Electrical fire intelligent identification system based on multi-dimensional sensor fusion

Abstract

The invention discloses an electrical fire intelligent identification system based on multi-dimensional sensor fusion. The system comprises the following steps: constructing a reference environment model through multi-sensor scanning, data dimension reduction and intelligent node deployment; multi-sensor time sequence alignment is carried out through edge calculation denoising and dynamic time warping, and high-priority data is processed in real time through a layering mechanism; establishing a fire feature modeling system through LSTM time sequence analysis, mutual information correlation mining and self-supervised learning; through multi-level data fusion, GAN abnormal data generation and fuzzy logic reasoning; through grading alarm, an intelligent fire extinguishing strategy and remote control, full-process coverage from fire detection to emergency response is realized. A fire scene is visualized by means of a three-dimensional thermodynamic diagram, flame dynamic analysis and an augmented reality technology. The system is suitable for fire detection, alarm and response in a complex industrial scene, and can be widely applied to intelligent management of electrical fire.

Description

Technical Field

[0001] The present invention relates to the technical field of electrical fire monitoring, and particularly to an intelligent electrical fire recognition system based on multi-dimensional sensor fusion. Background Art

[0002] With the wide application of electrical equipment, electrical fires have become a type of fire with high frequency and strong harmfulness. Most traditional fire detection systems rely on a single sensor for monitoring. Affected by the environmental complexity and noise interference, false alarms and missed alarms are likely to occur, and it is difficult to balance the real-time performance and accuracy of fire detection.

[0003] The existing technologies lack effective algorithm support in dealing with dynamic and multi-dimensional data, resulting in the inability to timely identify potential risks in the early stage of a fire. The existing fire alarm and response systems usually rely on fixed thresholds and are difficult to adapt to the changing environmental characteristics, which limits the extensiveness of their application scenarios. The algorithm complexity of multi-sensor data fusion is relatively high and the real-time performance is not strong; some systems fail to effectively combine noise processing and feature extraction in a dynamic environment, resulting in limited fire feature modeling capabilities; most of the existing response systems are passive in design and lack intelligent self-learning and dynamic optimization functions, making it difficult to meet the fire handling requirements in complex scenarios.

[0004] The present invention provides an intelligent electrical fire recognition system based on multi-dimensional sensor fusion. By integrating edge computing, deep learning models, adaptive filtering, and dynamic time series analysis, it realizes the efficient extraction and fusion of multi-dimensional environmental features; the system can dynamically evaluate the fire risk level according to real-time data and optimize the fire extinguishing strategy by combining reinforcement learning, so as to achieve early warning, precise positioning, and efficient response to fires. Summary of the Invention

[0005] Aiming at the above problems, the present invention provides an intelligent electrical fire recognition system based on multi-dimensional sensor fusion to solve the problems of high false alarm rate, high missed alarm rate, slow response speed, and insufficient adaptability of traditional fire detection systems.

[0006] To solve the above technical problems, the present invention provides the following technical solutions:

[0007] The present invention provides an intelligent electrical fire recognition system based on multi-dimensional sensor fusion, which is characterized by including the following steps:

[0008] Step S1, constructing a benchmark environment model through multi-sensor scanning, data dimensionality reduction, and intelligent node deployment;

[0009] Among them, in step S1, the following sub-steps are further included:

[0010] S1-1. Use temperature sensors, smoke sensors, gas sensors, and pyrolysis particle sensors to scan the initial environmental data through a distributed node layout. Utilize the Bayesian optimization algorithm to dynamically adjust the sensor response error, as shown in Equation (1):

[0011] P(θ|D) ∝ P(D|θ)P(θ) Equation (1)

[0012] Among them, P(θ|D) is the posterior distribution of the sensor parameters, P(D|θ) is the current scan result, and P(θ) is the historical data;

[0013] S1-2. Collect the initial data matrix X of the sensors, including temperature, smoke concentration, gas concentration, and humidity data. Calculate the covariance matrix, extract eigenvalues and eigenvectors, and select the main components for dimensionality reduction to form a benchmark model, as shown in Equation (2):

[0014]

[0015] Among them, is the sample normalization factor, n is the number of samples, X is the initial data matrix, and X T is the transpose of the initial data matrix;

[0016] S1-3. Deploy portable intelligent nodes, arrange multiple nodes in high-fire-risk areas to form a monitoring coverage network, and arrange coefficient nodes in areas with lower risks. The portable nodes are equipped with magnetic bases and movable brackets.

[0017] Step S2. Perform denoising through edge computing, align the multi-sensor time series by dynamic time warping, and use a hierarchical mechanism to process high-priority data in real time;

[0018] Among them, in Step S2, the following sub-steps are also included:

[0019] S2-1. Introduce edge computing and use the Kalman filtering algorithm for data preprocessing, which is completed locally on the sensor to remove high-frequency random noise, as shown in Equation (3):

[0020] x k = x k-1 + K k (z k - Hx k-1 ) Equation (3)

[0021] Among them, x k is the state estimate value at time k, x l-1 is the state estimate value at the previous time k - 1, z l is the observation value at the current time k, K l is the Kalman gain, and H is the observation model matrix;

[0022] S2-2. Use dynamic time warping (DTW) to perform time alignment on time series data from multiple sensors, capture cross-dimensional correlation trends, and identify abnormal patterns in sensor data using cluster analysis, as shown in Equations (4)-(5);

[0023] DTW(i,j) = d(x i ,y j ) + min{DTW(i - 1,j), DTW(i,j - 1), DTW(i - 1,j - 1)} Equation (4)

[0024]

[0025] where DTW(i,j) is the cumulative distance of the optimal alignment path to point (i,j), d(x i ,y j ) is the Euclidean distance between the i-th point and the j-th point, DTW(i - 1,j) matches from the upper point to the current point, DTW(i,j - 1) matches from the left point to the current point, and DTW(i - 1,j - 1) matches from the upper left diagonal point to the current point; J is the objective function of clustering, K is the number of clusters, C k is the set of the k-th cluster, x i is the i-th data point in the dataset, and μ k is the centroid of the k-th cluster;

[0026] S2-3. Add a dynamic data stratification mechanism, divide it into high-priority and medium-low-priority based on the change rate and risk level of sensor data, use a priority queue, high-priority data triggers alarms and further analysis in real time, and medium-low-priority data is processed later.

[0027] Step S3. Establish a fire feature modeling system through LSTM time series analysis, mutual information correlation mining, and self-supervised learning;

[0028] Among them, in step S3, the following sub-steps are also included:

[0029] S3-1. Use the LSTM time series feature analysis algorithm to capture the early fire trend and model long-term dependence relationships through the forget gate, input gate, and output gate mechanisms, as shown in Equations (6)-(10):

[0030] f t = σ(W f ·[h t-1 ,x t ) + b f ) (6)

[0031] i t = σ(W i ·[ht-1 , x t + b i ) (7)

[0032] C t = f t ·C t-1 + i t ·tanh(W C ·[h t-x , x t + b C ) (8)

[0033] o t = σ(W o ·[h t-1 , x t + b o ) (9)

[0034] h t = o t ·tanh(C t ) (10)

[0035] Among them, f t is the forgetting gate, i t is the input gate, C t is the cell state update, o t is the output gate, h t is the hidden state update;

[0036] S3-2. Use the mutual information analysis algorithm to analyze the multi-dimensional feature correlation, locate the fire risk signal, and combine historical data to construct a fire risk model based on feature correlation, focusing on capturing the temperature rise caused by short circuits and the abnormal gas concentration generated by arcs. Specifically, as shown in Equation (11):

[0037]

[0038] Among them, I(X; Y) is the mutual information between variables X and Y, X and Y are random variables, P(x, y) is the joint probability distribution of X = x and Y = y, P(x) is the marginal probability distribution of X = x, and P(y) is the marginal probability distribution of Y = y.

[0039] Step S4, through multi-level data fusion, GAN to generate abnormal data, and fuzzy logic reasoning;

[0040] Among them, in step S4, the following sub-steps are also included:

[0041] S4-1. Use the weighted average method to fuse multi-sensor data, and the weights are dynamically adjusted according to the noise variances of the sensors and the real-time environment. Specifically, as shown in Equations (12) - (13):

[0042]

[0043] Among them, w i is the weight, and x o is the sensor data; is the noise variance of sensor i;

[0044] S4-2. Use the generative adversarial network (GAN) to generate simulated abnormal data. The generator G(z) inputs the random noise z and outputs the simulated abnormal data; the discriminator D(x) is used to distinguish between real data and generated data. The objective function of GAN is specifically as shown in Equation (14):

[0045]

[0046] Among them, G(z) is the generator, D(x) is the discriminator, z is the random noise vector, and x is the real data sample, is the average value of the real data distribution on pdata(x), is the average value of the real data distribution on p x (z), pdata(x) is the probability distribution of the real data, and p x (z) is the probability distribution of the random noise;

[0047] S4-3. In a complex environment, use fuzzy logic decision-making to comprehensively analyze multi-dimensional features and judge the fire risk level. Fuzzy logic consists of three steps: fuzzy rules, fuzzification function, and defuzzification, specifically as shown in Equations (15)-(17):

[0048] IF AND THEN Equation (15)

[0049]

[0050] Among them, μ(x) is the membership value of the input x in the interval [a, b]; y is the output after defuzzification, and μ i is the membership degree, y i is the scalar value corresponding to the risk level, and n is the total number of fuzzy rules;

[0051] Fuzzy logic decision-making maps the input of multi-dimensional features to the risk level by setting a rule set; according to the range of feature changes, the fire risk is divided into four levels: low risk, medium risk, high risk, and emergency risk.

[0052] Step S5. Achieve full-process coverage from fire detection to emergency response through hierarchical alarm, intelligent fire extinguishing strategy, and remote control;

[0053] Among them, in step S5, the following sub-steps are also included:

[0054] S5-1. Divide the alarm level according to the fire rating and real-time environmental data, add a progressive alarm mechanism, which is divided into "Warning", "Danger", and "Emergency", and determine the response strategy according to the risk level:

[0055] Warning level: When the fire risk level is medium risk and the change rate of real-time data is low, the system triggers an alarm at the "Warning" level; at this time, the user is reminded to pay attention to potential risks, but no emergency response is required.

[0056] Danger level: When the fire risk level is high risk, regardless of the change rate of real-time data, the system triggers an alarm at the "Danger" level, or when the fire risk level is medium risk but the change rate of real-time data is high, the system also triggers an alarm at the "Danger" level; at this time, it is recommended that the user immediately check for abnormalities and take preventive measures.

[0057] Emergency level: When the fire risk level is emergency risk, the system immediately triggers an alarm at the "Emergency" level, or when the fire risk level is high risk and the change rate of real-time data increases significantly, the system also upgrades the alarm to the "Emergency" level and activates the fire extinguishing device and emergency linkage response.

[0058] S5-2. Judge the fire type according to the sensor characteristics, formulate the mapping rules between the fire type and the fire extinguishing equipment, and use the reinforcement learning algorithm to optimize the action strategy of the fire extinguishing equipment, specifically as shown in Equation (18):

[0059]

[0060] Among them, Q(s,a) is the value of executing action a in the current state s, α is the learning rate, and r is the reward value; the mapping rules are that dry powder fire extinguishers are suitable for arc fires, and gas fire extinguishers are suitable for fires in equipment areas.

[0061] S5-3. Deploy an Internet of Things-based remote control system, use the Internet of Things platform to send alarm information to users in real time, including fire registration, location, and sensor data, and allow staff to view the alarm information through mobile devices and manually or automatically control the fire extinguishing equipment.

[0062] Step S6. Visualize the fire scene with the help of 3D heat maps, flame dynamic analysis, and augmented reality technology.

[0063] Among them, in step S6, the following sub-steps are also included:

[0064] S6-1. Combine infrared camera and sensor data to generate a 3D heat map, provide the three-dimensional position of the fire source and the heat distribution, and calculate the fire source position using the triangulation method based on multi-point ranging, specifically as shown in Equation (19):

[0065] (x,y,z) = f(d1,d2,d3) Equation (19)

[0066] Among them, d1, d2, and d3 are the distances between the infrared camera and the fire source, and the three-dimensional coordinates of the fire source are calculated through the angles between nodes.

[0067] S6-2: Adopt edge AI technology to analyze the dynamic characteristics and abnormal behaviors of the flame in real time. The dynamic characteristics include the change rate of the flame area, and the abnormal behaviors include equipment explosion and smoke filling. Use a convolutional neural network to extract the flame texture and dynamic characteristics, determine the nature of the fire source, and provide the flame category and dynamic information.

[0068] S6-3: Use augmented reality technology to overlay the fire source location map and the on-site equipment layout, and display the fire source location in real time through an AR device to achieve the visualization of the fire source location; provide an evacuation route in combination with the layout of the fire scene; overlay the location of the fire extinguishing equipment to guide the rescue personnel to perform operations and provide fire extinguishing assistance.

[0069] Compared with the prior art, the beneficial effects of the present invention are as follows:

[0070] The present invention uses a variety of sensors to collect multi-dimensional environmental data, and realizes the deep fusion of sensor data through dynamic time warping and mutual information analysis, which can capture abnormal features in complex environments and significantly improve the accuracy of fire recognition.

[0071] The present invention uses edge computing to perform data preprocessing locally at the sensor nodes, removes noise through Kalman filtering and dynamic hierarchical mechanisms, improves the data quality, and ensures the reliability and real-time performance of the system in the case of high noise or data loss.

[0072] The present invention introduces an LSTM network to extract long-term dependence features from time series data, and combines a fuzzy logic decision system to dynamically evaluate the fire risk level, realizing the early recognition and hierarchical alarm of fires.

[0073] The present invention uses reinforcement learning to dynamically optimize the fire extinguishing strategy, automatically selects the optimal fire extinguishing equipment such as dry powder and gas according to the fire type, and at the same time has a hierarchical alarm function to ensure efficient response and optimal resource allocation in different fire scenarios. BRIEF DESCRIPTION OF THE DRAWINGS

[0074] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the following will briefly introduce the drawings required to be used in the embodiments. It should be understood that the following drawings only show some embodiments of the present invention, and therefore should not be regarded as limiting the scope. For those of ordinary skill in the art, without creative efforts, other related drawings can also be obtained based on these drawings.

[0075] Figure 1 It is the system flowchart of the present invention. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0076] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the following will clearly and completely describe the technical solutions in the embodiments of the present invention with reference to the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are some, but not all, of the embodiments of the present invention. All other embodiments obtained by those of ordinary skill in the art based on the embodiments of the present invention without creative efforts fall within the scope of protection of the present invention. Therefore, the following detailed description of the embodiments of the present invention provided in the drawings is not intended to limit the scope of the claimed present invention, but is merely for the selected embodiments of the present invention. All other embodiments obtained by those of ordinary skill in the art based on the embodiments of the present invention without creative efforts fall within the scope of protection of the present invention.

[0077] Please refer to Figure 1 which is a flowchart of an intelligent electrical fire identification system based on multi-dimensional sensor fusion provided by an embodiment of the present invention, including the following steps:

[0078] Step S1, constructing a reference environment model through multi-sensor scanning, data dimensionality reduction, and intelligent node deployment;

[0079] Among them, in step S1, the following sub-steps are further included:

[0080] S1-1, using temperature sensors, smoke sensors, gas sensors, and pyrolysis particle sensors, scanning the initial environmental data through a distributed node layout, and dynamically adjusting the sensor response error using the Bayesian optimization algorithm, specifically as shown in Equation (1):

[0081] P(θ|D)∝P(D|θ)P(θ) Equation (1)

[0082] Among them, P(θ|D) is the posterior distribution of sensor parameters, P(D|θ) is the current scanning result, and P(θ) is historical data;

[0083] S1-2, collecting the initial data matrix X of the sensors, including temperature, smoke concentration, gas concentration, and humidity data, calculating the covariance matrix, extracting eigenvalues and eigenvectors, and selecting the main components for dimensionality reduction to form a reference model, specifically as shown in Equation (2):

[0084]

[0085] Among them, is the sample normalization factor, n is the number of samples, X is the initial data matrix, and X T is the transpose of the initial data matrix;

[0086] S1-3. Deploy portable intelligent nodes, arrange multiple nodes in high-fire-risk areas to form a monitoring coverage network, and arrange coefficient nodes in areas with lower risks. The portable nodes are equipped with magnetic bases and movable brackets.

[0087] It should be noted that the temperature sensor is used to monitor the temperature change of the environment or the surface of the device in real time. In the initial stage of a fire, the abnormal increase in temperature is often an important feature. The temperature sensor can detect the trend change of temperature and the abnormal heating rate, providing data support for early fire warning.

[0088] The smoke sensor is used to detect the concentration of smoke particles in the air. The generation of smoke during fire combustion is an obvious feature. The smoke sensor can identify the change in concentration. For example, the light scattering method or the ionization method can be used to detect smoke particles.

[0089] The gas sensor is used to monitor specific gases, including the concentration of carbon monoxide and carbon dioxide. Electrical fires may release harmful gases. For example, carbon monoxide is released when insulation materials decompose due to short circuits or overloads. The gas sensor provides key combustion chemical characteristics.

[0090] The pyrolysis particle sensor is used to detect pyrolysis particles released due to material heating or combustion. The pyrolysis particle sensor is very effective for detecting the decomposition of plastics and insulation materials in electrical fires and can capture signals in the non-flaming stage at the initial stage of a fire.

[0091] The distributed nodes are arranged in the monitoring area. Through the arrangement of multiple sensor nodes, omni-directional data collection and fire risk monitoring within the area are realized. According to the risk characteristics and monitoring requirements of different areas, the sensor nodes are flexibly arranged to optimize the monitoring accuracy and coverage. In high-risk areas, more nodes are arranged to form a high-density monitoring network; in low-risk areas, the node arrangement is relatively sparse, only providing basic monitoring.

[0092] Step S2. Perform denoising through edge computing, align the multi-sensor time series by dynamic time warping, and use a hierarchical mechanism to process high-priority data in real time;

[0093] Among them, in step S2, the following sub-steps are also included:

[0094] S2-1. Introduce edge computing and use the Kalman filtering algorithm for data preprocessing, which is completed locally on the sensor to remove high-frequency random noise. Specifically, as shown in Equation (3):

[0095] x k = x k-1 + K k (z k - Hx k-1 ) Equation (3)

[0096] Among them, x kis the state estimate at time k, x k-1 is the state estimate at the previous time k-1, z k is the observation at the current time k, K k is the Kalman gain, and H is the observation model matrix;

[0097] S2-2. Use dynamic time warping (DTW) to perform time alignment on time series data from multiple sensors, capture cross-dimensional correlation trends, and use cluster analysis to identify abnormal patterns in sensor data, as shown in Equations (4)-(5);

[0098] DTW(i,j) = d(x i ,y j )

[0099] + min{DTW(i-1,j), DTW(i,j-1), DTW(i-1,j

[0100] -1)} Equation (4)

[0101]

[0102] where DTW(i,j) is the cumulative distance of the optimal alignment path to point (i,j), d(x i ,y j ) is the Euclidean distance between the i-th point and the j-th point, DTW(i-1,j) matches from the upper point to the current point, DTW(i,j-1) matches from the left point to the current point, and DTW(i-1,j-1) matches from the upper left diagonal point to the current point; J is the objective function of clustering, K is the number of clusters, C k is the set of the k-th cluster, x i is the i-th data point in the dataset, μ k is the centroid of the k-th cluster;

[0103] S2-3. Add a dynamic data stratification mechanism, divide it into high-priority and medium-low-priority according to the change rate and risk level of sensor data, use a priority queue, high-priority data triggers alarms and further analysis in real time, and medium-low-priority data is processed later.

[0104] It should be noted that edge computing is a data processing technology that deploys computing resources close to the data source to directly complete some data processing tasks locally, rather than transmitting all data to the central server for analysis; using edge computing to complete data preprocessing locally at the sensor, and using Kalman filtering to remove high-frequency noise to generate smooth data input, avoiding the time delay required for data transmission to the central server, thereby significantly improving the real-time response ability of the system. Even if the network connection is interrupted, the edge node can still independently complete local analysis and synchronize data after the connection is restored.

[0105] The temperature sensor provides ambient temperature data, the smoke sensor detects the concentration of smoke particles, and the gas sensor detects the gases released during the combustion process. There is a co-variation relationship among these data; by using dynamic time warping to align the time series data of different sensors, capturing the time consistency of cross-dimensional features, using DTW to calculate the minimum path distance between the temperature and smoke concentration time series, determining whether there is co-variation, and extracting potential fire signals; since the sensor data of electrical fires exhibits non-linear and short-term abnormal characteristics, the DTW algorithm combines historical environmental data to dynamically capture the time consistency of these abnormal signals.

[0106] Step S3, establish a fire feature modeling system through LSTM time series analysis, mutual information correlation mining, and self-supervised learning;

[0107] Among them, in step S3, the following sub-steps are also included:

[0108] S3-1, use the LSTM time series feature analysis algorithm to capture the early trend of fire, and model long-term dependencies through the forget gate, input gate, and output gate mechanisms, specifically as shown in equations (6)-(10):

[0109] f t =σ(W f ·[h t-1 ,x t +b f ) (6)

[0110] i t =σ(W i ·[h t-1 ,x t +b i ) (7)

[0111] C t =f t ·C t-1 +i t ·tanh(W C ·[h t-1 ,x t +b C) (8)

[0112] o t = σ(W o · [h t-1 , x t + b o ) (9)

[0113] h t = o t · tanh(C t ) (10)

[0114] Among them, f t is the forget gate, i t is the input gate, C t is the cell state update, o t is the output gate, h t is the hidden state update;

[0115] S3-2. Using the mutual information analysis algorithm, analyze the multi-dimensional feature correlation, locate the fire risk signal, and combine historical data to construct a fire risk model based on feature correlation, focusing on capturing the temperature rise caused by short circuits and the abnormal gas concentration generated by arcs, specifically as shown in Equation (11):

[0116]

[0117] Among them, I(X; Y) is the mutual information between variables X and Y, X and Y are random variables, P(x, y) is the joint probability distribution of X = x and Y = y, P(x) is the marginal probability distribution of X = x, and P(y) is the marginal probability distribution of Y = y.

[0118] It should be noted that in the identification of fire risk signals, the mutual information analysis algorithm is mainly used to measure the correlation between multi-dimensional features, analyze the correlation of multi-dimensional features, and find fire risk signals through feature association. For example, current and voltage fluctuations, short circuits or overloads may cause temperature increases; temperature and gas concentration, high temperatures may cause material decomposition and release toxic gases; smoke and temperature changes, an increase in smoke concentration may be accompanied by an abnormal increase in temperature.

[0119] The acquisition and processing of feature data collect time series data through multi-dimensional sensors, preprocess the data, calculate the mutual information between any two features, and determine whether there is a strong correlation between them. If the mutual information value is high, it indicates that the two may simultaneously exhibit abnormalities and are associated with fire risks.

[0120] Historical data provides the basic statistical information of abnormal features. By the distribution of historical normal and abnormal data, define the normal range and abnormal range of each feature, and statistically calculate the mutual information values of multi-dimensional features in normal and abnormal states to find feature combinations with high correlation degrees.

[0121] Step S4, through multi-level data fusion, GAN generates abnormal data and fuzzy logic reasoning;

[0122] Among them, in step S4, the following sub-steps are also included:

[0123] S4-1, Use the weighted average method to fuse multi-sensor data, and the weights are dynamically adjusted according to the noise variance of the sensors and the real-time environment, specifically as shown in equations (12)-(13):

[0124]

[0125] where, w i is the weight, and x i is the sensor data; is the noise variance of sensor i;

[0126] S4-2, Use the generative adversarial network (Gan) to generate simulated abnormal data. The generator G(z) inputs the random noise z and outputs the simulated abnormal data; the discriminator D(x) is used to distinguish real data from generated data. The objective function of Gan is specifically as shown in equation (14):

[0127]

[0128] where, G(z) is the generator, D(x) is the discriminator, z is the random noise vector, x is the real data sample, is the average value of the real data distribution on pdata(x), is the average value of the real data distribution on p x (z), pdata(x) is the probability distribution of the real data, and p x (z) is the probability distribution of the random noise;

[0129] S4-3, In a complex environment, use fuzzy logic decision-making to comprehensively analyze multi-dimensional features and judge the fire risk level. Fuzzy logic consists of three steps: fuzzy rules, fuzzification function, and defuzzification, specifically as shown in equations (15)-(17):

[0130] IF AND THEN Equation (15)

[0131]

[0132] where, μ(x) is the membership value of the input x in the interval [a, b]; y is the output after defuzzification, μ i is the membership degree, y i is the scalar value corresponding to the risk level, and n is the total number of fuzzy rules;

[0133] Fuzzy logic decision-making maps multi-dimensional feature inputs to risk levels by setting a rule set; according to the range of feature changes, the fire risk is divided into four levels: low risk, medium risk, high risk, and emergency risk.

[0134] It should be noted that the risk levels are usually divided according to the combination of multi-dimensional features, including the following key parameters: temperature change rate, smoke concentration, gas concentration, humidity, dust concentration. According to the range and combination of the index values, the following risk levels can be divided:

[0135] Low risk, the environment is normal or there are no fire characteristics, temperature < 30°C, low smoke concentration < 5 ppm, and normal gas concentration;

[0136] Medium risk, abnormalities are initially detected but not significantly harmful, temperature 30 - 50°C, slightly higher smoke concentration 5 - 10 ppm, and slightly abnormal gas concentration;

[0137] High risk, there are precursors of fire and timely intervention is required, temperature > 50°C, higher smoke concentration 10 - 20 ppm, and abnormal increase in gas concentration;

[0138] Emergency risk, obvious fire characteristics, immediate response is required, temperature > 70°C, extremely high smoke concentration > 20 ppm, and sharp increase in gas concentration;

[0139] The fuzzy logic system maps multi-dimensional feature inputs to risk levels by setting a rule set:

[0140] Rule 1: IF the temperature is normal AND the smoke concentration is low THEN low risk;

[0141] Rule 2: IF the temperature rises AND the smoke concentration is medium THEN medium risk;

[0142] Rule 3: IF the temperature is high AND the smoke concentration is high THEN high risk;

[0143] Rule 4: IF the temperature is high AND the smoke concentration is high AND the gas concentration is high THEN emergency risk.

[0144] Step S5, through hierarchical alarm, intelligent fire extinguishing strategy, and remote control, achieve full-process coverage from fire detection to emergency response;

[0145] Among them, in step S5, the following sub-steps are also included:

[0146] S5-1, divide the alarm level according to the fire level division and real-time environmental data, add a progressive alarm mechanism, which is divided into "warning", "danger", and "emergency", and determine the response strategy according to the risk level:

[0147] Warning level: When the fire risk level is medium risk and the real-time data change rate is low, the system triggers an alarm at the "Warning" level; at this time, the user is reminded to pay attention to potential risks, but no emergency response is required;

[0148] Danger level: When the fire risk level is high risk, regardless of the real-time data change rate, the system triggers an alarm at the "Danger" level, or when the fire risk level is medium risk but the real-time data change rate is high, the system also triggers an alarm at the "Danger" level; at this time, it is recommended that the user immediately check for abnormalities and take preventive measures;

[0149] Emergency level: When the fire risk level is emergency risk, the system immediately triggers an alarm at the "Emergency" level, or when the fire risk level is high risk and the real-time data change rate increases significantly, the system also upgrades the alarm to the "Emergency" level and activates the fire extinguishing device and emergency linkage response;

[0150] S5-2: Determine the fire type based on the sensor characteristics, formulate the mapping rules between the fire type and the fire extinguishing equipment, and use the reinforcement learning algorithm to optimize the action strategy of the fire extinguishing equipment, specifically as shown in Equation (18):

[0151]

[0152] Among them, Q(s,a) is the value of executing action a in the current state a, α is the learning rate, and r is the reward value; mapping rules: dry powder fire extinguishers are suitable for arc fires, and gas fire extinguishers are suitable for fires in equipment areas;

[0153] S5-3: Deploy an Internet of Things-based remote control system, use the Internet of Things platform to send alarm information to users in real time, including fire registration, location, and sensor data, and allow staff to view the alarm information through mobile devices and manually or automatically control the fire extinguishing equipment.

[0154] It should be noted that the characteristics of the electrical fire scene are that the temperature rises abnormally, the gas concentration is normal, and the smoke concentration is low; this is an early characteristic of a typical electrical fire, which may be caused by short circuits, overloads, or arcs.

[0155] Selection and strategy of fire extinguishing equipment: For the special needs of electrical fires, dry powder fire extinguishers and CO fire extinguishers are preferred; due to their non-conductive characteristics, dry powder fire extinguishers can safely handle fires in live environments and are suitable for fire sources caused by short circuits or overloads; CO fire extinguishers effectively control the spread of fire by quickly cooling and isolating oxygen; the system determines the fire level based on sensor data in the initial stage of the fire and automatically matches the appropriate fire extinguishing equipment.

[0156] The system associates the electrical fire characteristics with the fire extinguishing equipment based on the rule engine and sets the following rules:

[0157] Rule 1: IF fire type = electrical fire THEN choose dry powder fire extinguisher.

[0158] Rule 2: IF fire level = high risk THEN activate CO2 fire extinguisher and dry powder fire extinguisher linkage response.

[0159] Step S6, visualizing the fire scene with the help of three-dimensional thermal images, flame dynamics analysis and augmented reality technology.

[0160] Wherein, in step S6, the following sub-steps are also included:

[0161] S6-1, by combining the infrared camera and sensor data, a three-dimensional thermal map is generated to provide the three-dimensional position and heat distribution of the fire source, and the triangulation positioning method based on multi-point ranging is used to calculate the fire source position, as shown in formula (19):

[0162] (x, y, z) = f(d1, d2, d3) Formula (19)

[0163] Among them, d1, d2, and d3 are the distances between the infrared camera and the fire source, and the three-dimensional coordinates of the fire source are calculated by the angles between the nodes;

[0164] S6-2 uses edge AI technology to analyze flame dynamic characteristics and abnormal behaviors in real time: dynamic characteristics include flame area change rate, and abnormal behaviors include equipment explosion and smoke diffusion; convolutional neural network is used to extract flame texture and dynamic characteristics, determine the nature of the fire source, and provide flame category and dynamic information;

[0165] S6-3, use augmented reality technology to superimpose the fire source location map and the on-site equipment layout, and visualize the fire source location by displaying the fire source location in real time through AR equipment; provide an evacuation route in combination with the fire scene layout; superimpose the location of fire-fighting equipment, guide rescue personnel to perform operations and provide fire-fighting assistance.

[0166] It should be noted that convolutional neural networks (CNNs) are used to extract the texture and dynamic features of flames. Flame texture features include flame color, texture pattern (such as smooth or jumping edges), and shape (such as spikes or fan-shaped expansions), which are usually identified through spatial feature extraction; flame dynamic features are the patterns of flame changes over time, including edge jumping, area expansion, and brightness fluctuations. It is necessary to combine time series analysis to extract dynamic behaviors. For example: for ordinary flames, red, yellow and orange textures, fluctuating edges and diffusion dynamic features are extracted; for arc flames, high-brightness edges, stable distributions and tiny dynamic features are extracted; for oil flames, fast jumping edges and obvious diffusion trends are detected.

[0167] Extract the flame region from the video frame, remove interference information through image segmentation or background subtraction algorithms, standardize the size of the image data, and normalize the pixel values to the [0,1] interval to ensure unified input. Convert to a color space suitable for flame analysis, enhance the hue and saturation characteristics of the flame, enhance the diversity of training data, and simulate different flame scenarios.

[0168] The above are only the preferred embodiments of the present invention and are not used to limit the present invention. For those skilled in the art, various changes and modifications can be made to the present invention. Any modifications, equivalent replacements, improvements, etc. made within the spirit and principle of the present invention shall be included within the protection scope of the present invention.

Claims

1. An intelligent electrical fire recognition system based on multi-dimensional sensor fusion, characterized in that, It includes the following steps: Step S1, constructing a benchmark environment model through multi-sensor scanning, data dimensionality reduction, and intelligent node deployment; Step S2, denoising through edge computing, aligning multi-sensor time series by dynamic time warping, and processing high-priority data in real time through a hierarchical mechanism; Step S3, establishing a fire feature modeling system through LSTM time series analysis, mutual information correlation mining, and self-supervised learning; Step S4, through multi-level data fusion, generating abnormal data by GAN, and fuzzy logic reasoning; Step S5, achieving full-process coverage from fire detection to emergency response through hierarchical alarm, intelligent fire extinguishing strategy, and remote control; Step S6, visualizing the fire scene with the help of 3D heat maps, flame dynamic analysis, and augmented reality technology.

2. The intelligent electrical fire identification system based on multi-dimensional sensor fusion according to claim 1, characterized in that: Wherein in step S1, the following sub-steps are further included: S1-1, using temperature sensors, smoke sensors, gas sensors, and pyrolysis particle sensors, scanning the initial environmental data through distributed node layout, and dynamically adjusting the sensor response error by using the Bayesian optimization algorithm, specifically as shown in formula (1): P(θ|D)∝P(D|θ)P(θ) Formula (1) Wherein, P(θ|D) is the posterior distribution of the sensor parameters, P(D|θ) is the current scanning result, and P(θ) is the historical data; S1-2, collecting the initial data matrix X of the sensors, including temperature, smoke concentration, gas concentration, and humidity data, calculating the covariance matrix, extracting eigenvalues and eigenvectors, and selecting the main components for dimensionality reduction to form a benchmark model, specifically as shown in formula (2): Among them, is the sample normalization factor, n is the number of samples, X is the initial data matrix, and X T is the transpose of the initial data matrix; S1-3, deploying portable intelligent nodes, arranging multiple nodes in high-fire areas to form a monitoring coverage network, arranging coefficient nodes in areas with lower risks, and the portable nodes are equipped with magnetic bases and movable brackets.

3. The intelligent electrical fire identification system based on multi-dimensional sensor fusion according to claim 1, characterized in that: Wherein in step S2, the following sub-steps are further included: S2-1, introducing edge computing, using the Kalman filtering algorithm for data preprocessing, which is completed locally on the sensor to remove high-frequency random noise, specifically as shown in formula (3): x k = x k-1 + K k (Z k - Hx k-1 ) Equation (3) where, x k is the state estimate at time k, x k-1 is the state estimate at the previous time k-1, z k is the observation at the current time k, K k is the Kalman gain, and H is the observation model matrix; S2-2, using dynamic time warping (DTW) to perform time alignment on the time series data from multiple sensors, capturing the cross-dimensional correlation trend, and using clustering analysis to identify abnormal patterns in the sensor data, specifically as shown in formula (4)-formula (5); DTW(i,j) = d(x i , y j ) +min{DTW(i-1,j),DTW(i,j-1),DTW(i-1,j -1)} Formula (4) where DTW(i,j) is the cumulative distance of the optimal alignment path to the point (i,j), and d(x i ,y j 0 is the Euclidean distance between the i-th point and the j-th point, DTW(i - 1,j) matches from the upper point to the current point, DTW(i,j - 1) matches from the left point to the current point, and DTW(i - 1,j - 1) matches from the upper left diagonal point to the current point; J is the objective function of clustering, K is the number of clusters, C k is the set of the k-th cluster, x i is the i-th data point in the dataset, μ k is the centroid of the k-th cluster; S2-3, adding a dynamic data hierarchical mechanism, classifying the sensor data into high-priority and medium-low-priority according to the change rate and risk level of the sensor data, using a priority queue, and the high-priority data triggers alarms and further analysis in real time, while the medium-low-priority data is processed later.

4. The intelligent electrical fire identification system based on multi-dimensional sensor fusion according to claim 1, characterized in that: Wherein in step S3, the following sub-steps are further included: S3-1. Use the LSTM time series feature analysis algorithm to capture the early trends of fires, and model long-term dependencies through the mechanisms of forget gates, input gates, and output gates, as shown in Equations (6)-(10): f t = σ(W f · [h t-1 , x t + b f ) (6) i t = σ(W i · [h t-1 , x t + b i )(7) C t = f t ·C t-1 + i t ·tanh(W C ·[h t-1 , x t + b C )(8) o t = σ(W o · [h t-1 , x t + b o ) (9) h t = o t ·tanh(C t ) (10) where f t is the forget gate, i t is the input gate, C t is the cell state update, o t is the output gate, h t is the hidden state update; S3-2. Use the mutual information analysis algorithm to analyze the correlations of multi-dimensional features, locate fire risk signals, and construct a fire risk model based on feature correlations by combining historical data. Focus on capturing the temperature rise caused by short circuits and abnormal gas concentrations generated by electric arcs, as shown in Equation (11): Among them, I(X; Y) is the mutual information between variables X and Y, X and Y are random variables, P(x,y) is the joint probability distribution of X = x and Y = y, P(x) is the marginal probability distribution of X = x, and P(y) is the marginal probability distribution of Y = y.

5. The intelligent electrical fire identification system based on multi-dimensional sensor fusion according to Claim 1, wherein: Among them, in step S4, the following sub-steps are further included: S4-1. Use the weighted average method to fuse multi-sensor data, and the weights are dynamically adjusted according to the noise variances of the sensors and the real-time environment, as shown in Equations (12)-(13): where, w i is the weight, and x o is the sensor data; is the noise variance of sensor i; S4-2. Use the generative adversarial network (Gan) to generate simulated abnormal data. The generator G(z) inputs random noise z and outputs simulated abnormal data; the discriminator D(x) is used to distinguish real data from generated data. The objective function of Gan is as shown in Equation (14): Among them, G(z) is the generator, D(x) is the discriminator, z is the random noise vector, and x is the real data sample. is the average value of the real data distribution on pdata(x). is the average value of the real data distribution on p x (z), pdata(x) is the probability distribution of the real data, and p z (z) is the probability distribution of the random noise. S4-3. In a complex environment, use fuzzy logic decision-making to comprehensively analyze multi-dimensional features and judge the fire risk level. Fuzzy logic consists of three steps: fuzzy rules, fuzzification functions, and defuzzification, as shown in Equations (15)-(17): IF AND THEN Equation (15) Among them, μ(x) is the membership value of the input x in the interval [a, b]; y is the output after defuzzification, and μ i is the membership degree, and y i is the scalar value corresponding to the risk level, and n is the total number of fuzzy rules; Fuzzy logic decision-making maps the input of multi-dimensional features to the risk level by setting a rule set; according to the range of feature changes, the fire risk is divided into four levels: low risk, medium risk, high risk, and emergency risk.

6. The intelligent electrical fire identification system based on multi-dimensional sensor fusion according to Claim 1, wherein: Among them, in step S5, the following sub-steps are further included: S5-1. Divide the alarm levels according to the fire level classification and real-time environmental data, and add a progressive alarm mechanism, which is divided into "warning", "danger", and "emergency", and determine the response strategy according to the risk level: Warning level: When the fire risk level is medium risk and the change rate of real-time data is slow, the system triggers a "warning" level alarm; at this time, the user is reminded to pay attention to potential risks, but no emergency response is required; Danger level: When the fire risk level is high risk, regardless of the change rate of real-time data, the system triggers a "danger" level alarm, or when the fire risk level is medium risk but the change rate of real-time data is fast, the system also triggers a "danger" level alarm; Emergency level: When the fire risk level is emergency risk, the system immediately triggers an "emergency" level alarm, or when the fire risk level is high risk and the change rate of real-time data increases significantly, the system also upgrades the alarm to the "emergency" level and activates the fire extinguishing device and emergency linkage response; S5-2. Determine the fire type based on the sensor characteristics, formulate the mapping rules between the fire type and the fire extinguishing equipment, and use the reinforcement learning algorithm to optimize the action strategy of the fire extinguishing equipment, as shown in Equation (18): Among them, Q(s,a) is the value of executing action a in the current state s, α is the learning rate, and r is the reward value; the mapping rule is that dry powder fire extinguishers are applicable to arc fires, and gas fire extinguishers are applicable to fires in the equipment area; S5-3. Deploy an Internet of Things-based remote control system, and use the Internet of Things platform to send alarm information to users in real time, including fire registration, location, and sensor data, allowing staff to view the alarm information through mobile devices and manually or automatically control the fire extinguishing equipment.

7. The intelligent electrical fire identification system based on multi-dimensional sensor fusion according to claim 1, characterized in that: Among them, in step S6, the following sub-steps are further included: S6-1. Combine the infrared camera and sensor data to generate a three-dimensional thermal map, provide the three-dimensional position of the fire source and the heat distribution, and use the triangulation method based on multi-point ranging to calculate the position of the fire source, as shown in Equation (19): (x,y,z) = f(d1,d2,d3) Equation (19) Among them, d1, d2, and d3 are the distances between the infrared camera and the fire source, and the three-dimensional coordinates of the fire source are calculated through the angles between the nodes; S6-2. Adopt edge AI technology to analyze the dynamic characteristics and abnormal behaviors of the flame in real time: the dynamic characteristics include the change rate of the flame area, and the abnormal behaviors include equipment explosion and smoke filling; use the convolutional neural network to extract the flame texture and dynamic characteristics, determine the nature of the fire source, and provide the flame category and dynamic information; S6-3. Use augmented reality technology to superimpose the fire source location map and the on-site equipment layout, and display the fire source location in real time through the AR device to realize the visualization of the fire source location; provide the evacuation path in combination with the layout of the fire scene; superimpose the location of the fire extinguishing equipment to guide the rescue personnel to perform operations and provide fire extinguishing assistance.

Images