogeneous data with different features, and to fuse unstructured images and structured stress data to extract image data texture, dynamic features, structured current data wavelet features and information entropy features.

[0077] Specifically, feature-level fusion is the fusion of feature levels. The fire hazard identification fusion extraction module diagnoses hidden dangers and faults based on the electrical quantity characteristics such as current amplitude distortion, fault energy value, and the fusion of switch quantity information characteristics, and extracts features from multimodal data according to data characteristics, physical meanings and fusion targets; then, based on the convolutional neural network (CNN), homogeneous data and heterogeneous data with different features, unstructured images and structured stress data are fused to extract image data texture, dynamic features and structured current data wavelet and information entropy features to obtain a unified interpretation of the target object.

[0078] Convolutional neural network (CNN) maps each feature to the same space or the same coordinate system, assimilates heterogeneous feature vectors, extracts features for each parameter respectively, takes one of the feature vectors as the benchmark, and maps other feature vectors to the space where the feature vector is located; then the feature vectors of the same vector space are fused in the common subspace / sub-coordinate system, and feature fusion is performed in the common subspace / sub-coordinate system. Finally, the multi-network fusion target between the networks that solve the objective function is obtained to obtain a comprehensive perception of the target.

[0079] In one embodiment of the present application, in the process of fusing the normalized fire hazard data based on multimodal data to identify the fire hazard of the substation and obtain multimodal fusion data, the method further includes:

[0080] In the case where the fusion method of the multimodal data is decision-level fusion, the modal data of the same group or different modal data are fused with different decision results to perform multimodal data fusion;

[0081] The substations are determined as nodes, and the transmission lines connecting the substations are determined as edges. Each point directly transmits to the gateway center, and the properties of the nodes and edges are set as the network topology.

[0082] There are three main fusion methods for multimodal data fusion: front-end fusion or data level fusion, back-end fusion or decision level fusion and intermediate fusion.

[0083] This paper reviews the methods of fusing multiple modal data, such as sensor fusion, image fusion, text fusion, speech fusion, etc. Multimodal data fusion is a technology that effectively integrates and utilizes data from different modalities and maps them to the same vector space. Traditional multimodal fusion methods are generally classified into feature-level fusion, decision-level fusion, and hybrid methods that combine the first two according to the stage of modal fusion. There are many reviews related to the multimodal field, most of which focus on multimodal research in specific fields. The multimodal fusion method hopes to effectively utilize the complementary information between data, which is no different from multimodal representation learning. In addition, integrating multimodal information helps to achieve more robust predictions.

[0084] Specifically, decision-level fusion integrates decision-levels, makes decisions first with a specific fusion mechanism, synthesizes all decisions, and integrates the same set of modal data or different modal data and different decision results based on the comprehensive results obtained by sampling frequency and data structure to achieve multi-modal data fusion; then selects substations as nodes, and transmission lines connecting substations as edges, and each point directly transmits to the gateway center; analyzes its vulnerability, elasticity and stability through graph convolutional neural networks, correlation characteristic matrices, structural characteristic correlation relationships and maximum connected subsets, and uses this as the analysis object for target perception, defines the attributes of nodes and edges as network topology, and abstracts different networks and their coupling relationships into graphs or interdependent networks composed of nodes and edges based on topological structures. Among them, each point directly transmits to the gateway center, which saves transmission time, improves transmission accuracy, and can achieve fast connection with the gateway center point, with the advantages of simple network structure and low transmission latency.

[0085] Step S103, using the multimodal fusion data as the input of the neural network model, and continuously adjusting the weights and thresholds of each layer in the neural network model through the neural network technology and the stochastic gradient descent algorithm to obtain a final neural network model, wherein the output of the final neural network model includes the probability of no fire, the probability of open fire, the probability of smoldering fire, and the degree of fire hazard;

[0086] The intelligent identification algorithm of smoke particles uses the different angular distribution characteristics and polarization characteristics of the scattered light intensity of different smoke particles. The photodiodes installed at multiple angles receive multi-angle polarization signals, and the multi-angle scattered polarization signals of various particles are obtained as mode quantities through the photoelectric detector. After the sensor array receives the particle scattering signal, it generates photocurrent through photoelectric conversion, which is then amplified and sent to the computer through analog-to-digital A / D conversion, and finally the particle identification result is output through signal processing.

[0087] The detection system first pre-processes the scattering intensity at various angles of 30°, 60°, 90°, and 150° collected by the detector, and then accurately determines the particle type through fuzzy judgment. The fire hazard situation is judged according to the output result of the smoke particle intelligent recognition algorithm and the detection results are output. The signal features are extracted, and the computer fuses and calculates the multi-channel detection signals and multi-channel information parameters, and then sends them to the neural network for learning and training to determine the connection weights.

[0088] The fire alarm intelligent generation module is designed to shorten the sampling time. The probability graph model RBM is symmetrically connected to the random neural network model, and the parameters corresponding to the minimum energy function of the network are calculated. The probability distribution of the RBM is calculated by the energy function, and the probability value is obtained to optimize the RBM parameters. At the same time, the weights are randomly initialized, and the initial values ​​are optimized and pre-trained to obtain a relatively better initial value.

[0089] According to the smoke particle intelligent identification algorithm, the probability graph model RBM can convert the decision table into the connection weight matrix of the neural network to adaptively represent various input situations and give results close to the expected value. When a certain input mode is sent to the input layer, the weight W1 between the input layer and the hidden layer and the weight W2 between the hidden layer and the output layer are adjusted to minimize the output error Ep of the neural network. After adjusting the weights, the system transfers from the training state to the working state, sends the particle preprocessing signal to the input layer of the neural network, and obtains the output value of the trained weights for distinguishing the type of smoke particles.

[0090] In one embodiment of the present application, in the process of using multimodal fusion data as the input of the neural network model, and continuously adjusting the weights and thresholds of each layer in the neural network model through neural network technology and stochastic gradient descent algorithm, the method further includes:

[0091] The neural network model is symmetrically connected using a probabilistic graphical model to obtain the parameters corresponding to the minimum energy function of the neural network model, and the probability distribution of the probabilistic graphical model is calculated by the energy function to obtain the probability value to optimize the parameters of the probabilistic graphical model;

[0092] At the same time, the weights are randomly initialized, and the initial values ​​of the weights are optimized and pre-trained to optimize the initial values ​​of the weights.

[0093] Specifically, the sliding average model is used to substitute the values ​​of each detection parameter in the monitoring environment into the neural network probability graph model RBM, and the distribution of the training set is learned to obtain the probability value of the fire occurrence, and the effect of fire warning is initially achieved. Through the effective feature extraction of the RBM model, the fire temperature, CO and CO2 gas parameters in the substation are used as the data samples of the visible variables corresponding to the visible units of the RBM model, and the humidity and smoke are used as the hidden units of the fire risk model corresponding to the hidden variables, forming a binary risk variable of visible variables and hidden variables; it can improve the generalization ability of the deep learning model, and can identify the risk data in the substation fire according to the big data information, and accurately infer the data structure under the substation fire risk area. According to the established relevant areas, the intelligent substation fire terminal layout identification is implemented to determine the perception system and operation plan that actually constitutes the substation information.

[0094] Step S104, based on the optical time domain reflection and backward Raman scattering temperature effect of the optical fiber, the current real-time temperature measurement value is compared with the preset alarm parameter value to obtain a comparison result, and based on the output result of the final neural network model and the comparison result, it is determined whether to generate an early warning message.

[0095] In the above steps, at least three fire sensors are used to make subsequent judgments more consistent with actual working conditions. The normalized fire hazard data is fused based on multimodal data to identify fire hazards in substations and obtain multimodal fusion data. The multimodal fusion data is used as the input of the neural network model. The weights and thresholds of each layer in the above neural network model are continuously adjusted through neural network technology and stochastic gradient descent algorithm to obtain the final neural network model. The probability of no fire, open fire, smoldering fire and the degree of fire hazard are calculated. The multi-sensor detection parameters and their changes are used as the basis of fire judgment data. According to the optical time domain reflection and back Raman scattering temperature effect of the optical fiber, the disadvantages of the BP algorithm, such as low learning efficiency, slow convergence speed and easy to fall into the local minimum state, are avoided. Finally, according to the output result and comparison result of the final neural network model, it is determined whether to generate warning information, thereby improving the accuracy of fire warning, thereby solving the problem that the fire warning method of the substation in the existing scheme can only collect a single monitoring parameter, but the abnormality of a single monitoring parameter cannot fully explain the occurrence of a fire, resulting in low accuracy of the warning.

[0096] According to the requirements of high fire detection accuracy and fast response in power places, point smoke, linear smoke, point temperature, point flame, linear smoke, and cable temperature sensors are mainly used to assist the manual alarm function to complete the detection alarm function, and the alarm method is flexible; the manual alarm function is assisted to complete the detection alarm function; through the information configuration of temperature and smoke sensors in the substation, the integrated control of communication data information can be realized at the first time, greatly expanding the detection range of the substation safety monitoring system in time and space, and improving the timeliness and accuracy of the entire substation safety monitoring system. Apply multi-sensor data fusion technology in the fire warning system. Compared with traditional point-type smoke fire detection equipment, the detection sensitivity of the aspirating smoke fire detector is extremely high, which is 1,000 times that of the traditional point-type smoke fire detector. The selected multiple sensors monitor multiple parameters of the fire process, especially the abnormal information generated in the early stage of the fire, which makes up for the shortcomings of using a single sensor and expands the observation range in time and space. It changes the original single mode of satellite monitoring and solves the problems of small monitoring range, low timeliness and great weather influence caused by the limitations of satellite monitoring on the ground. It can accurately detect fire conditions near transmission lines, and also solves the problem that fires cannot be timely and accurately detected and precisely located due to blind spots in satellite monitoring.

[0097] Aiming at the problem that the existing fire alarm system based on wireless network transmission cannot judge and control the disaster situation due to network paralysis when a fire occurs, the fire risk assessment and early warning module is used to monitor the surrounding intelligent terminals in the training neural network, avoiding the influence of environmental changes. The hidden layer is sampled by the weight W and bias C of the data terminal. By continuously adjusting the weights and thresholds of each layer, the disadvantage of the traditional method of only setting fixed thresholds is overcome, so that the network error is minimized or the desired requirements are met. After adjusting the weights, the feedback of the data time period and the input of the information layer are set in real time to calculate the probability of no fire, open fire, smoldering fire and the degree of fire hazard. The multi-sensor detection parameters and their changes are used as the basis of fire judgment data. Based on the optical time domain reflectometry (OTDR) and backward Raman scattering temperature effect of optical fiber, the defects of BP algorithm such as low learning efficiency, slow convergence speed and easy to fall into local minimum state are avoided. Using the current real-time measurement value as the criterion, the temperature information is compared with the preset alarm parameter value, and the fire hazard is judged according to the output of the neural network. The probability judgment result is given in real time. The multi-parameter full-electricity intelligent sensing technology can be innovatively applied to break through the shackles of the traditional electrical fire monitoring system that relies on the single electrical parameters of residual current and temperature. When the alarm conditions are met, an alarm message is issued. By integrating the intelligent terminal in the substation, instructions can be issued to the substation fire extinguishing equipment terminal to quickly and automatically realize fire fighting. Ensure the effective prevention of substation equipment fires. In addition, different alarm methods such as fixed temperature, differential temperature, and differential fixed temperature are adopted; point or regional alarms can be used.

[0098] In one embodiment of the present application, determining whether to generate warning information according to the output result and the comparison result of the final neural network model includes:

[0099] Obtaining a comparison mapping relationship, wherein the comparison mapping relationship represents a mapping relationship between an output result of the final neural network model, the comparison result, and a judgment result, wherein the judgment result represents whether the warning information is generated;

[0100] According to the above-mentioned comparison mapping relationship, the output result of the above-mentioned final neural network model and the above-mentioned comparison result, it is determined whether to generate the above-mentioned warning information.

[0101] Specifically, by setting a comparison mapping relationship, the accuracy of whether to generate early warning information to warn of an impending fire is improved.

[0102] The collected information is processed and analyzed to obtain the operating status of each sensor equipment and facility. When the environmental indicators of the fire protection facilities in the monitored substation area exceed the normal range, a fault occurs or the status is abnormal, the fire point recognition algorithm is used to issue a fault warning or alarm instruction in advance, and at the same time, the fire early warning model and the video surveillance system associated with the alarm location are linked.

[0103] The important primary equipment of the substation includes: transformer, high-voltage circuit breaker, disconnector, busbar, lightning arrester, capacitor, reactor. The important areas of the substation include: main control room, relay protection room, power distribution room, station power room, battery room, communication room, cable tray, cable shaft, cable mezzanine.

[0104] It should be noted that the steps shown in the flowcharts of the accompanying drawings can be executed in a computer system such as a set of computer executable instructions, and that, although a logical order is shown in the flowcharts, in some cases, the steps shown or described can be executed in an order different from that shown here.

[0105] The embodiment of the present application also provides a fire warning device for a substation. It should be noted that the fire warning device for a substation in the embodiment of the present application can be used to execute the fire warning method for a substation provided in the embodiment of the present application. The device is used to implement the above-mentioned embodiments and preferred implementation modes, and the descriptions that have been made will not be repeated. As used below, the term "module" can implement a combination of software and / or hardware for a predetermined function. Although the device described in the following embodiments is preferably implemented in software, the implementation of hardware, or a combination of software and hardware, is also possible and conceivable.

[0106] The fire warning device for a substation provided in an embodiment of the present application is introduced below.

[0107] Figure 2 FIG. 1 is a structural block diagram of a fire warning device for a substation provided according to an embodiment of the present application. Figure 2 As shown, the device comprises:

[0108] The acquisition unit 21 is used to acquire fire hazard data collected by at least three fire sensors, and normalize the fire hazard data to obtain normalized fire hazard data. The fire sensors include point smoke detectors, linear smoke detectors, temperature detectors, flame detectors, smoke sensors, microwave sensors, and temperature sensors.

[0109] The first processing unit 22 is used to use the smoke particle intelligent recognition algorithm to identify the electrical fire hazard of the line through edge computing, and fuse the normalized fire hazard data based on multimodal data to identify the fire hazard of the substation and obtain multimodal fusion data;

[0110] The second processing unit 23 is used to use the multimodal fusion data as the input of the neural network model, and continuously adjust the weights and thresholds of each layer in the neural network model through the neural network technology and the stochastic gradient descent algorithm to obtain a final neural network model, wherein the output of the final neural network model includes the probability of no fire, the probability of open fire, the probability of smoldering fire, and the degree of fire hazard;

[0111] The third processing unit 24 is used to compare the current real-time temperature measurement value with the preset alarm parameter value based on the optical time domain reflection and backward Raman scattering temperature effect of the optical fiber to obtain a comparison result, and determine whether to generate an early warning message based on the output result of the final neural network model and the comparison result.

[0112] In the above device, at least three fire sensors are used, so that subsequent judgments can be more in line with actual working conditions. The above normalized fire hazard data is fused based on multimodal data to identify fire hazards in substations and obtain multimodal fusion data. The multimodal fusion data is used as the input of the neural network model. Through neural network technology and stochastic gradient descent algorithm, the weights and thresholds of each layer in the above neural network model are continuously adjusted to obtain the final neural network model, and the probability of no fire, open fire, smoldering fire and the degree of fire hazard are calculated. The multi-sensor detection parameters and their changes are used as the basis of fire judgment data. According to the optical time domain reflection and back Raman scattering temperature effect of optical fiber, the disadvantages of BP algorithm, such as low learning efficiency, slow convergence speed and easy to fall into local minimum state, are avoided. Finally, according to the output result and comparison result of the final neural network model, it is determined whether to generate warning information, thereby improving the accuracy of fire warning, thereby solving the problem that the fire warning method of the substation in the existing scheme can only collect a single monitoring parameter, but the abnormality of a single monitoring parameter cannot fully explain the occurrence of fire, resulting in low accuracy of warning.

[0113] In one embodiment of the present application, the acquisition unit includes a determination module.

[0114] Determine the module used to Determine the above normalized fire hazard data;

[0115] Among them, f(x) is the normalized fire hazard data mentioned above, x0 is the minimum value of the range in the above fire hazard data, a is the normalization coefficient, and x is the characteristic parameter measurement value in the above fire hazard data.

[0116] In one embodiment of the present application, the second processing unit includes a first processing module and a second processing module. In the process of using multimodal fusion data as input of the neural network model, the weights and thresholds of each layer in the above neural network model are continuously adjusted through neural network technology and stochastic gradient descent algorithm.

[0117] The first processing module is used to symmetrically connect the above-mentioned neural network model using a probability graph model, obtain the parameters corresponding to the minimum energy function of the above-mentioned neural network model, calculate the probability distribution of the above-mentioned probability graph model by the energy function, and obtain the probability value to optimize the parameters of the above-mentioned probability graph model;

[0118] The second processing module is used to randomly initialize the above weights and perform optimization pre-training on the initial values ​​of the above weights to optimize the initial values ​​of the above weights.

[0119] In one embodiment of the present application, the first processing unit includes a third processing module, in the process of fusing the normalized fire hazard data based on the multimodal data to identify the fire hazard of the substation and obtaining the multimodal fusion data,

[0120] The third processing module is used to fuse the data of the structured wide-area measurement system and the data acquisition and monitoring system based on the data-level fusion when the fusion method of the above-mentioned multimodal data is data-level fusion, and to fuse the visible light and infrared image data of the unstructured substation power equipment.

[0121] In one embodiment of the present application, the first processing unit includes a fourth processing module and a fifth processing module. In the process of fusing the normalized fire hazard data based on the multimodal data to identify the fire hazard of the substation and obtaining the multimodal fusion data,

[0122] The fourth processing module is used for diagnosing hidden faults based on the current amplitude distortion, fault energy value and switch quantity information feature fusion when the fusion mode of the above multimodal data is feature-level fusion, and extracting features from the multimodal data according to data features, physical meanings and fusion targets;

[0123] The fifth processing module is used to fuse homogeneous data and heterogeneous data with different features using a convolutional neural network, and to fuse unstructured images and structured stress data to extract image data texture, dynamic features, structured current data wavelet features, and information entropy features.

[0124] In one embodiment of the present application, the first processing unit includes a sixth processing module and a seventh processing module. In the process of fusing the normalized fire hazard data based on the multimodal data to identify the fire hazard of the substation and obtaining the multimodal fusion data,

[0125] The sixth processing module is used for fusing the same group of modal data or different modal data with different decision results to perform multimodal data fusion when the fusion mode of the multimodal data is decision-level fusion;

[0126] The seventh processing module is used to determine the substation as a node, and determine the transmission line connecting the above substation as an edge. Each point directly transmits to the gateway center, and the attributes of the node and the edge are set as the network topology structure.

[0127] In one embodiment of the present application, the third processing unit includes an eighth processing module and a ninth processing module.

[0128] The eighth processing module is used to obtain a comparison mapping relationship, wherein the comparison mapping relationship represents a mapping relationship between an output result of the final neural network model, the comparison result, and a judgment result, wherein the judgment result represents whether the warning information is generated;

[0129] The ninth processing module is used to determine whether to generate the above-mentioned warning information based on the above-mentioned comparison mapping relationship, the output result of the above-mentioned final neural network model and the above-mentioned comparison result.

[0130] The fire warning device of the above-mentioned substation includes a processor and a memory, the above-mentioned acquisition unit, and the first processing unit. The second processing unit and the third processing unit are all stored in the memory as program units, and the processor executes the above-mentioned program units stored in the memory to realize the corresponding functions. The above-mentioned modules are all located in the same processor; or, the above-mentioned modules are located in different processors in any combination.

[0131] The processor includes a kernel, which calls the corresponding program unit from the memory. One or more kernels can be set, and the kernel parameters can be adjusted to solve the problem that the fire warning method of the substation in the existing solution can only collect a single monitoring parameter, but the abnormality of a single monitoring parameter cannot fully explain the occurrence of a fire, resulting in a low accuracy of the warning.

[0132] The memory may include non-permanent memory in a computer-readable medium, random access memory (RAM) and / or non-volatile memory, such as read-only memory (ROM) or flash RAM, and the memory includes at least one memory chip.

[0133] An embodiment of the present invention provides a computer-readable storage medium, wherein the computer-readable storage medium includes a stored program, wherein when the program is executed, the device where the computer-readable storage medium is located is controlled to execute the fire warning method for the substation.

[0134] An embodiment of the present invention provides a processor, and the processor is used to run a program, wherein the fire warning method for the substation is executed when the program is run.

[0135] An embodiment of the present invention provides a device, which includes a processor, a memory, and a program stored in the memory and executable on the processor. When the processor executes the program, at least the following steps are implemented: obtaining fire hazard data collected by at least three fire sensors, and normalizing the fire hazard data to obtain normalized fire hazard data, wherein the fire sensors include point-type smoke detectors, linear smoke detectors, temperature detectors, flame detectors, smoke sensors, microwave sensors, and temperature sensors; using a smoke particle intelligent recognition algorithm to identify electrical fire hazards in a line through edge computing, and performing normalized fire hazard recognition on the normalized fire hazard data based on multimodal data; The hidden danger data is fused to identify the hidden danger of fire in the substation and obtain multimodal fusion data; the multimodal fusion data is used as the input of the neural network model, and the weights and thresholds of each layer in the above neural network model are continuously adjusted through neural network technology and stochastic gradient descent algorithm to obtain the final neural network model. The output of the above final neural network model includes the probability of no fire, the probability of open fire, the probability of smoldering fire and the degree of fire hazard; according to the optical time domain reflection and back Raman scattering temperature effect of the optical fiber, the current real-time temperature measurement value is compared with the preset alarm parameter value to obtain the comparison result, and according to the output result and comparison result of the final neural network model, it is determined whether to generate early warning information. The equipment in this article can be a server, PC, PAD, mobile phone, etc.

[0136] The present application also provides a computer program product, which, when executed on a data processing device, is suitable for executing a program that initializes at least the following method steps: obtaining fire hazard data collected by at least three fire sensors, and normalizing the fire hazard data to obtain normalized fire hazard data, wherein the fire sensors include point-type smoke detectors, linear smoke detectors, temperature detectors, flame detectors, smoke sensors, microwave sensors, and temperature sensors; using a smoke particle intelligent recognition algorithm to identify electrical fire hazards in a line through edge computing, and fusing the normalized fire hazard data based on multimodal data. The multimodal fusion data is obtained by combining the multimodal fusion data to identify the fire hazards of the substation; the multimodal fusion data is used as the input of the neural network model, and the weights and thresholds of each layer in the above neural network model are continuously adjusted through the neural network technology and the stochastic gradient descent algorithm to obtain the final neural network model. The output of the above final neural network model includes the probability of no fire, the probability of open fire, the probability of smoldering fire and the degree of fire hazard; based on the optical time domain reflection and back Raman scattering temperature effect of the optical fiber, the current real-time temperature measurement value is compared with the preset alarm parameter value to obtain a comparison result, and whether to generate early warning information is determined according to the output result and the comparison result of the final neural network model.

[0137] The present application also provides a fire warning system for a substation, the system comprising: one or more processors, a memory, and one or more programs, wherein the one or more programs are stored in the memory and are configured to be executed by the one or more processors, and the one or more programs include means for executing any one of the above methods. Using at least three fire sensors can make subsequent judgments more consistent with actual working conditions. Based on multimodal data, the above normalized fire hazard data is fused to identify the fire hazards of the substation and obtain multimodal fusion data. The multimodal fusion data is used as the input of the neural network model. Through the neural network technology and the stochastic gradient descent algorithm, the weights and thresholds of each layer in the above neural network model are continuously adjusted to obtain the final neural network model. The probability of no fire, open fire, smoldering fire and the degree of fire hazard are calculated. The multi-sensor detection parameters and their changes are used as the basis of fire judgment data. According to the optical time domain reflection and back Raman scattering temperature effect of the optical fiber, the disadvantages of the BP algorithm, such as low learning efficiency, slow convergence speed and easy to fall into the local minimum state, are avoided. Finally, according to the output result and comparison result of the final neural network model, it is determined whether to generate warning information, thereby improving the accuracy of fire warning, thereby solving the problem that the fire warning method of the substation in the existing scheme can only collect a single monitoring parameter, but the abnormality of a single monitoring parameter cannot fully explain the occurrence of a fire, resulting in low accuracy of the warning.

[0138] Obviously, those skilled in the art should understand that the above modules or steps of the present invention can be implemented by a general computing device, they can be concentrated on a single computing device, or distributed on a network composed of multiple computing devices, they can be implemented by a program code executable by a computing device, so that they can be stored in a storage device and executed by the computing device, and in some cases, the steps shown or described can be executed in a different order than here, or they can be made into individual integrated circuit modules, or multiple modules or steps therein can be made into a single integrated circuit module for implementation. Thus, the present invention is not limited to any specific combination of hardware and software.

[0139] Those skilled in the art will appreciate that the embodiments of the present application may be provided as methods, systems, or computer program products. Therefore, the present application may adopt the form of a complete hardware embodiment, a complete software embodiment, or an embodiment in combination with software and hardware. Moreover, the present application may adopt the form of a computer program product implemented in one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) that include computer-usable program code.

[0140] The present application is described with reference to the flowcharts and / or block diagrams of the methods, devices (systems), and computer program products according to the embodiments of the present application. It should be understood that each process and / or box in the flowchart and / or block diagram, as well as the combination of the processes and / or boxes in the flowchart and / or block diagram, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing device to generate a machine, so that the instructions executed by the processor of the computer or other programmable data processing device generate instructions for implementing the processes in the flowchart and / or block diagram. Figure 1 A process or multiple processes and / or boxes Figure 1 A device that provides the functions specified in a block or multiple blocks.

[0141] These computer program instructions may also be stored in a computer-readable memory capable of directing a computer or other programmable data processing device to operate in a specific manner, so that the instructions stored in the computer-readable memory produce an article of manufacture comprising an instruction device, which implements the process Figure 1 A process or multiple processes and / or boxes Figure 1 A function specified in one or more boxes.

[0142] These computer program instructions can also be loaded onto a computer or other programmable data processing device so that a series of operating steps are executed on the computer or other programmable device to produce a computer-implemented process, thereby providing instructions for implementing the process. Figure 1 A process or multiple processes and / or boxes Figure 1 The steps for the functions specified in one or more boxes.

[0143] In a typical configuration, a computing device includes one or more processors (CPU), input / output interfaces, network interfaces, and memory.

[0144] The memory may include non-permanent memory in a computer-readable medium, random access memory (RAM) and / or non-volatile memory in the form of read-only memory (ROM) or flash RAM. The memory is an example of a computer-readable medium.

[0145] Computer readable media include permanent and non-permanent, removable and non-removable media that can be implemented by any method or technology to store information. Information can be computer readable instructions, data structures, program modules or other data. Examples of computer storage media include, but are not limited to, phase change memory (PRAM), static random access memory (SRAM), dynamic random access memory (DRAM), other types of random access memory (RAM), read-only memory (ROM), electrically erasable programmable read-only memory (EEPROM), flash memory or other memory technology, compact disk read-only memory (CD-ROM), digital versatile disk (DVD) or other optical storage, magnetic cassettes, magnetic disk storage or other magnetic storage devices or any other non-transmission media that can be used to store information that can be accessed by a computing device. As defined herein, computer readable media does not include temporary computer readable media (transitory media), such as modulated data signals and carrier waves.

[0146] It should also be noted that the terms "include", "comprises" or any other variations thereof are intended to cover non-exclusive inclusion, so that a process, method, commodity or device including a series of elements includes not only those elements, but also other elements not explicitly listed, or also includes elements inherent to such process, method, commodity or device. In the absence of more restrictions, the elements defined by the sentence "comprises a ..." do not exclude the existence of other identical elements in the process, method, commodity or device including the elements.

[0147] From the above description, it can be seen that the above embodiments of the present application achieve the following technical effects:

[0148] 1) The fire warning method for the substation of the present application uses at least three fire sensors, which can make the subsequent judgment more in line with the actual working conditions. The normalized fire hazard data is fused based on multimodal data to identify the fire hazards of the substation, and the multimodal fusion data is obtained. The multimodal fusion data is used as the input of the neural network model. Through the neural network technology and the stochastic gradient descent algorithm, the weights and thresholds of each layer in the above neural network model are continuously adjusted to obtain the final neural network model. The probability of no fire, open fire, smoldering fire and the degree of fire hazard are calculated. The multi-sensor detection parameters and their changes are used as the basis of fire judgment data. According to the optical time domain reflection and back Raman scattering temperature effect of the optical fiber, the disadvantages of the BP algorithm, which usually have low learning efficiency, slow convergence speed, and easy to fall into the local minimum state, are avoided. Finally, according to the output result and comparison result of the final neural network model, it is determined whether to generate warning information, thereby improving the accuracy of the fire warning, thereby solving the problem that the fire warning method for the substation in the existing scheme can only collect a single monitoring parameter, but the abnormality of a single monitoring parameter cannot fully explain the occurrence of a fire, resulting in low accuracy of the warning.

[0149] 2) The fire warning device of the substation of the present application uses at least three fire sensors, which can make subsequent judgments more in line with actual working conditions. The above-mentioned normalized fire hazard data is fused based on multimodal data to identify the fire hazards of the substation and obtain multimodal fusion data. The multimodal fusion data is used as the input of the neural network model. Through the neural network technology and the stochastic gradient descent algorithm, the weights and thresholds of each layer in the above-mentioned neural network model are continuously adjusted to obtain the final neural network model. The probability of no fire, open flame, smoldering fire and the degree of fire hazard are calculated. The multi-sensor detection parameters and their changes are used as the basis of fire judgment data. According to the optical time domain reflection and back Raman scattering temperature effect of the optical fiber, the disadvantages of the BP algorithm, which usually have low learning efficiency, slow convergence speed, and easy to fall into the local minimum state, are avoided. Finally, according to the output result and comparison result of the final neural network model, it is determined whether to generate warning information, thereby improving the accuracy of the fire warning, thereby solving the problem that the fire warning method of the substation in the existing scheme can only collect a single monitoring parameter, but the abnormality of a single monitoring parameter cannot fully explain the occurrence of a fire, resulting in low accuracy of the warning.

[0150] The above description is only the preferred embodiment of the present application and is not intended to limit the present application. For those skilled in the art, the present application may have various modifications and variations. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present application shall be included in the protection scope of the present application.

Claims

1. A fire early warning method for a substation, characterized in that: include: Acquire fire hazard data collected by at least three fire sensors, and perform normalization processing on the fire hazard data to obtain normalized fire hazard data, wherein the fire sensors include point smoke fire detectors, linear smoke fire detectors, temperature fire detectors, flame detectors, smoke sensors, microwave sensors, and temperature sensors; Using the smoke particle intelligent recognition algorithm, the electrical fire hazards of the line are identified through edge computing, and the normalized fire hazard data is fused based on multimodal data to identify the fire hazards of the substation and obtain multimodal fusion data; Taking the multimodal fusion data as the input of the neural network model, the weights and thresholds of each layer in the neural network model are continuously adjusted through the neural network technology and the stochastic gradient descent algorithm to obtain the final neural network model, and the output of the final neural network model includes the probability of no fire, the probability of open fire, the probability of smoldering fire and the degree of fire hazard; According to the optical time-domain reflection and backward Raman scattering temperature effects of the optical fiber, the current real-time temperature measurement value is compared with the preset alarm parameter value to obtain a comparison result. According to the output result and comparison result of the final neural network model, it is determined whether to generate an early warning information.

2. The method according to claim 1, characterized in that The fire hazard data is normalized to obtain normalized fire hazard data, including: according to Determining the normalized fire hazard data; Among them, f(x) is the normalized fire hazard data, x0 is the minimum value of the range in the fire hazard data, a is the normalization coefficient, and x is the characteristic parameter measurement value in the fire hazard data.

3. The method according to claim 1, characterized in that In the process of using the multimodal fusion data as the input of the neural network model and continuously adjusting the weights and thresholds of each layer in the neural network model through the neural network technology and the stochastic gradient descent algorithm, the method further includes: The neural network model is symmetrically connected using a probabilistic graphical model to obtain the parameters corresponding to the minimum energy function of the neural network model, and the probability distribution of the probabilistic graphical model is calculated by the energy function to obtain a probability value to optimize the parameters of the probabilistic graphical model; At the same time, the weights are randomly initialized, and the initial values ​​of the weights are optimized and pre-trained to optimize the initial values ​​of the weights.

4. The method according to claim 1, characterized in that: In the process of fusing the normalized fire hazard data based on the multimodal data to identify the fire hazard of the substation and obtain the multimodal fusion data, the method further includes: When the fusion method of the multimodal data is data-level fusion, the data of the structured wide-area measurement system and the data acquisition and monitoring system are fused based on the data-level fusion, and the visible light and infrared image data of the unstructured substation power equipment are fused at the same time.

5. The method according to claim 1, characterized in that In the process of fusing the normalized fire hazard data based on the multimodal data to identify the fire hazard of the substation and obtain the multimodal fusion data, the method further includes: When the fusion mode of the multimodal data is feature-level fusion, the hidden danger fault is diagnosed based on the fusion of the current amplitude distortion, fault energy value and switch quantity information characteristics, and the multimodal data is feature extracted according to the data characteristics, physical meaning and fusion target; Convolutional neural networks are used to fuse homogeneous data and heterogeneous data with different features, and to fuse unstructured images and structured stress data to extract image data texture, dynamic features, structured current data wavelet features and information entropy features.

6. The method according to claim 1, characterized in that In the process of fusing the normalized fire hazard data based on the multimodal data to identify the fire hazard of the substation and obtain the multimodal fusion data, the method further includes: In the case where the fusion mode of the multimodal data is decision-level fusion, the modal data of the same group or different modal data are fused with different decision results to perform multimodal data fusion; The substation is determined as a node, and the transmission line connecting the substation is determined as an edge. Each point directly transmits to the gateway center, and the attributes of the node and the edge are set as the network topology structure.

7. The method according to any one of claims 1 to 6, characterized in that According to the output results and comparison results of the final neural network model, determine whether to generate warning information, including: Acquire a comparison mapping relationship, wherein the comparison mapping relationship represents a mapping relationship between an output result of the final neural network model, the comparison result, and a judgment result, wherein the judgment result represents whether to generate the warning information; Whether to generate the warning information is determined according to the comparison mapping relationship, the output result of the final neural network model and the comparison result.

8. A fire warning device for a substation, characterized in that: include: an acquisition unit, configured to acquire fire hazard data collected by at least three fire sensors, and normalize the fire hazard data to obtain normalized fire hazard data, wherein the fire sensors include point-type smoke detectors, linear smoke detectors, temperature detectors, flame detectors, smoke sensors, microwave sensors, and temperature sensors; The first processing unit is used to use a smoke particle intelligent recognition algorithm to identify electrical fire hazards of the line through edge computing, and to fuse the normalized fire hazard data based on multimodal data to identify fire hazards of the substation and obtain multimodal fusion data; A second processing unit is used to use the multimodal fusion data as an input of the neural network model, and continuously adjust the weights and thresholds of each layer in the neural network model through the neural network technology and the stochastic gradient descent algorithm to obtain a final neural network model, wherein the output of the final neural network model includes a probability of no fire, a probability of open fire, a probability of smoldering fire, and a degree of fire hazard; The third processing unit is used to compare the current real-time temperature measurement value with the preset alarm parameter value based on the optical time domain reflection and backward Raman scattering temperature effect of the optical fiber to obtain a comparison result, and determine whether to generate an early warning message based on the output result of the final neural network model and the comparison result.

9. A computer-readable storage medium, characterized in that: The computer-readable storage medium includes a stored program, wherein when the program is executed, the device where the computer-readable storage medium is located is controlled to execute the method according to any one of claims 1 to 7.

10. A fire warning system for a substation, characterized in that: include: One or more processors, a memory, and one or more programs, wherein the one or more programs are stored in the memory and are configured to be executed by the one or more processors, and the one or more programs include methods for executing any one of claims 1 to 7.

Images