Extremely early-stage monitoring and early-warning method and device for over-temperature pyrolysis electrical fire

Abstract

The invention discloses an ultra-early monitoring and early warning method and device for an over-temperature pyrolysis electrical fire hazard, and belongs to the technical field of electrical equipment safety monitoring and fire hazard early warning, and the method comprises the steps: obtaining multi-dimensional environment parameters in a power distribution facility in real time; inputting the multi-dimensional environmental parameters into a pre-trained neural network model, and outputting an early warning probability reflecting a real pyrolysis hidden danger and a false alarm probability reflecting environmental interference; on the basis of the early warning probability and the false alarm probability, dynamically updating the weight of each environment parameter, and inputting the environment parameters with the updated weights into the neural network model again to obtain the updated early warning probability and false alarm probability; and performing comprehensive judgment by combining a preset multi-stage judgment rule so as to distinguish real pyrolysis and interference factors, and outputting an early warning signal when the real pyrolysis is judged. According to the invention, through a neural network and dynamic weight cooperation mechanism, the false alarm rate is significantly reduced, and extremely early precise early warning of the electrical fire is realized.

Description

Technical Field

[0001] This invention belongs to the field of electrical equipment safety monitoring and fire early warning technology, and more specifically, relates to a method and device for early monitoring and warning of electrical fires caused by overheating pyrolysis. Background Technology

[0002] In power system operation, substations, ring main units, high and low voltage switchgear, cable trays, and other power distribution facilities are the core links in power transmission and distribution, and their safe and stable operation directly determines the overall security of the power grid. During long-term operation, electrical equipment within these facilities is prone to heat accumulation due to resistance losses, iron losses, dielectric losses, overload, external heat sources, and poor heat dissipation in some facilities. This heat accumulation leads to the thermal decomposition of organic insulation materials, causing a decline in the insulation performance and mechanical strength of the equipment, ultimately significantly increasing the risk of electrical fires.

[0003] Residual current detection suffers from numerous false alarms, and fault arc detection is difficult to locate. Existing pyrolysis particle detectors generally suffer from poor response, narrow applicability, insufficient multi-parameter analysis, and high false alarm rates. Traditional monitoring methods are also limited by voltage levels, alarm lag, reliance on single parameters leading to false alarms or missed alarms, and monitoring blind spots.

[0004] These limitations collectively restrict the reliability, accuracy, and practicality of early detection technologies for electrical fires. Summary of the Invention

[0005] To address the shortcomings of existing technologies, this invention provides a method and device for early monitoring and warning of electrical fires caused by overheating pyrolysis.

[0006] The present invention adopts the following technical solution.

[0007] The first aspect of the present invention provides a method for very early monitoring and warning of electrical fires caused by superheated pyrolysis, comprising: Real-time acquisition of multi-dimensional environmental parameters inside power distribution facilities, including temperature, temperature change rate, thermal decomposition gas concentration, thermal decomposition gas concentration change rate, and pyrolysis particle concentration with different particle size characteristics. Multidimensional environmental parameters are input into a pre-trained neural network model, which outputs the early warning probability reflecting the actual pyrolysis hazard and the false alarm probability reflecting environmental disturbance. Based on the early warning probability and false alarm probability, the weights of each environmental parameter are dynamically updated. The environmental parameters with updated weights are then input into the neural network model to obtain the updated early warning probability and false alarm probability. Based on the updated warning probability and false alarm probability, a comprehensive judgment is made in combination with the preset multi-level judgment rules to distinguish between real pyrolysis and interference factors. When it is determined to be real pyrolysis, a warning signal is output.

[0008] Optionally, the neural network model is a BP neural network comprising an input layer, a hidden layer, and an output layer; The number of input layer nodes is consistent with the number of multidimensional environment parameters; The output layer is configured with two nodes, which output the early warning probability and the false alarm probability, respectively.

[0009] Optionally, the adjustment of the dynamic weights actually includes periodic updates and triggered updates, wherein: Triggered updates are initiated when the real-time value of any environmental parameter fluctuates beyond a set percentage of its base value. The adjustment rules for dynamic weights are as follows: The dynamic weight value of each environmental parameter is equal to its preset base weight value multiplied by an adjustment factor; The adjustment factor includes the product of the false alarm probability and the first coefficient factor, and the product of the operating condition correction factor and the second coefficient factor.

[0010] Optionally, the preset base weights are set according to the environmental parameter categories, including: Set a first basic weight value for temperature and the rate of temperature change; A second basic weight value is set for the concentration of thermally decomposed gases and the rate of change of thermally decomposed gases concentration; A third basic weighting value is set for the concentration of pyrolysis particles with different particle size characteristics; The third basic weight value is greater than the second basic weight value, and the second basic weight value is greater than the first basic weight value.

[0011] Optionally, the operating condition correction coefficient is determined based on the environmental parameter category and the current operating conditions, and its determination logic includes: Regarding temperature and temperature change rate, if the current temperature is not lower than the preset temperature threshold, its operating condition correction coefficient value is adjusted to the first operating condition correction coefficient value; otherwise, its operating condition correction coefficient remains at 0. Based on the concentration of thermally decomposed gases and the rate of change of thermally decomposed gases, the current humidity inside the power distribution facility is obtained. If the current humidity is not lower than the preset humidity threshold, its operating condition correction coefficient value is adjusted to the second operating condition correction coefficient value; otherwise, its operating condition correction coefficient value remains at 0. The operating condition correction factor for the pyrolysis particle concentration with different particle size characteristics is determined jointly based on the current humidity and the false alarm probability: If the current humidity is not lower than the preset humidity threshold and the false alarm probability is not lower than the preset false alarm threshold, the operating condition correction coefficient is the sum of the second operating condition correction coefficient and the third operating condition correction coefficient. If the current humidity is not lower than the preset humidity threshold and the false alarm probability is lower than the preset false alarm threshold, the operating condition correction coefficient value is the second operating condition correction coefficient value; If the current humidity is lower than the preset humidity threshold and the false alarm probability is not lower than the preset false alarm threshold, the working condition correction coefficient value is the third working condition correction coefficient value; If the current humidity is lower than the preset humidity threshold and the false alarm probability is lower than the preset false alarm threshold, the operating condition correction coefficient value remains at 0; Among them, the correction coefficient value for the first working condition is greater than 0, while the correction coefficient values ​​for the second and third working conditions are less than 0.

[0012] Optionally, a comprehensive judgment can be made by combining preset multi-level judgment rules, including: If the updated warning probability is not lower than the first alarm threshold and the updated false alarm probability is lower than the first false alarm threshold, then it is determined to be a real pyrolysis and a warning signal is triggered. If the updated warning probability is lower than the second alarm threshold, or the updated false alarm probability is not lower than the second false alarm threshold, it is determined that there is no pyrolysis hazard or environmental interference, and the warning signal is not triggered. If the updated warning probability and the updated false alarm probability do not meet the above two conditions, then the system is determined to enter the review interval and the high-frequency sampling review procedure is initiated.

[0013] Optionally, the high-frequency sampling verification procedure includes: Multiple high-frequency samples are taken within a preset time period, and the mean early warning probability and the mean false alarm probability are recalculated based on the sampling results. If the average early warning probability is not lower than the third alarm threshold and the average false alarm probability is lower than the third false alarm threshold, then it is determined to be a real pyrolysis and an early warning signal is triggered; otherwise, it is determined to be that there is no pyrolysis hazard or environmental interference, and the corresponding environmental parameter data and probability data are recorded for iterative optimization of the neural network model.

[0014] The second aspect of this invention provides an early warning device for superheated pyrolysis electrical fires, used to implement the early warning method for superheated pyrolysis electrical fires described in the first aspect of this invention, comprising: The monitoring module, communication module, and power supply module are connected via electrical connections or wireless communication protocols. The monitoring module is used to acquire multi-dimensional environmental parameters inside the power distribution facility in real time, including temperature, temperature change rate, thermal decomposition gas concentration, gas concentration change rate, and pyrolysis particle concentration with different particle size characteristics. These multi-dimensional environmental parameters are then input into its built-in pre-trained neural network model to obtain the early warning probability reflecting actual pyrolysis hazards and the false alarm probability reflecting environmental interference. Based on the early warning and false alarm probabilities, the weights of each environmental parameter are dynamically updated. The weighted and updated environmental parameters are then input into the neural network model again to obtain updated early warning and false alarm probabilities. Based on these updated probabilities, a pre-set multi-level judgment rule is invoked for comprehensive judgment to distinguish between actual pyrolysis and interference factors, and an early warning command is generated when actual pyrolysis is determined. The communication module is connected to the monitoring module and is used to upload the early warning instructions, monitoring data and alarm information generated by the monitoring module to the remote monitoring platform; The power module is connected to the monitoring module and the communication module to provide operating power for the entire device.

[0015] A third aspect of the present invention provides an electronic device, including a memory, a processor, and a computer program stored in the memory and executable on the processor. When the computer program is loaded onto the processor, it implements a method for early monitoring and warning of superheated pyrolysis electrical fires according to a first aspect of the present invention.

[0016] A fourth aspect of the present invention provides a computer-readable storage medium storing a computer program, which, when executed by a processor, implements a method for early monitoring and warning of superheated pyrolysis electrical fires according to a first aspect of the present invention.

[0017] Compared with the prior art, the beneficial effects of the present invention include at least the following: 1. This invention introduces a false alarm suppression mechanism that combines a BP neural network with dynamic weights. By combining the false alarm probability calculated by the neural network with the parameter weights that are adjusted in real time based on the operating conditions, it achieves dual shielding against environmental interference such as water mist and dust. This effectively solves the problems of high false alarm rate and poor reliability of traditional monitoring technologies, reducing the system false alarm rate to below 5%.

[0018] 2. This invention adopts a non-contact, multi-sensor collaborative full-area monitoring solution, integrating temperature, gas and multi-size particulate matter sensors. It does not require direct contact with electrical equipment or fixed-point installation, solving the problems of coverage blind spots and installation limitations of traditional contact or single-point monitoring, and realizing comprehensive and blind-spot-free monitoring of potential pyrolysis areas inside the cabinet.

[0019] 3. This invention, through multi-dimensional feature perception and neural network fusion analysis, can capture trace amounts of pyrolysis gas and characteristic particles released by insulating materials at temperatures far below their ignition point (such as 90°C), solving the problem of alarm lag in traditional temperature and smoke detectors and achieving very early warning of fire hazards.

[0020] 4. This invention, through the preset multi-level intelligent judgment rules and model iterative optimization mechanism, can perform high-frequency verification of fuzzy risk signals and record and self-learn the judgment results. It solves the problems of rigidity and inability to adapt to complex working conditions in traditional threshold alarm methods, and improves the long-term adaptability and early warning accuracy of the system in different operating environments. Attached Figure Description

[0021] Figure 1 This is a flowchart of a method provided according to an embodiment of the present invention. Detailed Implementation

[0022] To make the objectives, technical solutions, and advantages of this invention clearer, the technical solutions of this invention will be clearly and completely described below with reference to the accompanying drawings of the embodiments of this invention. The described embodiments are merely some embodiments of this invention, and not all embodiments. Based on the spirit of this invention, all other embodiments obtained by those skilled in the art without creative effort are within the protection scope of this invention.

[0023] In Example 1, this invention provides a method for very early monitoring and warning of electrical fires caused by superheated pyrolysis, such as... Figure 1 As shown, it includes the following steps: Step 1: Real-time acquisition of multi-dimensional environmental parameters inside the power distribution facility, including temperature, temperature change rate, thermal decomposition gas concentration, thermal decomposition gas concentration change rate, and pyrolysis particle concentration with different particle size characteristics.

[0024] Preferably, the concentrations of pyrolysis particles with different particle size characteristics include the concentrations of PM1.0, PM2.5, and PM10.

[0025] For example, the power distribution facility is a low-voltage distribution cabinet, and the real-time values ​​of seven parameters inside the low-voltage distribution cabinet are obtained, including temperature. Temperature change rate Concentration of gases decomposed by heat Rate of change of gas concentration due to thermal decomposition PM1.0 particle concentration PM2.5 particle concentration and PM10 particle concentration .

[0026] Furthermore, in S1: temperature and the rate of temperature change Real-time data is collected via a temperature sensor; concentration of thermally decomposed gases. and the rate of change of gas concentration due to thermal decomposition Real-time data collection is performed using pollutant gas sensors; PM1.0 particle concentration. PM2.5 particle concentration and PM10 particle concentration Real-time data is collected using a laser-based inhalable dust sensor.

[0027] Step 2: Input the multidimensional environmental parameters into the pre-trained neural network model and output the early warning probability reflecting the actual pyrolysis hazard and the false alarm probability reflecting environmental interference.

[0028] Preferably, the neural network model is a BP neural network comprising an input layer, a hidden layer, and an output layer; The number of input layer nodes is consistent with the number of multidimensional environment parameters; The output layer is configured with two nodes, which output the early warning probability and the false alarm probability, respectively.

[0029] Furthermore, step 2 specifically includes: Using a BP neural network model with 7 parameters , , , , , , As The vector is input into the input layer; In the hidden layer, through the activation function Perform nonlinear operations and output ,in For the input parameter vector, This is the weight matrix. The bias vector is calculated as follows: In the output layer, a logistic regression is performed on the computation results of the hidden layer. Finally, the output vector The specific calculations are as follows:

[0030] vector It is divided into two categories, representing the probability of normal alarms and false alarms, respectively.

[0031] Step 3: Based on the warning probability and false alarm probability, dynamically update the weights of each environmental parameter, and input the environmental parameters with updated weights back into the neural network model to obtain the updated warning probability and false alarm probability.

[0032] Preferably, the adjustment of the dynamic weights actually includes periodic updates and triggered updates, wherein: Triggered updates are initiated when the real-time value of any environmental parameter fluctuates beyond a set percentage of its base value.

[0033] Preferably, the adjustment rule for the dynamic weights is as follows: The dynamic weight value of each environmental parameter is equal to its preset base weight value multiplied by an adjustment factor; The adjustment factor includes the product of the false alarm probability and the first coefficient factor, and the product of the operating condition correction factor and the second coefficient factor.

[0034] More preferably, the preset base weights are set according to the categories of environmental parameters, including: Set a first basic weight value for temperature and the rate of temperature change; A second basic weight value is set for the concentration of thermally decomposed gases and the rate of change of thermally decomposed gases concentration; A third basic weighting value is set for the concentration of pyrolysis particles with different particle size characteristics; The third basic weight value is greater than the second basic weight value, and the second basic weight value is greater than the first basic weight value.

[0035] More preferably, the operating condition correction coefficient is determined based on the environmental parameter category and the current operating conditions, and its determination logic includes: Regarding temperature and temperature change rate, if the current temperature is not lower than the preset temperature threshold, its operating condition correction coefficient value is adjusted to the first operating condition correction coefficient value; otherwise, its operating condition correction coefficient remains at 0. Based on the concentration of thermally decomposed gases and the rate of change of thermally decomposed gases, the current humidity inside the power distribution facility is obtained. If the current humidity is not lower than the preset humidity threshold, its operating condition correction coefficient value is adjusted to the second operating condition correction coefficient value; otherwise, its operating condition correction coefficient value remains at 0. The operating condition correction factor for the pyrolysis particle concentration with different particle size characteristics is determined jointly based on the current humidity and the false alarm probability: If the current humidity is not lower than the preset humidity threshold and the false alarm probability is not lower than the preset false alarm threshold, the operating condition correction coefficient is the sum of the second operating condition correction coefficient and the third operating condition correction coefficient. If the current humidity is not lower than the preset humidity threshold and the false alarm probability is lower than the preset false alarm threshold, the operating condition correction coefficient value is the second operating condition correction coefficient value; If the current humidity is lower than the preset humidity threshold and the false alarm probability is not lower than the preset false alarm threshold, the working condition correction coefficient value is the third working condition correction coefficient value; If the current humidity is lower than the preset humidity threshold and the false alarm probability is lower than the preset false alarm threshold, the operating condition correction coefficient value remains at 0; Among them, the correction coefficient value for the first working condition is greater than 0, while the correction coefficient values ​​for the second and third working conditions are less than 0.

[0036] For example, the adjustment of the dynamic weights actually includes periodic updates and triggered updates, wherein: The periodic update cycle can be set to 30 seconds; the triggered update is initiated when the real-time value fluctuation of any environmental parameter exceeds 20% of its base value.

[0037] The adjustment rules for dynamic weights are as follows: The dynamic weight value of each environmental parameter is equal to its preset base weight value multiplied by an adjustment factor; The adjustment factor includes the product of the false alarm probability and the first coefficient factor, and the product of the operating condition correction factor and the second coefficient factor.

[0038] In a specific example, the first coefficient factor can be 0.3, and the second coefficient factor can be 0.2. The weight adjustment formula is established as follows: Weight W for each parameter. n =Basic weight W 0n ×[1+(O2×0.3)+(Operating condition correction factor K×0.2)]; Basic weight value W 0n The preset values ​​are: a first basic weight value for temperature and the rate of temperature change, for example, 0.25; a second basic weight value for the concentration of thermally decomposed gases and their rate of change, for example, 0.3; and a third basic weight value for the concentration of pyrolysis particles with different particle size characteristics (such as PM1.0, PM2.5, PM10), for example, 0.45.

[0039] In this configuration, the third basic weight value is greater than the second basic weight value, and the second basic weight value is greater than the first basic weight value, reflecting the priority given to particle concentration signals.

[0040] The operating condition correction coefficient (K) is determined based on the parameter category and real-time operating conditions, and its value selection logic is as follows: For temperature and temperature change rate: If the current temperature is not lower than the preset temperature threshold (e.g., 70℃), its K value is taken as the first working condition correction coefficient value (e.g., +0.2) which is greater than 0, so as to increase the weight of temperature parameters; otherwise, its K value is 0.

[0041] For gas concentration and its rate of change: if the current humidity is not lower than the preset humidity threshold (e.g., 80%RH), its K value is taken as the second working condition correction coefficient value (e.g., -0.1) which is less than 0, so as to reduce the weight of gas parameters that may be affected by water mist in high humidity environment; otherwise, its K value is 0.

[0042] For pyrolysis particle concentration: its K value is determined by both humidity and false alarm probability. If the humidity is not lower than the threshold (e.g., 80%RH), a second operating condition correction factor (e.g., -0.1) is introduced; if the false alarm probability is not lower than its preset threshold (e.g., 0.4), a third operating condition correction factor (e.g., -0.15) less than 0 is introduced. The final K value of the particle concentration is the algebraic sum of the above introduced values.

[0043] For example, when both humidity and false alarm probability conditions are met, K = (-0.1) + (-0.15) = -0.25; when only one condition is met, the corresponding correction coefficient value is taken; when neither condition is met, the value of K is 0.

[0044] Step 4: Based on the updated warning probability and false alarm probability, and combined with the preset multi-level judgment rules, a comprehensive judgment is made to distinguish between real pyrolysis and interference factors. When it is determined to be real pyrolysis, a warning signal is output.

[0045] Preferably, the comprehensive judgment based on preset multi-level judgment rules includes: If the updated warning probability is not lower than the first alarm threshold and the updated false alarm probability is lower than the first false alarm threshold, then it is determined to be a real pyrolysis and a warning signal is triggered. If the updated warning probability is lower than the second alarm threshold, or the updated false alarm probability is not lower than the second false alarm threshold, it is determined that there is no pyrolysis hazard or environmental interference, and the warning signal is not triggered. If the updated warning probability and the updated false alarm probability do not meet the above two conditions, then the system is determined to enter the review interval and the high-frequency sampling review procedure is initiated.

[0046] More preferably, the high-frequency sampling verification procedure includes: Multiple high-frequency samples are taken within a preset time period, and the mean early warning probability and the mean false alarm probability are recalculated based on the sampling results. If the average early warning probability is not lower than the third alarm threshold and the average false alarm probability is lower than the third false alarm threshold, then it is determined to be a real pyrolysis and an early warning signal is triggered; otherwise, it is determined to be that there is no pyrolysis hazard or environmental interference, and the corresponding environmental parameter data and probability data are recorded for iterative optimization of the neural network model.

[0047] Before the aforementioned early warning method is put into application, the BP neural network model needs to be trained, specifically including: Multiple sets of historical data (including the seven environmental parameters) of different electrical materials during the simulated pyrolysis process were collected. After preprocessing and analysis, the data were divided into training and validation sets in a ratio of approximately 4:1 for training and optimization of the model. The trained and converged model is deployed in the intelligent analysis module for processing real-time monitoring data; The intelligent analysis module generates corresponding control commands based on the judgment results output by the model (such as the probability of early warning and the probability of false alarm), which drive the early warning execution modules such as buzzers and alarm lights to issue alarms.

[0048] This invention aims to overcome several key deficiencies in existing electrical fire monitoring technologies, specifically including: (1) Traditional methods are limited by voltage levels, have poor adaptability, and alarms mostly occur after electrical ignition, resulting in delayed early warning and inability to effectively intervene in the early stages of fire hazards. (2) Most detectors rely on a single parameter, which makes it difficult to comprehensively reflect the true state of hidden dangers. They are easily affected by environmental interference and equipment aging, resulting in frequent false alarms and missed alarms, and insufficient reliability. (3) The installation of contact or fixed-point temperature measurement equipment is limited, there are monitoring blind spots, it is difficult to fully cover potential temperature abnormality areas, and it is impossible to detect hidden thermal hazards in a timely manner; (4) Some existing multi-sensor fusion schemes lack dynamic weight adjustment mechanisms and adopt fixed weight algorithms, which cannot adapt to complex and ever-changing actual working conditions, resulting in decreased monitoring accuracy and difficulty in accurately assessing the degree of deterioration and risk level of insulation materials.

[0049] The technical effects of this invention include at least the following: Collaborative False Alarm Suppression Based on Backpropagation Neural Network and Dynamic Weights: The false alarm probability is calculated in real time using a neural network, and a dynamic weighting mechanism is integrated to adaptively adjust the importance of various environmental parameters, forming a dual false alarm shielding logic. This collaborative mechanism can effectively distinguish between real pyrolysis signals and common environmental interferences such as water mist and smoke. Experimental verification shows that it can reduce the system's false alarm rate to below 5%, improving the reliability of early warning systems.

[0050] Non-contact multi-sensor collaborative monitoring and early warning capabilities: By integrating temperature, gas, and multi-size particulate matter sensors and employing non-contact deployment, multi-dimensional, blind-spot-free monitoring of the environment within the distribution cabinet is achieved. This design can comprehensively capture trace gases and characteristic particles released during the initial thermal decomposition of insulating materials. Experiments show that when the temperature of the insulating material rises to 90℃ (far below its ignition point), the system can accurately detect pyrolysis gases (such as m-xylene, ethylbenzene, and other VOCs) and trace particulate matter, achieving an alarm time more than 30 minutes earlier than traditional smoke and heat detectors, thus providing ample time for hazard mitigation.

[0051] As an optional or auxiliary decision-making method, the system can directly perform cross-validation and false alarm identification on the monitoring signal by analyzing the following three types of quantifiable features: Particle Concentration Time-Series Characteristic Analysis: The system analyzes the change patterns of pyrolysis particle concentrations (especially PM2.5 and PM10) in real time. If the concentration increases by more than a certain percentage (e.g., 50%) within a very short time (e.g., 10 seconds) and then rapidly decreases within a short period (e.g., 1 minute), the signal is determined to conform to the typical characteristics of transient interference (such as water mist or dust). Conversely, if the particle concentration exhibits a step-like, continuous growth pattern, such as a cumulative increase exceeding a certain threshold (e.g., 10%) within several minutes (e.g., 5 minutes) without significant decrease, the signal is determined to conform to the release characteristics of pyrolysis of insulating materials.

[0052] Gas-Particle Concentration Correlation Analysis: The system calculates the correlation between changes in pyrolysis particle concentration and VOC gas concentration generated by thermal decomposition. If the two show a strong positive correlation, such as a correlation coefficient reaching or exceeding a high correlation threshold (e.g., ≥0.6), it is determined to be a synergistic release characteristic of the pyrolysis process, supporting the judgment of potential pyrolysis hazards. If only the particle concentration is abnormal, while the gas concentration shows no significant change and the correlation is very low (e.g., correlation coefficient <0.3), it is more likely to originate from non-pyrolysis interference such as water mist or dust.

[0053] Temperature Co-location Analysis: The system simultaneously monitors the ambient temperature and its rate of change. Real pyrolysis processes are typically accompanied by significant thermal effects. Therefore, when particle or gas concentrations are abnormal, if the temperature simultaneously reaches or exceeds a high-temperature threshold (e.g., ≥70℃), or the rate of temperature change exceeds a rapid temperature rise threshold (e.g., ≥0.5℃ / min), this constitutes strong evidence supporting real pyrolysis. Conversely, if during periods of abnormal particle concentration, the temperature remains below a safe threshold (e.g., <60℃) and the temperature rise is extremely slow (e.g., rate of change <0.2℃ / min), it is more likely to be determined as environmental disturbance without a heat source.

[0054] By comprehensively comparing the characteristics of one or more of the above dimensions and their preset judgment thresholds, the system can directly assist in verifying or independently determine the authenticity of an alarm event, thereby effectively distinguishing between real pyrolysis hazards and common environmental interferences.

[0055] The technical solution described in this invention can be modified or expanded through the following alternative implementation methods to adapt to different application needs and scenario constraints: Replacement of neural network models The original typical solution uses a 3-layer BP neural network, configured with 7 neurons in the input layer, 12 neurons in the hidden layer, and 2 neurons in the output layer. The hidden layer uses the ReLU activation function, and the output layer uses the Softmax function to calculate the alarm probability and false alarm probability.

[0056] An alternative solution is to use an LSTM (Long Short-Term Memory) model instead of the BP neural network. LSTM models excel at capturing the dynamic changes in time-series data and have superior trend analysis capabilities for parameters such as temperature change rate and gas concentration change rate that fluctuate over time. In the early stages of electrical equipment failure (such as the slow degradation process of insulation materials), this alternative can improve the early warning accuracy by approximately 5%-8%. However, its algorithm complexity is higher, requiring increased computing power from the main control chip, potentially increasing hardware costs by about 25%, and requiring a longer model training time. This solution is suitable for critical power distribution facilities (such as substation high-voltage switchgear) that have extremely high requirements for early warning accuracy.

[0057] Replacement of dynamic weight adjustment mechanism The original typical solution: Based on operating parameters (such as temperature and humidity), a fixed adjustment ratio is set (for example, increasing the weight of temperature change rate by 20% in high-temperature environments) to achieve dynamic weight adjustment of parameters.

[0058] An alternative solution is to use fuzzy control algorithms to replace the fixed-ratio dynamic weight adjustment mechanism. Fuzzy control does not require a precise mathematical model; by defining semantic fuzzy rules, it can adapt to different complex operating conditions and reduce the subjective error of manually setting weights. However, the formulation of its fuzzy rules relies on the accumulation of a large amount of experimental data, and in scenarios where the operating condition rules are unclear, the adjustment accuracy may be slightly lower than the original solution. This alternative solution is suitable for application scenarios with complex and diverse operating conditions and sufficient accumulated experimental data.

[0059] The original typical solution: The core is to achieve very early warning and false alarm suppression through the collaboration of "multi-sensor (temperature, gas, particulate matter) acquisition + BP neural network + dynamic weight". The key is to integrate three types of sensors to achieve non-contact full-area coverage.

[0060] An alternative solution is a combination of an infrared thermal imager and a gas sensor. The infrared thermal imager provides real-time imaging of the temperature field within the distribution cabinet, allowing for direct location of high-temperature points, while the gas sensor detects pyrolysis gases. This solution reduces the number of sensors and provides wider temperature monitoring coverage. However, infrared thermal imagers are more expensive, and in scenarios with many obstructions (such as dense cables) within the cabinet, blind spots can easily appear in the temperature imaging. Therefore, this alternative solution is suitable for scenarios with ample cabinet space, sufficient budget, and a need for visual temperature monitoring. While it achieves the same early warning effect, it has limitations in terms of cost and application scenarios.

[0061] In Example 2, this invention provides an early warning device for superheated pyrolysis electrical fires, used to implement the early warning method for superheated pyrolysis electrical fires described in Example 1, comprising: The monitoring module, communication module, and power supply module are connected via electrical connections or wireless communication protocols. The monitoring module is used to acquire multi-dimensional environmental parameters inside the power distribution facility in real time, including temperature, temperature change rate, thermal decomposition gas concentration, gas concentration change rate, and pyrolysis particle concentration with different particle size characteristics. These multi-dimensional environmental parameters are then input into its built-in pre-trained neural network model to obtain the early warning probability reflecting actual pyrolysis hazards and the false alarm probability reflecting environmental interference. Based on the early warning and false alarm probabilities, the weights of each environmental parameter are dynamically updated. The weighted and updated environmental parameters are then input into the neural network model again to obtain updated early warning and false alarm probabilities. Based on these updated probabilities, a pre-set multi-level judgment rule is invoked for comprehensive judgment to distinguish between actual pyrolysis and interference factors, and an early warning command is generated when actual pyrolysis is determined. The communication module is connected to the monitoring module and is used to upload the early warning instructions, monitoring data and alarm information generated by the monitoring module to the remote monitoring platform; The power module is connected to the monitoring module and the communication module to provide operating power for the entire device.

[0062] Specifically, the monitoring module performs filtering (using Kalman filtering to eliminate noise) and normalization (mapping parameter values ​​to the 0~1 range) on the collected raw data; Data processing and decision-making are performed by an embedded neural network model, which employs a three-layer feedforward structure: The input layer has 7 neurons, corresponding to the 7 parameters mentioned above, to ensure full-dimensional feature input; The hidden layer has 12 neurons. Comparative experiments have verified that this size can achieve a balance between feature extraction depth and embedded hardware computing power, effectively avoiding overfitting or underfitting. The output layer has two neurons, corresponding to the two decision outcomes: "normal alarm" and "false alarm." The hidden layer uses the ReLU activation function (Y=ReLU(p...). W1+b1), the output layer uses the Softmax function (O=Softmax(Y). W2+b2), the final output vector is O=[O1,O2], where O1 represents the probability of a normal alarm and O2 represents the probability of a false alarm.

[0063] For example, the communication module adopts wired (RS485 bus, communication rate 9600bps) + wireless (LoRa module, frequency band 33MHz, transmission distance ≤5km) dual-mode communication to upload monitoring data and alarm information to the distribution network monitoring platform, and at the same time receive parameter configuration instructions issued by the platform; The power module adopts a wide voltage input (AC220V / DC12V~48V), which is converted to DC5V / 3.3V by a switching power supply module to power each module. It has a built-in lithium battery (capacity 5000mAh) to achieve ≥4 hours of battery life after power failure.

[0064] It should be noted that, as a key design feature of the monitoring solution, this invention employs a non-contact, full-area coverage monitoring method. Specifically, the monitoring module integrates temperature, gas, and multi-size particulate matter sensors. These sensors do not require direct physical contact with electrical equipment within the power distribution facility, nor do they need to be installed at fixed locations. Instead, they collect environmental parameters within the cabinet through the air medium. The detectors are deployed within the cabinet space, and by monitoring heat diffusion and airflow distribution, they can achieve coverage of the entire potential pyrolysis area within the cabinet, thereby effectively eliminating monitoring blind spots and improving the comprehensiveness and reliability of hazard detection.

[0065] Embodiment 3 of the present invention provides an electronic device, including a memory, a processor, and a computer program stored in the memory and executable on the processor. When the computer program is loaded onto the processor, it implements an early monitoring and warning method for superheated pyrolysis electrical fires as described in Embodiment 1.

[0066] Embodiment 4 of the present invention provides a computer-readable storage medium storing a computer program, which, when executed by a processor, implements a method for early monitoring and warning of superheated pyrolysis electrical fires as described in Embodiment 1.

[0067] (a) Neural Network Model Replacement Original solution: Use a 3-layer BP neural network (7 input, 12 hidden, 2 output neurons), with ReLU activation function in the hidden layer and Softmax function in the output layer to calculate alarm probability and false alarm probability.

[0068] Alternative solution: Employ an LSTM (Long Short-Term Memory) model. LSTM effectively captures the dynamic changes in time-series data, exhibiting better trend analysis capabilities for parameters such as temperature change rate and gas concentration change rate that fluctuate over time. In the early stages of electrical equipment failures (such as the slow degradation of insulation materials), the early warning accuracy can be improved by 5%-8%. However, the algorithm is more complex, requiring increased computing power from the main control chip, increasing hardware costs by approximately 25%, and requiring longer model training time. It is suitable for critical power distribution facilities with extremely high early warning accuracy requirements (such as high-voltage switchgear in substations).

[0069] (ii) Dynamic weight adjustment mechanism as an alternative Original solution: Set a fixed adjustment ratio based on operating conditions (such as temperature and humidity) (e.g., increase the weight of temperature change rate by 20% in high-temperature environments) to achieve dynamic adjustment of parameter weights.

[0070] Alternative solution: Employ fuzzy control algorithms. This eliminates the need for precise mathematical models; by defining fuzzy rules, it can adapt to various complex operating conditions, reducing the subjective errors of manually setting weights. However, the formulation of fuzzy rules relies on the accumulation of a large amount of experimental data, and in scenarios where the operating condition rules are unclear, the adjustment accuracy is slightly lower than the original solution. It is suitable for applications with complex operating conditions and sufficient experimental data.

[0071] (III) Complete technical solution replacement Original solution: Achieve very early warning and false alarm suppression through "multi-sensor (temperature, gas, particulate matter) acquisition + BP neural network + dynamic weighting". The core is to integrate three types of sensors to achieve full area coverage.

[0072] Alternative Solution: A combination of an infrared thermal imager and a gas sensor. The infrared thermal imager provides real-time imaging of the temperature field inside the distribution cabinet, allowing for direct location of high-temperature points. Combined with a gas sensor to detect pyrolysis gases, this reduces the number of sensors required. This solution offers wider temperature monitoring coverage, but infrared thermal imagers are more expensive, and blind spots can easily appear in scenarios with significant obstructions inside the cabinet (such as dense cabling). It is suitable for scenarios with ample cabinet space, sufficient budget, and a need for visual temperature monitoring, achieving the same early warning effect as the original solution, but with limitations in cost and application scenarios.

[0073] This disclosure can be a system, method, and / or computer program product. A computer program product may include a computer-readable storage medium having computer-readable program instructions loaded thereon for causing a processor to implement various aspects of this disclosure.

[0074] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and not to limit it. Although the present invention has been described in detail with reference to the above embodiments, those skilled in the art should understand that modifications or equivalent substitutions can still be made to the specific implementation of the present invention. Any modifications or equivalent substitutions that do not depart from the spirit and scope of the present invention should be covered within the protection scope of the claims of the present invention.

Claims

1. A method for monitoring and early warning of electrical fire in the super-temperature pyrolysis stage, characterized in that, The method comprises the following steps: Real-time acquisition of multi-dimensional environmental parameters inside the power distribution facility, including temperature, temperature change rate, thermal decomposition gas concentration, thermal decomposition gas concentration change rate, and pyrolysis particle concentration of different particle size characteristics; Inputting the multi-dimensional environmental parameters into a pre-trained neural network model to output an early warning probability reflecting the real pyrolysis hazard and a false alarm probability reflecting the environmental interference; Based on the early warning probability and the false alarm probability, dynamically updating the weights of each environmental parameter, and inputting the environmental parameters with updated weights into the neural network model again to obtain updated early warning probability and false alarm probability; Based on the updated early warning probability and false alarm probability, combined with the preset multi-level judgment rule for comprehensive judgment to distinguish real pyrolysis and interference factors, and outputting an early warning signal when real pyrolysis is determined.

2. The method according to claim 1, wherein: The neural network model is a BP neural network comprising an input layer, a hidden layer, and an output layer; The number of input layer nodes is consistent with the number of multi-dimensional environmental parameters; The output layer is configured with two nodes to output early warning probability and false alarm probability respectively.

3. The method according to claim 1, wherein: The dynamic weight adjustment actually includes periodic updating and trigger updating, wherein: Trigger updating is started when the real-time value of any environmental parameter fluctuates more than a set proportion of its base value; The dynamic weight adjustment rule is: The dynamic weight value of each environmental parameter is equal to its preset base weight value multiplied by an adjustment factor; The adjustment factor includes the product of false alarm probability and a first coefficient factor, and the product of a working condition correction coefficient and a second coefficient factor.

4. The method according to claim 3, wherein: The preset base weight is set according to the environmental parameter category, including: Setting a first base weight value for temperature and temperature change rate; Setting a second base weight value for thermal decomposition gas concentration and thermal decomposition gas concentration change rate; Setting a third base weight value for pyrolysis particle concentration of different particle size characteristics; The third base weight value is greater than the second base weight value, and the second base weight value is greater than the first base weight value.

5. The method according to claim 4, wherein: The working condition correction coefficient is determined according to the environmental parameter category and the current working condition, and the determination logic includes: For temperature and temperature change rate, if the current temperature is not lower than the preset temperature threshold, the working condition correction coefficient value is adjusted to the first working condition correction coefficient value, otherwise the working condition correction coefficient remains 0; For thermal decomposition gas concentration and thermal decomposition gas concentration change rate, the current humidity inside the power distribution facility is obtained, and if the current humidity is not lower than the preset humidity threshold, the working condition correction coefficient value is adjusted to the second working condition correction coefficient value, otherwise the working condition correction coefficient value remains 0; For pyrolysis particle concentration of different particle size characteristics, the working condition correction coefficient is determined according to the current humidity and the false alarm probability: If the current humidity is not lower than the preset humidity threshold and the false alarm probability is not lower than the preset false alarm threshold, the working condition correction coefficient value is the sum of the second working condition correction coefficient value and the third working condition correction coefficient value; If the current humidity is not lower than the preset humidity threshold and the false alarm probability is lower than the preset false alarm threshold, the working condition correction coefficient value is the second working condition correction coefficient value; If the current humidity is lower than the preset humidity threshold and the false alarm probability is not lower than the preset false alarm threshold, the working condition correction coefficient value is the third working condition correction coefficient value; If the current humidity is lower than the preset humidity threshold and the false alarm probability is lower than the preset false alarm threshold, the working condition correction coefficient value remains 0; Wherein, the first working condition correction coefficient value is greater than 0, and the second working condition correction coefficient value and the third working condition correction coefficient value are less than 0.

6. The super-temperature pyrolysis electrical fire extremely early monitoring and warning method according to claim 1, characterized in that: comprehensive determination in combination with the preset multi-level determination rule comprises: If the updated warning probability is not lower than the first alarm threshold, and the updated false alarm probability is lower than the first false alarm threshold, it is determined that there is real pyrolysis and a warning signal is triggered; If the updated warning probability is lower than the second alarm threshold, or the updated false alarm probability is not lower than the second false alarm threshold, it is determined that there is no pyrolysis risk or environmental interference, and no warning signal is triggered; If the updated warning probability and the updated false alarm probability do not meet the above two conditions, it is determined that it enters the review interval, and a high-frequency sampling review program is started.

7. The super-temperature pyrolysis electrical fire extremely early monitoring and warning method according to claim 6, characterized in that: the high-frequency sampling review program comprises: multiple high-frequency samplings are performed within a preset time period, and the average warning probability and the average false alarm probability are recalculated based on the sampling results; If the average warning probability is not lower than the third alarm threshold, and the average false alarm probability is lower than the third false alarm threshold, it is determined that there is real pyrolysis and a warning signal is triggered; otherwise, it is determined that there is no pyrolysis risk or environmental interference, and the corresponding environmental parameter data and probability data are recorded for iterative optimization of the neural network model.

8. An apparatus for monitoring and warning of an electrical fire in the very early stage of super-temperature pyrolysis, for implementing the method of monitoring and warning of an electrical fire in the very early stage of super-temperature pyrolysis according to any one of claims 1-7, characterized in that, comprises: a monitoring module, a communication module and a power supply module, each module being connected through electrical connection or wireless communication protocol; The monitoring module is used to acquire multi-dimensional environmental parameters inside the power distribution facility in real time, including temperature, temperature change rate, heat-decomposed gas concentration, gas concentration change rate and pyrolysis particle concentration of different particle sizes; and input the multi-dimensional environmental parameters into the pre-trained neural network model built-in the monitoring module to obtain a warning probability reflecting real pyrolysis risk and a false alarm probability reflecting environmental interference; based on the warning probability and the false alarm probability, the weights of each environmental parameter are dynamically updated, and the environmental parameters after weight updating are input into the neural network model again to obtain updated warning probability and false alarm probability; Based on the updated warning probability and the false alarm probability, the preset multi-level determination rule is called for comprehensive determination to distinguish real pyrolysis from interference factors, and a warning instruction is generated when real pyrolysis is determined; The communication module is connected with the monitoring module and is used to upload the warning instruction, monitoring data and alarm information generated by the monitoring module to the remote monitoring platform; The power module is connected with the monitoring module and the communication module, and is used for providing working power for the whole device.

9. An electronic device comprising a memory, a processor, and a computer program stored on the memory and executable on the processor, characterized in that, The computer program, when loaded into the processor, implements the super-temperature pyrolysis electrical fire extremely early monitoring and warning method according to any one of claims 1-7.

10. A computer-readable storage medium storing a computer program, the computer program comprising instructions that, when executed by a computer, cause the computer to perform the method of any one of claims 1 to 9. The computer program, when executed by the processor, implements the super-temperature pyrolysis electrical fire extremely early monitoring and warning method according to any one of claims 1-7.

Images