Fire-fighting electrical fire early warning method and system based on Internet of Things

Abstract

The invention relates to the technical field of fire early warning, and particularly discloses a fire-fighting electrical fire early warning method and system based on the Internet of Things, and the method comprises the steps: configuring an edge computing gateway, and collecting a current waveform signal and a high-frequency electromagnetic radiation signal of an electrical loop based on the edge computing gateway; extracting harmonic characteristics and electromagnetic radiation characteristics; constructing a risk value prediction and propagation model based on the graph neural network model according to the power grid topological structure, generating a data packet when the coupling relationship between the harmonic characteristics and the electromagnetic radiation characteristics meets a joint triggering condition, inputting the data packet into the risk value prediction and propagation model, and transmitting the data packet to the power grid. A dynamic risk value and a fault source position are output based on a risk value prediction and propagation model, through bimodal acquisition, electrical fire hazards can be found in an early stage, real-time feature extraction is performed by using an edge computing gateway, the data transmission amount is reduced, modeling is performed on power grid topology through a graph neural network model, a fault propagation path can be accurately predicted, and the fault propagation efficiency is improved. And the early warning precision is improved.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of fire warning, in particular to a fire-fighting electrical fire warning method and system based on Internet of Things. BACKGROUND

[0002] Electrical fire has always been a major challenge in the field of fire safety due to its strong concealment, high suddenness, rapid spread and serious consequences. According to relevant statistics, electrical reasons have long occupied the first place in the causes of various fires. Fire accidents caused by electrical hazards such as line aging, overload, short circuit, poor contact, and electric leakage occur frequently, causing huge casualties and property losses.

[0003] Traditional electrical fire monitoring and warning technology mainly relies on independent and localized detection equipment such as temperature sensors, residual current detectors (electric leakage current detectors), and fault arc detectors. Although these technologies have played a role to some extent, they have obvious limitations:

[0004] Information silos: equipment usually operates independently or is only connected in a small range (such as a single power distribution box or building floor), making it difficult to integrate data and form a global risk view.

[0005] Early warning lag: local alarms rely on on-site personnel response, response speed is subject to personnel arrival time, and lack of remote instant notification mechanism, making it difficult to intervene in early-stage hazards in a timely manner.

[0006] Weak data analysis capability: local devices have limited processing power and mainly perform threshold judgment (such as temperature exceeding XX degrees for alarm), lack of deep correlation analysis and trend prediction capability for multi-parameter and long-time series data, and are difficult to identify complex and slowly developing hidden hazard patterns (such as gradual temperature rise caused by poor contact).

[0007] Maintenance and management difficulties: equipment is scattered, state monitoring, fault diagnosis, and regular maintenance rely on manual inspection, which is low in efficiency, high in cost, and prone to omissions.

[0008] Lack of intelligence: unable to use historical data and learning algorithms to continuously optimize the warning model, with limited room for improvement in warning accuracy and reliability. SUMMARY

[0009] The present application aims to provide a fire-fighting electrical fire warning method and system based on Internet of Things to solve the problems raised in the background.

[0010] To achieve the above-mentioned purpose, the present application provides the following technical solution: a fire-fighting electrical fire warning method based on Internet of Things, the method comprising:

[0011] An edge computing gateway is configured to collect current waveform signals and high-frequency electromagnetic radiation signals of an electric circuit based on the edge computing gateway;

[0012] Real-time spectrum analysis is performed on the current waveform to extract harmonic features, including odd harmonic content and change rate, and pulse density detection is performed on the high-frequency electromagnetic radiation signals to generate electromagnetic radiation features;

[0013] The power grid topology is obtained, and a risk value prediction and propagation model is constructed based on the power grid topology and a graph neural network model, and the risk value prediction and propagation model is deployed to the cloud;

[0014] Based on the coupling relationship between the harmonic features and the electromagnetic radiation features, it is determined whether the joint triggering condition is met;

[0015] When the joint triggering condition is met, a data packet is generated, the data packet is input to the risk value prediction and propagation model, a dynamic risk value and a fault source position are output based on the risk value prediction and propagation model, and an early warning execution scheme is generated based on the dynamic risk value.

[0016] As a further scheme of the present application, the step of obtaining the power grid topology and constructing the risk value prediction and propagation model based on the power grid topology and the graph neural network model specifically comprises:

[0017] The physical power grid structure is obtained, the power grid topology in the physical power grid structure is extracted, the electrical nodes are mapped to graph vertices, and the electrical connections are mapped to weighted edges;

[0018] The node risk state is iteratively updated based on the graph neural network model through a multi-layer message passing mechanism to generate the risk value prediction and propagation model.

[0019] As a further scheme of the present application, the step of determining whether the joint triggering condition is met based on the coupling relationship between the harmonic features and the electromagnetic radiation features specifically comprises:

[0020] The harmonic change rate sequence and the electromagnetic pulse density sequence in a preset time window are obtained;

[0021] The standard deviations of the harmonic change rate sequence and the electromagnetic pulse density sequence in the preset time window are calculated;

[0022] The coupling factor is calculated based on the standard deviations;

[0023] When the coupling factor is greater than 85% and the harmonic change rate continues to grow positively, it is determined that the joint triggering condition is met.

[0024] As a further scheme of the present application, the step of outputting the dynamic risk value and the fault source position based on the risk value prediction and propagation model specifically comprises:

[0025] Determine the basic hazard degree of each node, which is determined by the harmonic exceeding degree, the electromagnetic pulse density and the temperature rise rate;

[0026] Determine the propagation risk degree of each node, which refers to the predicted probability of propagation to the key equipment;

[0027] Determine the dynamic risk value of each node based on the basic hazard degree and the propagation risk degree and the correction factor;

[0028] Determine the fault source position based on the dynamic risk value.

[0029] As a further scheme of the present application, the step of determining the fault source position based on the dynamic risk value specifically comprises:

[0030] Calculate the partial derivative of the risk value of each node to the adjacent node;

[0031] Select the adjacent node with the maximum absolute value of the partial derivative for recursive tracing;

[0032] Determine the initial fault source when the second-order time derivative of the node risk value exceeds a preset threshold.

[0033] As a further scheme of the present application, the correction factor is determined by the weighted fusion of the environmental layer correction factor and the device layer correction factor, the device layer correction factor is determined by the device type, the aging grade and the load rate, and the environmental layer correction factor is determined by the environmental temperature, the environmental humidity, the dust concentration and the oxygen concentration.

[0034] As a further scheme of the present application, the step of generating the early warning execution scheme based on the dynamic risk value specifically comprises:

[0035] Construct a multi-level early warning execution scheme and determine the risk value range corresponding to each level of the early warning execution scheme;

[0036] Determine the corresponding early warning execution scheme based on the dynamic risk value.

[0037] The present application also provides a fire-fighting electrical fire early warning system based on the Internet of Things, which is used for realizing the fire-fighting electrical fire early warning method based on the Internet of Things.

[0038] The acquisition module is used for configuring an edge computing gateway and acquiring the current waveform signal and the high-frequency electromagnetic radiation signal of the electrical circuit based on the edge computing gateway.

[0039] The extraction module is used for performing real-time spectrum analysis on the current waveform, extracting the harmonic feature, performing pulse density detection on the high-frequency electromagnetic radiation signal, and generating the electromagnetic radiation feature.

[0040] A model construction module is configured to acquire a power grid topology, construct a risk value prediction and propagation model based on the power grid topology and a graph neural network model, and deploy the risk value prediction and propagation model to a cloud;

[0041] A condition judgment module is configured to judge whether a joint triggering condition is met based on a coupling relationship between the harmonic feature and the electromagnetic radiation feature.

[0042] A scheme output module is configured to generate a data packet when the joint triggering condition is met, input the data packet to the risk value prediction and propagation model, output a dynamic risk value and a fault source position based on the risk value prediction and propagation model, and generate an early warning execution scheme based on the dynamic risk value.

[0043] As a further scheme of the present application, the model construction module specifically includes:

[0044] A topology unit is configured to acquire a physical power grid structure, extract a power grid topology in the physical power grid structure, map an electrical node to a graph vertex, and map an electrical connection to a weighted edge.

[0045] An iteration unit is configured to iteratively update a node risk state based on the graph neural network model through a multi-layer message passing mechanism to generate the risk value prediction and propagation model.

[0046] As a further scheme of the present application, the condition judgment module specifically includes:

[0047] A sequence acquisition unit is configured to acquire a harmonic change rate sequence and an electromagnetic pulse density sequence within a preset time window.

[0048] A variance calculation unit is configured to calculate a standard deviation of the harmonic change rate sequence and the electromagnetic pulse density sequence within the preset time window.

[0049] A coupling unit is configured to calculate a coupling factor based on the standard deviation.

[0050] A triggering unit is configured to determine that the joint triggering condition is met when the coupling factor is greater than 85% and the harmonic change rate continuously increases.

[0051] Compared with the prior art, the present application has the following advantages: through dual-mode (current harmonic and electromagnetic radiation) acquisition, electrical fire hazards can be detected early; through real-time feature extraction by an edge computing gateway, data transmission volume is reduced and cloud computing load is reduced; through graph neural network model-based modeling of the power grid topology, the risk value can be predicted and the fault propagation path can be accurately predicted, so that the fault source can be located in time and accurately when a fault occurs, thereby improving the early warning accuracy. BRIEF DESCRIPTION OF DRAWINGS

[0052] In order to more clearly illustrate the technical solutions in the embodiments of the present application, the drawings needed to be used in the embodiments or prior art description will be briefly introduced as follows. Obviously, the drawings in the following description only represent some embodiments of the present application.

[0053] Figure 1 A flow chart of a fire electrical fire early warning method based on the Internet of Things is provided for the embodiments of the present application.

[0054] Figure 2 A flow chart of the step of constructing a risk value prediction and propagation model according to the power grid topology and based on a graph neural network model is provided for the embodiments of the present application.

[0055] Figure 3 A flow chart of the step of judging whether the joint triggering condition is met based on the coupling relationship between the harmonic characteristics and the electromagnetic radiation characteristics is provided for the embodiments of the present application.

[0056] Figure 4 A composition structure block diagram of a fire electrical fire early warning system based on the Internet of Things is provided for the embodiments of the present application.

[0057] Figure 5 A composition structure block diagram of a model construction module is provided for the embodiments of the present application.

[0058] Figure 6 A composition structure block diagram of a condition judgment module is provided for the embodiments of the present application. DETAILED DESCRIPTION

[0059] In order to make the technical problems to be solved by the present application, the technical solutions and the beneficial effects more clearly, the present application will be further described in detail below in combination with the drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present application and do not limit the present application.

[0060] Figure 1 A flow chart of a fire electrical fire early warning method based on the Internet of Things is provided for the embodiments of the present application. In the embodiments of the present application, a fire electrical fire early warning method based on the Internet of Things, the method comprises steps S100 to S500:

[0061] Step S100, configuring an edge computing gateway, collecting current waveform signals and high-frequency electromagnetic radiation signals of an electrical circuit based on the edge computing gateway;

[0062] Step S200, performing real-time spectrum analysis on the current waveform, extracting harmonic characteristics, the harmonic characteristics including odd harmonic content rate and change rate, performing pulse density detection on the high-frequency electromagnetic radiation signal, and generating electromagnetic radiation characteristics;

[0063] In step S300, the power grid topology is acquired, and a risk value prediction and propagation model is constructed according to the power grid topology and based on a graph neural network model, and the risk value prediction and propagation model is deployed to the cloud;

[0064] In step S400, whether the joint triggering condition is satisfied is determined based on the coupling relationship between the harmonic characteristics and the electromagnetic radiation characteristics.

[0065] In step S500, when the joint triggering condition is satisfied, a data packet is generated, the data packet is input to the risk value prediction and propagation model, a dynamic risk value and a fault source position are output based on the risk value prediction and propagation model, and an early warning execution scheme is generated based on the dynamic risk value.

[0066] In this embodiment, the edge computing gateway is an embedded device with data acquisition, calculation and communication functions, and can be deployed in an intelligent sensor node at a key node of an electrical circuit (such as a distribution box, a switch or an important load inlet), to acquire in real time: a current instantaneous value, a current effective value, a current waveform, a voltage instantaneous value, a voltage effective value, a voltage waveform, a leakage current, a cable / joint temperature, an environmental temperature / humidity, an arc characteristic (such as a high-frequency transient pulse detection), a device switch state / operation mode and the like, and to synchronously acquire a current harmonic and an EMR pulse at an edge end through the edge computing gateway.

[0067] The current waveform signal is a sequence of current instantaneous values in a conductor over time, the high-frequency electromagnetic radiation signal refers to a 30-100 MHz frequency band electromagnetic field fluctuation generated by a fault arc, a current transformer captures a conductor current in real time and outputs an analog signal, and an EMR sensor (a wideband antenna + an amplification circuit) captures a spatial electromagnetic pulse, and the two signals are synchronously sampled by an ADC to generate time-aligned digital sequences.

[0068] The power data of each device in the power grid topology is acquired in real time, a current / voltage effective value, a power and a temperature are calculated or acquired, and an effective value or a content rate of a specific harmonic is calculated in real time through an optimized sliding window FFT algorithm (such as an FFT or a wavelet transform based on a sliding window, optimized for an embedded platform), the specific harmonic refers to a 3rd (150 Hz) harmonic and a 5th (250 Hz) harmonic, and the 3rd harmonic and the 5th harmonic are most likely to cause a fire

[0069] The harmonic content rate refers to a percentage of a harmonic current effective value to a fundamental wave current effective value, the change rate characteristic refers to an increase / decrease amount of the harmonic content rate per unit time, the pulse density refers to a number of electromagnetic pulses exceeding a threshold per unit time, and the coupling relationship refers to a statistical correlation between harmonic distortion and electromagnetic pulses in time sequence.

[0070] The pulse density detection process is: first set the voltage threshold (can be 0.5V, corresponding to the field intensity 10V / m), then count the number of signals exceeding the threshold per second, and output the stable density value after sliding average filtering (window width 5 seconds).

[0071] The spectrum analysis process uses a sliding time window to intercept current signal segments, calculates the fundamental and specific harmonic component amplitudes through fast spectrum transformation, and calculates the harmonic change rate based on the amplitude changes of the harmonic components of adjacent time windows.

[0072] The initial stage of electrical fault will generate 30-100MHz transient electromagnetic pulse, and the abnormal electrical energy (harmonic / surge) generated by the fault will propagate along the conductor, thereby causing adjacent equipment to be abnormal, such as high-frequency harmonic, i.e. distortion of current waveform; voltage fluctuation, i.e. sudden rise and fall of motor, electrical equipment has a tolerance threshold, when the incoming abnormal energy exceeds the design capacity (such as transformer magnetic saturation limit), it will trigger secondary fault.

[0073] In modern smart grid, devices communicate and interact with each other, and the misoperation of a node may trigger a chain reaction through control signals, such as a certain relay misoperation leading to load transfer collapsing adjacent lines. The key case is the 2019 London blackout in the UK: starting from a gas power plant failure, it triggered the overload trip of a 66kV line, and the remaining load was transferred to adjacent lines, triggering 22 substation chain tripping, the whole process only took 9 minutes. This is a typical manifestation of fault propagation.

[0074] In the embodiment, by collecting through dual modalities (current harmonic and electromagnetic radiation), electrical fire hazards can be detected early, real-time feature extraction is performed by using an edge computing gateway, data transmission volume is reduced, and cloud computing load is reduced; by modeling the power grid topology through a graph neural network model, the risk value can be predicted, and the fault propagation path can be accurately predicted, so that the fault source can be located in time and accurately when the fault occurs, thereby improving the warning accuracy.

[0075] As shown in Figure 2 As a preferred embodiment of the present application, the step of acquiring the power grid topology structure and constructing a risk value prediction and propagation model based on a graph neural network model comprises:

[0076] Step S301: acquiring the physical power grid structure, extracting the power grid topology structure in the physical power grid structure, mapping the electrical nodes to graph vertices, and mapping the electrical connections to weighted edges;

[0077] Step S302: generating a risk value prediction and propagation model by iteratively updating the node risk state based on a graph neural network model through a multi-layer message passing mechanism.

[0078] In the embodiment, the model is constructed by mapping the power grid topology into a graph structure, and relies on a large amount of historical data, and the propagation relationship between devices is embodied from a risk tree, so that the risk degree can be directly observed through a numerical value, and the dynamic risk value is a continuous value of 0-100, and the fire occurrence probability is quantified.

[0079] The node features fuse the harmonic, electromagnetic radiation and temperature data, and the edge weight is set according to the electrical distance and historical fault propagation strength;

[0080] The training data is derived from a large amount of historical fault data.

[0081] By converting the physical power grid into a graph structure, the electrical connection relationship can be directly reflected, and a basis is provided for fault propagation modeling; through multi-layer message passing, the propagation process of the fault in the power grid is simulated, and the accuracy of the risk value prediction is improved; the weighted edge design considers the electrical distance and the fault coupling relationship, so that the model is more in line with the actual physical law.

[0082] As shown in Figure 3 As a preferred embodiment of the present application, the step of judging whether the joint triggering condition is met based on the coupling relationship between the harmonic features and the electromagnetic radiation features specifically includes:

[0083] Step S401: acquiring a harmonic change rate sequence and an electromagnetic pulse density sequence in a preset time window;

[0084] Step S402: calculating the standard deviation of the harmonic change rate sequence and the electromagnetic pulse density sequence in the preset time window;

[0085] Step S403: calculating a coupling factor based on the standard deviation;

[0086] Step S404: when the coupling factor is greater than 85% and the harmonic change rate continuously increases, it is determined that the joint triggering condition is met.

[0087] In the embodiment, the HR3 sequence and the EMR sequence in a 5-minute time window are taken, and the coupling factor calculation formula is:

[0088]

[0089] wherein N is the number of 5-minute window data points, C couple is the coupling factor, HR3 i is the 3rd harmonic content rate of the i-th sampling point, is the average value of HR3 in the window, is the pulse density of the i-th sampling point, is the average value of the pulse density in the window, and σ HR3 is the standard deviation of the harmonic change rate sequence, The standard deviation of the electromagnetic pulse density sequence.

[0090] Delta HR3 is greater than 0.3% / s, the harmonic change rate growth rate is positive, and the duration is greater than 120 seconds.

[0091] Poor contact and other failures can cause rapid rise of harmonics, and transient fluctuations caused by load switching are effectively distinguished from real failures and transient interference by statistical correlation analysis, reducing the false alarm rate (measured <5%); the condition of continuous positive growth avoids false triggering caused by transient load changes, improving system stability.

[0092] As a preferred embodiment of the present application, the step of outputting dynamic risk values and fault source positions based on the risk value prediction and propagation model specifically includes:

[0093] Determine the basic hazard degree of each node, which is determined by weighting the harmonic exceeding degree, the electromagnetic pulse density and the temperature rise rate;

[0094] Determine the propagation risk degree of each node, which refers to the predicted probability of propagation to critical equipment;

[0095] Determine the dynamic risk value of each node based on the basic hazard degree and the propagation risk degree and the correction factor;

[0096] Determine the fault source position based on the dynamic risk value.

[0097] In the present embodiment, the propagation risk degree K=1+0.2*a, a is the number of critical nodes, and the failure of critical nodes will cause the collapse of the overall system, which can quantitatively evaluate the consequences of failure chain diffusion.

[0098] The basic hazard degree comprehensively considers various electrical parameters, fully evaluates single-point risk, the propagation risk quantifies the influence of failure diffusion, realizes system-level risk evaluation, and the dynamic risk value reflects the fire probability in real time, providing a basis for hierarchical response.

[0099] As a preferred embodiment of the present application, the step of determining the fault source position based on the dynamic risk value specifically includes:

[0100] Calculate the partial derivative of the risk value of each node to the adjacent node;

[0101] Select the adjacent node with the maximum absolute value of the partial derivative for recursive tracing;

[0102] When the second-order time derivative of the node risk value exceeds the preset threshold, the initial fault source is determined.

[0103] In the embodiment, the fault source is backtracked through the dynamic risk value, so that the fault source can be directly located during fault processing, thereby improving the processing efficiency and avoiding processing only the fault table box while ignoring the root fault position.

[0104] The gradient backtracking method is used in the process of determining the fault source position, so that the fault source can be quickly located, and the second-order time derivative criterion can effectively distinguish the fault source and the node in the middle of the propagation.

[0105] As a preferred embodiment of the present application, the correction factor is determined by the weighted fusion of the environmental layer correction factor and the equipment layer correction factor, the equipment layer correction factor is determined by the equipment type, the aging level and the load rate, and the environmental layer correction factor is determined by the environmental temperature, the environmental humidity, the dust concentration and the oxygen concentration.

[0106] In the embodiment, the correction factor is set to dynamically adapt to the equipment state and environmental changes, so that the risk value is closer to the actual risk, and the weighted fusion ensures that each factor has a reasonable influence and avoids being dominated by a single factor.

[0107] The equipment layer correction factor is determined by the equipment type, the aging level and the load rate, medical equipment is more sensitive than industrial equipment in different equipment types, the insulation of old equipment deteriorates to accelerate the fire risk in different aging levels, and the fire risk of equipment with a long-term load of more than 90% is significantly increased, and the environmental layer correction factor is determined by the environmental temperature, the environmental humidity, the dust concentration and the oxygen concentration, wherein, according to the Arrhenius equation, the aging doubles when the temperature rises by 10°C, and when the relative humidity (RH) is greater than 80%, the insulation strength will decrease by 50%. This is because in a high humidity environment, a water film is easily formed on the surface of the insulation material, thereby reducing its insulation performance and increasing the risk of flashover.

[0108] According to the IEEE-1459 standard, when the dust concentration is greater than 5mg / m 3 , the arc ignition probability will increase exponentially. This means that in a high dust environment, electrical equipment is more likely to have an arc fault, thereby increasing the risk of equipment damage and fire.

[0109] When the oxygen concentration is greater than 23%, the risk of deflagration increases. High concentration of oxygen accelerates the combustion reaction, so that once a fire occurs, the fire will be more violent and more difficult to control.

[0110] The weight coefficient is determined by expert experience or historical data fitting.

[0111] As a preferred embodiment of the present application, the step of generating a pre-warning execution scheme based on a dynamic risk value specifically includes:

[0112] A multi-level pre-warning execution scheme is constructed, and the risk value range corresponding to each level of the pre-warning execution scheme is determined.

[0113] The corresponding early warning execution scheme is determined based on the dynamic risk value.

[0114] In the embodiment, the early warning response level is divided according to the dynamic risk value, when the first level early warning response, the dynamic risk value is 85 or more, at this time, the building level heat dissipation system is started and the fire fighting platform is connected, when the second level early warning response, the dynamic risk value is 65-85, at this time, the non-key load is automatically cut off and the APP push maintenance work order is carried out, when the third level early warning response, the dynamic risk value is less than 65, at this time, the data is recorded and the monitoring frequency of the adjacent nodes of the fault source is enhanced.

[0115] The early warning response level is determined by historical fault data statistics results.

[0116] At the same time, the model parameters are optimized and updated based on the actual false negative and false positive conditions, the over-treatment is avoided through the hierarchical response, the normal power supply is ensured, and the overall safety is improved through the multi-system cooperation (power distribution, heating, and fire fighting).

[0117] As shown in Figure 4 The application also provides a fire-fighting electrical fire early warning system based on the Internet of Things, which is used to realize the fire-fighting electrical fire early warning method based on the Internet of Things, and the system comprises:

[0118] The acquisition module 100 is configured as an edge computing gateway, and is used to acquire the current waveform signal and the high-frequency electromagnetic radiation signal of the electrical circuit based on the edge computing gateway.

[0119] The extraction module 200 is used to perform real-time spectrum analysis on the current waveform, extract the harmonic feature, and perform pulse density detection on the high-frequency electromagnetic radiation signal to generate the electromagnetic radiation feature.

[0120] The model construction module 300 is used to acquire the power grid topology, construct a risk value prediction and propagation model based on the graph neural network model according to the power grid topology, and deploy the risk value prediction and propagation model to the cloud.

[0121] The condition judgment module 400 is used to judge whether the joint triggering condition is met based on the coupling relationship between the harmonic feature and the electromagnetic radiation feature.

[0122] The scheme output module 500 is used to generate a data packet when the joint triggering condition is met, input the data packet into the risk value prediction and propagation model, output the dynamic risk value and the fault source position based on the risk value prediction and propagation model, and generate the early warning execution scheme based on the dynamic risk value.

[0123] Through dual-mode (current harmonic and electromagnetic radiation) acquisition, electrical fire hazards can be found early, real-time feature extraction is performed by using an edge computing gateway, data transmission amount is reduced, and cloud computing load is reduced; through modeling of the power grid topology by using a graph neural network model, the risk value can be predicted, and the fault propagation path can be accurately predicted, the fault source can be located in time and accurately according to the fault propagation path when the fault occurs, and therefore the early warning accuracy can be improved.

[0124] As shown in Figure 5 , as a preferred embodiment of the present application, the model construction module specifically comprises:

[0125] The topology unit 301 is configured to obtain a physical power grid structure, extract a power grid topology structure in the physical power grid structure, map an electrical node to a graph vertex, and map an electrical connection to a weighted edge.

[0126] The iteration unit 302 is configured to iteratively update a node risk state based on a graph neural network model through a multi-layer message passing mechanism, and generate a risk value prediction and propagation model.

[0127] In this embodiment, through the deep learning model for mapping the power grid topology to the graph structure, the construction of the model depends on a large amount of historical data, the propagation relationship between various devices is embodied from the risk resin, the risk degree can be directly observed through the numerical value, the dynamic risk value is a continuous value of 0-100, and the fire occurrence probability is quantified.

[0128] The node feature fusion harmonic, electromagnetic radiation and temperature data, and the edge weight is set according to the electrical distance and historical fault propagation strength, and the training data is derived from a large amount of historical fault data.

[0129] Through the conversion of the physical power grid into a graph structure, the electrical connection relationship can be directly reflected, and a basis is provided for fault propagation modeling; through multi-layer message passing, the propagation process of the fault in the power grid is simulated, and the accuracy of the risk value prediction is improved; the design of the weighted edge considers the electrical distance and the fault coupling relationship, so that the model is more in line with the actual physical law.

[0130] As shown in Figure 6 , as a preferred embodiment of the present application, the condition judgment module specifically comprises:

[0131] The sequence acquisition unit 401 is configured to acquire a harmonic change rate sequence and an electromagnetic pulse density sequence in a preset time window.

[0132] The variance calculation unit 402 is configured to calculate the standard deviation of the harmonic change rate sequence and the electromagnetic pulse density sequence in the preset time window.

[0133] The coupling unit 403 is configured to calculate a coupling factor based on the standard deviation.

[0134] The triggering unit 404 is configured to determine that the joint triggering condition is met when the coupling factor is greater than 85% and the harmonic change rate continues to increase.

[0135] The above merely provides the preferred embodiments of the present application, and is not used to limit the present application, and any modification, equivalent replacement, and improvement made within the spirit and principle of the present application should be included in the protection scope of the present application.

Claims

1. A fire electrical fire early warning method based on Internet of Things, characterized in that, The method comprises: configuring an edge computing gateway, collecting current waveform signals and high-frequency electromagnetic radiation signals of an electric circuit based on the edge computing gateway; performing real-time spectral analysis on the current waveform, extracting harmonic features, the harmonic features including odd harmonic content rate and change rate, performing pulse density detection on the high-frequency electromagnetic radiation signals, and generating electromagnetic radiation features; obtaining a power grid topology, constructing a risk value prediction and propagation model based on the power grid topology and a graph neural network model, and deploying the risk value prediction and propagation model to the cloud; judging whether a joint triggering condition is met based on a coupling relationship between the harmonic features and the electromagnetic radiation features; when the joint triggering condition is met, generating a data packet, inputting the data packet into the risk value prediction and propagation model, outputting a dynamic risk value and a fault source position based on the risk value prediction and propagation model, and generating an early warning execution scheme based on the dynamic risk value.

2. The fire electrical fire early warning method based on the Internet of Things according to claim 1, characterized in that, The step of obtaining the power grid topology and constructing the risk value prediction and propagation model based on the power grid topology and the graph neural network model comprises: obtaining a physical power grid structure, extracting a power grid topology in the physical power grid structure, mapping electrical nodes to graph vertices, and mapping electrical connections to weighted edges; iteratively updating node risk states based on the graph neural network model through a multi-layer message passing mechanism to generate the risk value prediction and propagation model.

3. The fire electrical fire early warning method based on the Internet of Things according to claim 1, characterized in that, The step of judging whether the joint triggering condition is met based on the coupling relationship between the harmonic features and the electromagnetic radiation features comprises: obtaining a harmonic change rate sequence and an electromagnetic pulse density sequence within a preset time window; calculating standard deviations of the harmonic change rate sequence and the electromagnetic pulse density sequence within the preset time window; calculating a coupling factor based on the standard deviations; when the coupling factor is greater than 85% and the harmonic change rate continuously increases, it is determined that the joint triggering condition is met.

4. The fire electrical fire early warning method based on the Internet of Things according to claim 1, characterized in that, The step of outputting the dynamic risk value and the fault source position based on the risk value prediction and propagation model comprises: determining a basic hazard degree of each node, the basic hazard degree being determined by weighting a harmonic over-standard degree, an electromagnetic pulse density, and a temperature rise rate; determining a propagation risk degree of each node, the propagation risk degree referring to a predicted probability of being propagated to a critical device; determining a dynamic risk value of each node based on the basic hazard degree, the propagation risk degree, and a correction factor; determining a fault source position based on the dynamic risk value.

5. The fire electrical fire early warning method based on the Internet of Things according to claim 4, characterized in that, The step of determining the fault source position based on the dynamic risk value comprises: calculating partial derivatives of risk values of each node with respect to adjacent nodes; selecting an adjacent node with the largest absolute value of the partial derivative to perform recursive tracing; when a second-order time derivative of the node risk value exceeds a preset threshold, determining the adjacent node as an initial fault source.

6. The fire electrical fire early warning method based on Internet of Things according to claim 4, characterized in that, The correction factor is determined by weighting and fusing an environmental layer correction factor and a device layer correction factor, the device layer correction factor being determined by a device type, an aging level, and a load rate, and the environmental layer correction factor being determined by an environmental temperature, an environmental humidity, a dust concentration, and an oxygen concentration.

7. The fire electrical fire early warning method based on Internet of Things according to claim 1, characterized in that, The step of generating the early warning execution scheme based on the dynamic risk value comprises: constructing a multi-level early warning execution scheme and determining a risk value range corresponding to each level of the early warning execution scheme; determining a corresponding early warning execution scheme based on the dynamic risk value.

8. An Internet of Things based fire-fighting electrical fire warning system for implementing the Internet of Things based fire-fighting electrical fire warning method of any one of claims 1-7, characterized in that, The system comprises: The collection module is configured to configure an edge computing gateway, and collect current waveform signals and high-frequency electromagnetic radiation signals of an electric circuit based on the edge computing gateway; The extraction module is configured to perform real-time spectrum analysis on the current waveform, extract harmonic features, and perform pulse density detection on the high-frequency electromagnetic radiation signals to generate electromagnetic radiation features. The model construction module is configured to obtain a power grid topology, construct a risk value prediction and propagation model based on the power grid topology and a graph neural network model, and deploy the risk value prediction and propagation model to the cloud. The condition judgment module is configured to determine whether a joint triggering condition is met based on a coupling relationship between the harmonic features and the electromagnetic radiation features. The scheme output module is configured to generate a data packet when the joint triggering condition is met, input the data packet into the risk value prediction and propagation model, output a dynamic risk value and a fault source position based on the risk value prediction and propagation model, and generate an early warning execution scheme based on the dynamic risk value.

9. The fire electrical fire warning system based on Internet of Things according to claim 8, characterized in that, The model construction module specifically includes: A topology unit is configured to obtain a physical power grid structure, extract a power grid topology in the physical power grid structure, map electrical nodes to graph vertices, and map electrical connections to weighted edges. An iteration unit is configured to update node risk states through a multi-layer message passing mechanism based on a graph neural network model to generate a risk value prediction and propagation model.

10. The fire electrical fire warning system based on Internet of Things according to claim 8, characterized in that, The condition judgment module specifically includes: A sequence acquisition unit is configured to obtain a harmonic change rate sequence and an electromagnetic pulse density sequence within a preset time window. A variance calculation unit is configured to calculate standard deviations of the harmonic change rate sequence and the electromagnetic pulse density sequence within the preset time window. A coupling unit is configured to calculate a coupling factor based on the standard deviations. A triggering unit is configured to determine that the joint triggering condition is met when the coupling factor is greater than 85% and the harmonic change rate is continuously positively increasing.

Images