Intelligent fire-fighting electrical fire monitoring method, device, equipment and medium

Abstract

The invention relates to the technical field of fire safety, in particular to an intelligent fire-fighting electrical fire monitoring method, device and equipment and a medium, and the method comprises the steps: obtaining electrical parameters; according to the electrical parameters, a fire hazard identification model is established by using an LSTM algorithm and a CNN algorithm, the operation state of the electrical equipment is predicted according to the fire hazard identification model, an operation state trend prediction result is obtained, and a potential fire hazard type is identified; according to a fire hazard type identification result, in combination with a use environment of electrical equipment and a threshold determination algorithm, performing quantitative evaluation on a fire hazard risk to obtain a corresponding risk level, and when the risk level exceeds a set multi-layer threshold, automatically starting a linkage control mechanism; therefore, the problems of response lagging, high labor cost, high intermittent fault omission ratio and the like in the prior art are solved.

Description

Technical Field

[0001] The present application relates to the field of fire safety technology, and in particular to a smart fire electrical fire monitoring method, device, equipment and medium. Background Art

[0002] With the acceleration of urbanization, the number of electrical devices in high-rise buildings, commercial complexes, and densely populated residential areas has exploded. This has led to frequent hidden dangers such as aging lines, overheating due to overloads, and poor contact, posing a significant challenge to the safety of human life and property, as well as the stable development of the social economy. To effectively address this challenge, the research and application of intelligent fire protection electrical fire monitoring methods is particularly important.

[0003] However, traditional monitoring methods rely on threshold alarms and manual inspections of single-point sensors, and their drawbacks are prominent in modern electrical monitoring systems. Single-parameter monitoring can only make isolated judgments and cannot build a coupled disaster-causing model of multiple physical quantities. Not only are missed reports frequent, but they are also prone to false alarms due to rigid threshold settings. Manual inspections are limited by time and space coverage, and have a high missed detection rate for intermittent faults, which are completely impossible to capture. In addition, the labor cost is high and the response is delayed, making it difficult to adapt to the dynamic operation needs of smart grids. These defects have led to serious deficiencies in the warning capabilities of traditional solutions in complex electrical scenarios, and they are unable to support real-time, full-area, and multi-dimensional risk prevention and control. This makes it difficult for traditional monitoring methods to respond quickly and effectively when faced with sudden fire disasters, thus missing the best opportunity for prevention and disaster reduction. Summary of the Invention

[0004] The present application provides a smart fire protection electrical fire monitoring method, device, equipment and medium to solve the problems of high labor costs, high intermittent fault missed detection rate, and serious lack of early warning capabilities in related technologies.

[0005] The first embodiment of the present application provides a smart fire protection electrical fire monitoring method, comprising the following steps: obtaining electrical parameters; establishing a fire hazard identification model based on the electrical parameters using the LSTM algorithm and the CNN algorithm, predicting the operating status of the electrical equipment based on the fire hazard identification model, obtaining an operating status trend prediction result, and at the same time, identifying the potential fire hazard type; based on the fire hazard type identification result, combined with the use environment of the electrical equipment and the threshold judgment algorithm, quantitatively evaluating the fire risk to obtain the corresponding risk level; when the risk level exceeds the set multi-layer threshold, automatically starting the linkage control mechanism, wherein, when the risk level is low and exceeds the first-level threshold, a local early warning is triggered and the light flashes; when the risk level is intermediate and exceeds the second-level threshold, a remote alarm is triggered and a text message is pushed to the mobile phone; when the risk level is high and exceeds the third-level threshold, the fire extinguishing device is triggered and the power is automatically cut off.

[0006] Optionally, according to the electrical parameters, a fire hazard identification model is established using the LSTM algorithm and the CNN algorithm, including: constructing a training data set; based on the training data set, constructing a single-device anomaly detection and regional risk association model using the LSTM algorithm and the CNN algorithm; based on the single-device anomaly detection and regional risk association model, using a data labeling method, using the SMOTE algorithm to perform sample balancing, and training the model after five-fold cross-validation and confusion matrix evaluation; based on the trained model, using a multimodal data fusion algorithm, combined with edge-cloud collaborative deployment and continuous learning mechanism for optimization, to form a fire hazard identification model.

[0007] Optionally, the LSTM algorithm formula is:

[0008] in, Indicates the long-term and short-term timing dependency information of the electrical parameters stored at the current moment. Represents historical temporal memory, For the Gate of Forgetfulness, is the input gate, is the hyperbolic tangent activation function, For input to the weight matrix of the candidate state space, For input to the candidate state space to the weight matrix, is the temporal abstract feature after processing at the previous moment, The electrical parameter timing data collected at this moment, is an independent bias term.

[0009] Optionally, the CNN algorithm is:

[0010] in, The extracted electrical fire related features, To input the spatiotemporal data of electrical monitoring, is the row position of the output feature map, is the column position of the output feature map, is the offset of the convolution kernel in the row direction, is the offset of the convolution kernel in the column direction, Indicates that the convolution kernel is at position The weight value at , M is the number of rows of the convolution kernel, N is the number of columns of the convolution kernel, and b is the bias term.

[0011] Optionally, the operating status of the electrical equipment is predicted according to the fire hazard identification model to obtain an operating status trend prediction result, including: obtaining the operating status data of the electrical equipment; based on the fire hazard identification model, before training and prediction, performing Z-score normalization processing on the operating status data of the electrical equipment, using the early stopping method to prevent overfitting during training, and at the same time, using data enhancement to expand the sample optimization parameters, outputting the predicted value and change of the fire hazard point, and generating an equipment trend label according to the risk change rate; through threshold comparison, historical pattern matching and risk classification mapping, converting the equipment trend label into a visual risk classification signal, performing a comprehensive judgment on the risk classification signal, and generating a development trend prediction result.

[0012] Optionally, based on the fire hazard type identification result, combined with the use environment of the electrical equipment and the threshold judgment algorithm, a quantitative assessment of the fire risk is performed, including: obtaining fire hazard type data; based on the fire hazard type data, combined with the hazard disaster-causing characteristics, historical data and environmental parameters, determining the basic risk weight of the hazard type through the hierarchical analysis method; based on the basic risk weight of the hazard type, using a nonlinear mapping function to characterize the nonlinear coupling relationship between the basic risk of the hazard and the ambient temperature correction factor, and constructing a nonlinear risk model; according to the nonlinear risk model, calculating the risk index in real time, and through the threshold judgment algorithm, classifying the risk index into levels, and performing a quantitative assessment and level judgment on the fire risk.

[0013] Optionally, the risk quantification assessment formula is:

[0014] Among them, R is the comprehensive risk value, is the number of electrical parameters, is the total number of electrical parameters, For the The risk factor function of parameters, For the The weight of each electrical parameter, T is the trend persistence coefficient, and C is the confidence coefficient.

[0015] The second aspect of the present application provides an intelligent fire protection electrical fire monitoring device, including: an acquisition module for acquiring electrical parameters; a construction module for establishing a fire hazard identification model using an LSTM algorithm and a CNN algorithm; an identification module for predicting the operating status of electrical equipment based on the fire hazard identification model, obtaining an operating status trend prediction result, and at the same time, identifying potential fire hazard types; an evaluation module for quantitatively evaluating the fire risk based on the fire hazard type identification result, combined with the use environment of the electrical equipment and the threshold judgment algorithm, to obtain a corresponding risk level; a feedback module for automatically starting a linkage control mechanism when the risk level exceeds a set multi-layer threshold, wherein when the risk level is low and exceeds the first-level threshold, a local early warning is triggered and the light flashes; when the risk level is intermediate and exceeds the second-level threshold, a remote alarm is triggered and a text message is pushed to a mobile phone; when the risk level is high and exceeds the third-level threshold, a fire extinguishing device is triggered and the power is automatically cut off.

[0016] The third aspect of the present application provides an electronic device, comprising: a memory, a processor, and a computer program stored on the memory and executable on the processor, wherein the processor executes the program to execute the intelligent fire protection electrical fire monitoring method as described in the above embodiment.

[0017] The fourth aspect of the present application provides a computer-readable storage medium on which a computer program is stored. The program is executed by a processor to perform the intelligent fire protection electrical fire monitoring method as described in the above embodiment.

[0018] Therefore, this application has at least the following beneficial effects: The embodiment of the present application uses the LSTM algorithm to capture the time-series dependency features of electrical parameters and the CNN algorithm to extract spatial correlation and frequency domain anomaly features, constructs a multimodal fire hazard identification model, and combines real-time data preprocessing on the edge with a continuous learning mechanism on the cloud to solve the problems of weak intermittent fault feature extraction and delayed warning in traditional solutions; uses SMOTE sample balancing, five-fold cross-validation and other technologies to improve the model's recognition accuracy for minority hidden dangers, and cooperates with Z-score normalization and early stopping method to prevent overfitting and enhance early warning capabilities in complex scenarios; constructs a nonlinear risk model through the hierarchical analysis method, dynamically integrates environmental parameters to correct hidden danger weights, and combines a three-level threshold hierarchical linkage mechanism to automate the entire process from risk identification to disposal, avoiding response delays caused by manual intervention, reducing labor costs, and effectively solving the core problems of high labor dependence, high intermittent fault missed detection rate and insufficient warning capabilities in related technologies.

[0019] Additional aspects and advantages of the present application will be given in part in the description below, and in part will become apparent from the description below, or will be learned through practice of the present application. BRIEF DESCRIPTION OF THE DRAWINGS

[0020] The above and / or additional aspects and advantages of the present application will become apparent and easily understood from the following description of the embodiments in conjunction with the accompanying drawings, in which: Figure 1 This is a flow chart of a smart fire protection electrical fire monitoring method provided according to an embodiment of the present application; Figure 2 This is an example diagram of electrical fire monitoring in a large-scale smart park provided according to one embodiment of the present application; Figure 3 This is an example diagram of electrical fire monitoring in a smart manufacturing demonstration park provided according to one embodiment of the present application; Figure 4 This is an example diagram of a practical project for smart fire fighting on electrical fires provided according to one embodiment of the present application; Figure 5 This is an example diagram of electrical fire monitoring in a high-rise office building provided according to one embodiment of the present application; Figure 6 This is a flow chart of a smart fire protection electrical fire monitoring method provided according to one embodiment of the present application; Figure 7 This is a block diagram of an example of a smart fire protection electrical fire monitoring device provided according to an embodiment of the present application; Figure 8 A schematic diagram of the structure of an electronic device provided according to an embodiment of the present application. DETAILED DESCRIPTION

[0021] The following describes in detail embodiments of the present application. Examples of the embodiments are shown in the accompanying drawings, wherein the same or similar reference numerals throughout represent the same or similar elements or elements having the same or similar functions. The embodiments described below with reference to the accompanying drawings are exemplary and are intended to be used to explain the present application, and should not be construed as limiting the present application.

[0022] The following describes a method, device, equipment, and medium for monitoring electrical fires in an intelligent fire protection system according to an embodiment of the present application with reference to the accompanying drawings. In response to the high intermittent fault missed detection rate problem mentioned in the above background technology, the present application provides a method for monitoring electrical fires in an intelligent fire protection system. In this method, an LSTM algorithm is used to capture the temporal dependency features of electrical parameters, a CNN algorithm is used to extract spatial correlation and frequency domain anomaly features, and a multimodal fire hazard identification model is constructed. This method combines edge-side real-time data preprocessing with a cloud-based continuous learning mechanism to solve the problems of weak intermittent fault feature extraction and delayed early warning in traditional solutions. SMOTE sample balancing and five-fold cross-validation techniques are used to improve the model's recognition accuracy for a small number of hidden dangers. Z-score normalization and early stopping methods are used to prevent overfitting, thereby enhancing early warning capabilities in complex scenarios. A nonlinear risk model is constructed using a hierarchical analysis method, and environmental parameters are dynamically integrated to correct the hidden danger weights. Combined with a three-level threshold hierarchical linkage mechanism, the entire process from risk identification to disposal is automated, avoiding response delays caused by manual intervention and reducing labor costs. This effectively solves the core problems of high human dependence, high intermittent fault missed detection rate, and insufficient early warning capabilities in related technologies.

[0023] The following describes an intelligent fire protection electrical fire monitoring method, device, equipment and medium of an embodiment of the present application with reference to the accompanying drawings.

[0024] Specifically, Figure 1 A flow chart of a smart fire protection electrical fire monitoring method provided in an embodiment of the present application.

[0025] like Figure 1 As shown, the smart fire protection electrical fire monitoring method includes the following steps: In step S101 , electrical parameters are acquired.

[0026] Among them, electrical parameters refer to the real-time collection of key data related to electrical operating status through the deployment of intelligent sensing equipment, including current, voltage, temperature, and residual current.

[0027] It can be understood that the embodiments of the present application build a perception network by collecting key data such as current, voltage, temperature, and residual current in real time, providing a comprehensive and accurate basis for fire hazard identification model training and risk quantification assessment, thereby improving the accuracy and scientific nature of fire management.

[0028] In step S102, a fire hazard identification model is established based on electrical parameters using the LSTM algorithm and the CNN algorithm. The operating status of the electrical equipment is predicted based on the fire hazard identification model to obtain an operating status trend prediction result. At the same time, the potential fire hazard type is identified. Among them, the fire hazard identification model is a model that analyzes multi-dimensional data such as temperature, smoke concentration, electrical equipment status, and flammable material distribution in the environment, and combines algorithms to intelligently identify potential fire risks and issue warnings.

[0029] It is understandable that the embodiments of the present application utilize LSTM algorithms and CNN algorithms to establish a fire hazard identification model based on the electrical parameters of electrical equipment, such as current, voltage, and temperature, to predict the performance degradation trend of the equipment in advance. By learning from historical data and continuously optimizing the model, it accurately distinguishes normal fluctuations from abnormal changes, reducing the false alarm and missed alarm rate. At the same time, based on massive data mining of characteristic patterns of different types of hidden dangers, it quickly locates specific hidden dangers such as short circuits, overloads, and insulation aging, providing decision-making data for fire management, improving the foresight, accuracy, and intelligence level of fire risk prevention and control, and reducing the risk of property loss and casualties.

[0030] For example, in a smart fire protection pilot community, Mr. Li installed a comprehensive smart fire protection electrical fire monitoring platform in his home. Using intelligent sensors, the platform provides 24-hour, uninterrupted monitoring of key electrical parameters in his home, including current, voltage, line temperature, and residual current. Two months after installation, the platform's monitoring data revealed abnormal current fluctuations in the circuit of a socket in Mr. Li's kitchen. The circuit temperature rapidly climbed from the normal range of 30°C to 55°C, far exceeding the device's preset safety threshold of 45°C. Simultaneously, the residual current soared from a few milliamperes to 50 milliamperes. Based on its built-in algorithm, the platform identified this as a serious electrical fire hazard and immediately sent an alert to Mr. Li via its mobile app and a text message. Mr. Li quickly contacted a professional electrician for investigation, who discovered that the socket's internal wiring was aging and damaged, causing a local short circuit and overheating. Prompt wiring replacement eliminated the hazard. The platform has been installed in 500 households in the community. Within six months of installation, it successfully identified and issued warnings for 120 electrical fire hazards, effectively preventing household electrical fires and protecting residents' lives and property.

[0031] In an embodiment of the present application, a fire hazard identification model is established based on electrical parameters using the LSTM algorithm and the CNN algorithm, including: constructing a training data set; based on the training data set, constructing a single-device anomaly detection and regional risk association model using the LSTM algorithm and the CNN algorithm; based on the single-device anomaly detection and regional risk association model, using a data labeling method, using the SMOTE algorithm for sample balancing, and training the model after five-fold cross-validation and confusion matrix evaluation; based on the trained model, using a multimodal data fusion algorithm, combined with edge-cloud collaborative deployment and continuous learning mechanism for optimization, to form a fire hazard identification model.

[0032] Among them, the multimodal data fusion algorithm is an algorithm that integrates the features or decision results of different modal data such as text, images, and audio, and uses their complementarity and correlation to build a unified semantic representation or comprehensive decision to improve the performance of the model in complex tasks. The formula is: ,

[0033] in, For electrical characteristics The weight coefficient in the fusion, is the activation function, is the weight matrix, Splicing and The feature concatenation function, is the time series characteristic vector of electrical parameters, Semantic feature vector of visual image, is the bias vector, is the comprehensive feature vector after fusion, For visual features Weight coefficients in the fusion.

[0034] It can be understood that the embodiment of the present application uses a multimodal data fusion algorithm to integrate multiple types of data such as electrical parameters, environmental data, and visual data to conduct a comprehensive analysis of the data, breaking through the limitations of single data monitoring, capturing various abnormal signs before a fire occurs, providing early warning of electrical fire hazards, and reducing the incidence of electrical fires.

[0035] Among them, the SMOTE algorithm is a classic algorithm for solving the problem of class imbalance. The SMOTE algorithm formula is: , represents the synthesized virtual hidden danger sample, Represents a sample of historical fire hazards, express Neighborhood hidden danger samples, represents the interpolation weight.

[0036] It is understood that the embodiments of this application synthesize virtual samples that conform to the characteristic distribution of hidden dangers by performing linear interpolation between minority hidden danger samples and their nearest neighbors. This balances the category distribution of the training dataset, avoids the model's prediction bias due to the dominance of majority class data, improves the model's learning ability and recognition accuracy for minority hidden danger samples, and reduces the risk of missed detections. Simultaneously, by combining 5-fold cross-validation and confusion matrix evaluation, the model's generalization performance is optimized for unbalanced data, improving the sensitivity and robustness of single-device anomaly detection and regional risk association models for subtle fire hazards.

[0037] For example, Figure 2As shown, electrical fire hazards are common in large-scale smart campuses due to complex power usage environments and aging equipment. Traditional monitoring methods face data imbalance, making it difficult to accurately capture hazard characteristics. To address this, the campus implemented an electrical fire monitoring solution based on the SMOTE algorithm. Over 500 electrical parameter sensors were deployed throughout the campus, covering key locations such as power distribution rooms and production lines in each factory. Real-time data, such as current, voltage, and power, was collected every five seconds. Initial data showed a 20:1 ratio of normal operating data samples to potential fault data samples. This severely insufficient number of fault samples resulted in the initial model's prediction accuracy for electrical fire hazards being only 60%, leaving a significant risk of missed detection. To address this dilemma, the technical team applied the SMOTE algorithm to augment minority fault data samples. The algorithm first searches for neighboring samples in the feature space for each fault sample, determines an interpolation range based on the feature distance between samples, and then uses random interpolation to generate synthetic samples that match the true distribution characteristics. For example, the current data of an aging transformer in a factory exhibited abnormal fluctuations, which are typical of the minority class in the original data. The algorithm generates multiple new samples within the feature space of the sample and its neighboring samples, simulating the abnormal fluctuation characteristics, effectively increasing the diversity of the fault data. After processing with the SMOTE algorithm, the ratio of faulty samples to normal samples increased to 1:5, significantly improving data balance. Training a neural network model based on this balanced dataset significantly improved the model's ability to identify electrical fire hazards. In actual operational testing, the model issued accurate early warnings for three potential electrical fire hazards 45 minutes in advance, increasing its accuracy from 60% to 90%. In an incident where an abnormal increase in current was caused by aging equipment, the model accurately identified the hidden danger characteristics in the parameter fluctuations and issued a timely warning. Based on the warning information, staff quickly carried out maintenance and successfully avoided a potential fire accident. This practice fully demonstrates the key value of the SMOTE algorithm in smart fire protection electrical fire monitoring technology, building a more reliable line of defense for safe production and stable operations in the park.

[0038] In the embodiment of the present application, the LSTM algorithm formula is:

[0039] in, Indicates the long-term and short-term timing dependency information of the electrical parameters stored at the current moment. Represents historical temporal memory, For the Gate of Forgetfulness, is the input gate, is the hyperbolic tangent activation function, For input to the weight matrix of the candidate state space, For input to the candidate state space to the weight matrix, is the temporal abstract feature after processing at the previous moment, The electrical parameter timing data collected at this moment, is an independent bias term.

[0040] It can be understood that in the embodiment of the present application, through the synergistic effect of the forget gate, input gate, and output gate, LSTM can selectively retain long-term historical trends and capture short-term abnormal mutations, and conduct in-depth mining of multi-time scale hidden danger patterns. By storing long-term memory through cell state, encoding short-term features through hidden state, and fusing to form a hierarchical feature representation, it can distinguish normal fluctuations from fire precursors and process the temporal dependency of monitoring data; dynamically weight the current overload, voltage anomaly, and temperature surge parameters through the weight matrix to capture multi-parameter coupling risks and improve the ability to identify complex hidden dangers; through end-to-end training to adapt to different devices and scenarios, the weight matrix is ​​optimized to highlight fire-sensitive features and reduce the false positive rate and missed detection rate. Identify the progressive hidden dangers of equipment insulation aging and contact resistance increase, provide real-time and reliable risk assessment for smart fire protection, and improve the level of intelligent urban fire safety.

[0041] For example, Figure 3As shown in the figure, in a smart firefighting scenario at a smart manufacturing demonstration park, the LSTM algorithm achieves accurate early warning of electrical fire hazards through deep time series modeling. This deployment covers key equipment in the park's core power distribution rooms and 20 production lines. Using multiple sensors, including residual current transformers, thermocouple temperature sensors, and arc fault detectors, it collects eight core parameters, including current, voltage, temperature, and leakage current, in real time. The sampling rate reaches 200 times per second, generating a high-resolution time series data stream. Edge computing nodes enable feature extraction and primary diagnosis of local circuits within 5ms, minimizing data transmission delays. Taking a 380V motor group in a precision machining workshop in the park as an example, the parameter range during normal operation is 50-60A current, 375-385V voltage, and 40-50°C temperature. At 2:00 PM one afternoon, monitoring data showed that the current of a certain motor slowly increased from 55A to 78A within 30 minutes, the voltage simultaneously dropped to 360V, and the temperature exceeded 65°C. The LSTM model uses a 12-layer stacked bidirectional LSTM network combined with a 512-dimensional latent state space to capture abnormal characteristics such as long-term trends, short-term mutations, and coupling risks. The model was trained based on 200,000 pieces of historical data. The SMOTE algorithm was used to synthesize scarce initial overload samples, optimizing the ratio of positive and negative samples in the training set from 1:15 to 1:5. The Adam optimizer was used during training, with a learning rate set to 0.001. The model converged after 50 epochs. Test data showed that the model's accuracy in identifying early progressive hidden dangers increased from 68% using the traditional threshold method to 94%, and the response time to sudden short-circuit events was compressed from 120 seconds to 180 milliseconds. In this incident, the LSTM model dynamically assigned weights to different parameters (0.45 for current, 0.35 for temperature, and 0.2 for voltage) through a temporal attention mechanism. The model issued a Level 3 alert at 2:30 PM, predicting a 91.7% probability of a short-circuit fire within the next two hours. Testing revealed carbonization of the motor winding insulation due to long-term overload, increasing the contact resistance from 0.1Ω to 1.2Ω, which matched the model's predicted resistance mutation threshold. If not addressed promptly, this event would have resulted in an estimated 20 million RMB in equipment damage and 12 hours of production downtime. Since the implementation of this solution, the campus' electrical fire alarm accuracy has increased to 98.6%, the false alarm rate has dropped from 35% to 4.2%, annual equipment maintenance costs have been reduced by 28%, and the average equipment lifespan has been extended by 15%. Furthermore, the constructed electrical fault knowledge graph continuously optimizes the warning rule base by linking data from over 5,000 fire cases worldwide.

[0042] In the embodiment of the present application, the CNN algorithm formula is:

[0043] in, The extracted electrical fire related features, To input the spatiotemporal data of electrical monitoring, is the row position of the output feature map, is the column position of the output feature map, is the offset of the convolution kernel in the row direction, is the offset of the convolution kernel in the column direction, Indicates that the convolution kernel is at position The weight value at , M is the number of rows of the convolution kernel, N is the number of columns of the convolution kernel, and b is the bias term.

[0044] It can be understood that the embodiment of the present application adapts to the spatial correlation of electrical parameters and the recognition of local abnormal patterns through the local feature extraction capability of the convolution kernel. The convolution layer is used to perform sliding window operations on multi-dimensional sensor data to automatically capture the coupling characteristics of multiple parameters at the same time, as well as the dynamic evolution of local features at different time points. The spatial features of high-dimensional data are extracted to learn fire-sensitive patterns directly from the original sensor signals; feature dimensionality reduction and noise filtering are achieved through the pooling layer, and it has strong robustness to noise such as electromagnetic interference and measurement errors, thereby improving the signal-to-noise ratio of abnormal features; it supports end-to-end training, adapts to the parameter ranges of different devices, and quickly adapts to new monitoring objects through transfer learning. The electrical parameters are converted into parameter-to-time two-dimensional images according to time series, and a deep convolutional network is used to identify thermal map anomalies of equipment operation, thereby improving the recognition accuracy of hidden dangers such as short circuits, overloads, and abnormal contact resistance, and shortening the response time.

[0045] For example, large commercial complexes feature a complex array of electrical equipment and crisscrossing electrical wiring, posing a high fire risk. To enhance fire safety, an electrical fire monitoring solution based on the CNN algorithm was introduced. During deployment, high-definition cameras and various sensors were densely installed at key locations on electrical equipment, at electrical wiring nodes, and in public areas to collect data on the equipment's operating status, wiring temperature, and on-site imagery. The massive amount of data collected includes samples showing normal operation with stable electrical parameters, uniform surface temperature, and normal images, as well as samples showing abnormal conditions such as overloads and short circuits, resulting in electrical parameter fluctuations, sudden temperature rises, and localized smoke. This data, totaling over 100,000, includes 50,000 images of normal conditions, 30,000 images of early-stage faults, and 20,000 images of severe faults. The electrical parameter data includes multi-dimensional values ​​such as voltage, current, and power factor. After processing, the data is fed into a CNN algorithm model for training. The CNN model uses convolutional layers to extract image and electrical parameter features. Pooling layers reduce the data's dimensionality to reduce computational complexity. Fully connected layers integrate the features and output a classification result, determining whether the current electrical status is normal. After multiple rounds of training and optimization, the model achieved an accuracy rate of over 95%. The solution monitors electrical equipment in real time. Late one night, a distribution box on a floor experienced a local short circuit due to aging wiring, causing a rapid temperature rise and the generation of a small amount of smoke. Cameras captured the smoke image, sensors collected abnormal electrical parameters, and transmitted the data to the monitoring terminal. The CNN algorithm model quickly analyzed and identified an electrical fire hazard, triggering an alarm within 0.5 seconds. The specific location and type of fault were simultaneously transmitted to the fire control center and the property management personnel's mobile phones. On-duty personnel responded immediately, rushing to the scene with firefighting equipment. They promptly implemented measures such as power outages and fire extinguishing, successfully defusing the fire and preventing serious damage. Long-term operational statistics show that after implementing the CNN algorithm-based electrical fire monitoring solution, the commercial complex detected electrical fire hazards an average of 10 minutes earlier than traditional monitoring methods. Fire warning accuracy increased from 60% to 95%, and the false alarm rate dropped from 20% to 5%, effectively ensuring the safety of firefighters and property within the commercial complex.

[0046] In an embodiment of the present application, the operating status of electrical equipment is predicted based on a fire hazard identification model to obtain an operating status trend prediction result, including: obtaining the operating status data of the electrical equipment; based on the fire hazard identification model, before training and prediction, the operating status data of the electrical equipment is Z-score normalized, and the early stopping method is used during training to prevent overfitting. At the same time, data enhancement is used to expand the sample optimization parameters, and the predicted value and change of the fire hazard point are output, and the equipment trend label is generated according to the risk change rate; through threshold comparison, historical pattern matching and risk classification mapping, the equipment trend label is converted into a visual risk classification signal, the risk classification signal is comprehensively judged, and a development trend prediction result is generated.

[0047] Among them, early stopping refers to a common technique to prevent model overfitting and optimize training efficiency.

[0048] It is understandable that the embodiment of the present application provides a solution to the problem of model overfitting by constructing a dynamic mechanism for real-time monitoring of validation set indicators. During the model training process, the algorithm continuously monitors the key performance indicators of the validation set. When it is found that the validation set performance has not been optimized for multiple consecutive cycles, the mechanism of forced termination of training is immediately triggered to avoid the model from over-learning the noise or local features in the training data, improving the model's generalization ability for unknown data, suppressing overfitting, and improving the model's prediction accuracy in real scenarios. By automatically identifying the best training nodes, computing resources are saved, and it is suitable for complex scenarios where deep neural network training takes a long time; the training time is automatically determined by dynamic adaptive logic, which reduces the cost of manual parameter adjustment and enhances model stability.

[0049] For example, Figure 4As shown, in the Chongqing Bishan District Intelligent Electrical Fire Prevention Project, addressing the pain points of aging electrical equipment in older urban areas and high false alarm and omission rates of traditional monitoring, the project deployed over 20,000 IoT sensors to collect 10 types of electrical parameters, including current, voltage, and temperature, in real time. This project aggregated 21.2 billion historical data points to build a deep neural network model and introduced early stopping to optimize the training process, forming a comprehensive prevention and control model encompassing "data collection - intelligent analysis - precise early warning." Specifically, the project used the Long Short-Term Memory (LSTM) network as its core algorithm, designed a network structure with three hidden layers, and employed an early stopping mechanism with a patience value of 15 training cycles. The validation set mean squared error (MSE) was used as the termination metric, forcing training to terminate after 15 consecutive validation cycles without a decrease in the validation loss. This resulted in early stopping at the 85th cycle of the originally planned 200-cycle training, significantly saving 57.5% of computing resources. Furthermore, by suppressing overfitting, the test set accuracy increased from 82.3% to 88.7%, the false alarm rate decreased from 23% to 11%, and the omission rate decreased from 16% to 7%, significantly improving the model's generalization and computational efficiency. In the 18 months since the program went live, it has issued 14,542 warnings for electrical fire hazards, successfully assisted in the handling of over 60 initial incidents, and accurately located and guided the remediation of 65 hidden electrical faults. This has contributed to a 41% year-on-year decrease in the district's electrical fire rate, directly preventing over 1.8 million yuan in economic losses. The project's breakthrough lies in its intelligent prevention and control model, which combines dynamic threshold calibration, linkage, and continuous iterative optimization. Differentiated warning rules are automatically generated based on the power usage characteristics of different scenarios, achieving a 92% accuracy rate for nighttime warnings in residential areas. The warning terminal integrates millisecond-level linkage with the fire command center and community grid platform, automatically generating work orders with three-dimensional remediation plans when a Level III risk is triggered. The model is retrained monthly based on real-world cases, and seven major version updates have covered emerging scenarios such as new energy equipment and distributed photovoltaics, increasing recognition coverage to 85%. As a model for data-driven smart firefighting, this project leverages the dual mechanisms of "training efficiency optimization" and "monitoring dynamic parameter adjustment" to shift electrical fire prevention from "post-event response" to "pre-event prevention," reducing potential losses by 12 yuan for every 1 yuan invested.

[0050] In step S103, based on the fire hazard type identification results, combined with the use environment of the electrical equipment and the threshold judgment algorithm, the fire risk is quantitatively assessed to obtain the corresponding risk level. When the risk level exceeds the set multi-layer threshold, the linkage control mechanism is automatically started. Among them, when the risk level is low and exceeds the first-level threshold, the local early warning is triggered and the light flashes. When it is medium and exceeds the second-level threshold, the remote alarm is triggered and a text message is pushed to the mobile phone. When it is high and exceeds the third-level threshold, the fire extinguishing device is triggered and the power is automatically cut off.

[0051] The threshold determination algorithm is a basic decision-making method for binary classification or triggering response based on a preset critical value. The formula is:

[0052] S is the overall risk level of the current monitoring parameters, is the electrical parameter number, is the total number of electrical monitoring parameters, For the The weight of the parameters, For the The real-time value of the parameter, For the The safety upper threshold of the parameter, It is an alarm state.

[0053] It can be understood that the embodiments of the present application convert data into clear risk judgment standards by setting safety thresholds for various types of monitoring data, quickly identify abnormal signals, and implement risk-graded response; at the same time, by setting fluctuation tolerance ranges and dynamically adaptively adjusting thresholds, it effectively filters noise interference and avoids false alarms, providing a standardized decision-making basis for fire management, and accurately matching response strategies with risk levels, which is the core technical guarantee for fire warning and disposal.

[0054] For example, in smart firefighting scenarios within large commercial complexes, a large number of smoke and temperature sensors are deployed. Long-term data monitoring and analysis have determined that under normal circumstances, the average smoke concentration is 3 ppm with a standard deviation of 0.5 ppm, and the average temperature is 25°C with a standard deviation of 1°C. Using a threshold determination algorithm, the smoke concentration threshold is set at 4.5 ppm (the mean plus three standard deviations), and the temperature threshold is set at 28°C. When the smoke sensor reading in a store reaches 5 ppm, exceeding the 4.5 ppm threshold, and the temperature sensor reading is 29°C, exceeding the 28°C threshold, the monitoring terminal quickly determines that a fire is highly likely, immediately issues a high-decibel alarm, and pushes an alert to the fire control center, buying valuable time for rapid response and firefighting.

[0055] In an embodiment of the present application, a quantitative assessment of fire risk is performed based on the fire hazard type identification results, combined with the use environment of the electrical equipment and the threshold judgment algorithm, including: obtaining fire hazard type data; based on the fire hazard type data, combined with the hazard disaster-causing characteristics, historical data and environmental parameters, determining the basic risk weight of the hazard type through the hierarchical analysis method; based on the basic risk weight of the hazard type, using a nonlinear mapping function to characterize the nonlinear coupling relationship between the basic hazard risk and the ambient temperature correction factor, and constructing a nonlinear risk model; according to the nonlinear risk model, calculating the risk index in real time, and dividing the risk index into levels through the threshold judgment algorithm, and performing a quantitative assessment and level judgment on the fire risk.

[0056] Among them, the hierarchical analysis method is a multi-criteria decision analysis method that decomposes complex problems into a multi-level structure, determines the weight of each factor by comparing each level one by one, and realizes a combination of quantitative and qualitative methods. The formula is: , , ,CR is the random consistency ratio, is the consistency indicator, The average consistency index of the randomly generated judgment matrix, is the overall deviation degree of the comparison of the importance of each risk factor, is the number of risk factors for comparison, W is the weight vector of electrical fire risk index, and A is the judgment matrix of electrical fire risk factors.

[0057] It can be understood that the embodiments of the present application integrate equipment status, environmental factors, and operation and maintenance management data, and use a weight distribution mechanism that combines subjective and objective factors to dynamically identify high-risk areas, quantitatively evaluate and classify risk levels, and accurately locate key risk points such as cable aging and loose connections; based on the consistency test of the hierarchical structure, the credibility of the assessment results is improved, and full-process decision-making support from risk warning to hidden danger investigation is provided to the fire department, thereby improving the comprehensiveness, accuracy and emergency response efficiency of electrical fire monitoring.

[0058] It should be noted that the nonlinear risk model refers to an intelligent model used to characterize the complex nonlinear relationship between electrical parameters and fire risk.

[0059] It can be understood that the embodiments of the present application use nonlinear risk models to capture the coupling effects and dynamic correlations of current, temperature, and harmonic indicators, and identify hidden hidden dangers such as gradual insulation aging and sudden transient harmonic surges; use neural networks to automatically extract high-order interactive features such as current fluctuation slope and temperature changes, identify combined anomalies that cannot be captured by linear models, improve the recognition rate of hidden hidden dangers, dynamically calibrate the risk thresholds of equipment in different time periods based on time series data, and guide the differentiated scheduling of fire protection resources according to risk levels; integrate electrical parameters, environmental humidity, equipment life and other data to construct a multi-layer nonlinear model, improve the accuracy of early warning, and visualize high-risk loops through risk heat maps.

[0060] For example, in large chemical parks, nonlinear risk models provide a rigorous safety barrier. The park houses over 50 storage tanks for various hazardous chemicals and dozens of kilometers of pipelines. Numerous temperature, pressure, and combustible gas concentration sensors are installed, generating over 10,000 data points per minute. Nonlinear risk models are used to deeply explore the complex nonlinear relationships between these various data types. For example, on a summer afternoon in a large tank area storing Class A flammable liquids, high temperatures and strong sunlight caused sensor data to change: the tank temperature slowly climbed from 30°C to 35°C over two hours, the pressure slightly increased from 0.8 MPa to 0.85 MPa, and the surrounding combustible gas concentration rose from 50 ppm to 150 ppm. The nonlinear risk model does not rely on a single threshold to make its judgment, but instead comprehensively considers the trend, rate of change, and interactions between these data. Through complex calculations, it arrives at a risk value of 0.8. The monitoring terminal quickly determined that this was a high-risk situation, immediately triggering the sound and light alarms in the park, notifying the fire emergency team to rush to the scene, and automatically starting the cooling spray to cool the storage tank, successfully resolving the potential fire crisis and avoiding possible large-scale explosions and fire accidents.

[0061] In the embodiment of the present application, the risk quantification assessment formula is:

[0062] Among them, R is the comprehensive risk value, is the number of electrical parameters, is the total number of electrical parameters, For the The risk factor function of parameters, For the The weight of each electrical parameter, T is the trend persistence coefficient, and C is the confidence coefficient.

[0063] It can be understood that the embodiment of the present application converts abstract fire risks into visual quantitative indicators by integrating multi-source data such as electrical parameters, equipment status and environmental factors, and reflects the severity of fire hazards through weighted calculation of data such as current, temperature, and smoke concentration. The risk value is updated according to real-time data to grasp the trend of risk changes; at the same time, based on historical data, risk patterns are mined to provide data for management decisions, improve fire prevention and control capabilities in advance, and reduce false alarm and missed alarm rates and emergency response costs.

[0064] For example, Figure 5As shown in the figure, a smart fire monitoring system in a high-rise office building uses a risk quantification assessment formula for fire monitoring. This formula comprehensively considers multiple factors, including floor height, electrical equipment density, human density, and the amount of flammable materials. Based on historical data and expert experience, these factors were weighted: floor height with a weight of 0.2, electrical equipment density with a weight of 0.3, human density with a weight of 0.3, and the amount of flammable materials with a weight of 0.2. After calculation, the fire risk value for this area reached 0.7. Once the monitoring system determined this risk value, it immediately issued an early warning to the property fire control center. Relevant personnel swiftly took measures to strengthen patrols and firefighting equipment inspections in the area, successfully averting a potential fire.

[0065] The following will describe the smart fire protection electrical fire monitoring method through a specific embodiment. Figure 6 As shown, taking a large industrial park as an example, it includes: In the data acquisition and preprocessing stage: The industrial park houses numerous enterprises, with a vast and diverse array of electrical equipment. To ensure comprehensive and accurate data collection, a large number of intelligent sensors, including residual current sensors, temperature sensors, current sensors, and voltage sensors, have been installed in each enterprise's distribution boxes and rooms, at key electrical equipment, and at key electrical line nodes. These sensors offer high-precision data acquisition capabilities, capturing real-time changes in electrical parameters. For example, residual current data is collected every three seconds, and temperature, current, and voltage data are collected every five seconds, ensuring timely monitoring of the operating status of electrical equipment. Furthermore, high-definition cameras are deployed in key areas of the park, capturing high-definition images of the electrical equipment's operating status every 20 seconds to capture information about the equipment's appearance and surrounding environment. The collected data inevitably contains noise, so advanced filtering algorithms are used to process the sensor data. The Kalman filter smoothes out data susceptible to interference, such as current and voltage, making data fluctuations more stable and accurately reflecting the actual operating status of the electrical equipment. Image data is first grayscaled to reduce its dimensionality, then denoised using Gaussian filtering to remove noise. Finally, the image is normalized to uniform size, laying a solid foundation for subsequent analysis. Regarding data transmission, a complementary wired and wireless network approach is employed, taking into account the specific conditions of the park. Preprocessed data is transmitted to the central data server in real time and stably, ensuring efficient and reliable data transmission.

[0066] In the stage of establishing the fire hazard identification model: Based on massive electrical data and image data, the LSTM algorithm and CNN algorithm are used to build a fire hazard identification model. For electrical parameter data, The LSTM model constructed by the formula processes time series data and can deeply learn the dynamic changes of electrical parameters over time. During the model training process, the electrical equipment operation data of the past two years were selected as training samples, of which 70% were normal operation data samples and 30% were data samples of periods where electrical faults occurred. The internal parameters of the model were continuously adjusted to enable the model to accurately identify various electrical parameter abnormalities such as current overload, abnormal voltage fluctuation, and excessive residual current. For image data, the The CNN model is constructed using a specific formula. The model consists of multiple convolutional layers, pooling layers, and fully connected layers. The convolutional layers extract key features such as edges, textures, and shapes from the image. The pooling layers reduce feature dimensionality and computational complexity. The fully connected layers perform classification and judgment of the extracted features. To improve the model's generalization capabilities, a massive amount of images of normal equipment operation and images with fire hazards were collected for training. The total number of training samples reached 80,000, including 50,000 normal images and 30,000 images with fire hazards. After multiple rounds of intensive training, the CNN model achieved an accuracy rate of 96% for identifying images with electrical fire hazards.

[0067] During the hazard type identification phase: When the real-time collected electrical data and image data are input into the trained model, the LSTM model is used to analyze the changing trends, accelerations, and periodic patterns of electrical parameters such as current to achieve short-term and long-term load forecasts. At the same time, the CNN model is combined with the optical flow neural network and video sequence prediction model to trace the abnormal features of the image and simulate the future state. On this basis, the predicted results of the electrical parameters and images are aligned in time and space and multimodally fused. All prediction results are visualized through the digital twin engine, and the disposal plan is automatically matched, ultimately achieving a leap from real-time identification of hidden dangers to active prediction of future risks, buying more time for fire prevention and control.

[0068] During the risk quantification and assessment phase: Based on the identified fire hazard type, comprehensive consideration is given to environmental factors such as the duration of the hazard, the importance of the equipment area, and the distribution of flammable materials, combined with a threshold judgment algorithm. Apply it from the perspectives of hidden danger impact scope, potential loss degree, etc. Quantitative assessments are performed. For example, if a current overload hazard occurs in a core production facility and lasts for more than 40 minutes, it is assessed as a high risk due to the potential severe impact on park operations. If it only affects auxiliary areas and lasts for a short time, it is assessed as a medium-low risk. By constructing a risk assessment matrix covering hazard types such as short circuits, overloads, and equipment overheating, different hazard characteristics are matched to multiple risk thresholds. Once the risk level exceeds the preset threshold, a hierarchical linkage control mechanism is automatically triggered.

[0069] In the early warning and feedback stage: Once the monitoring system identifies a high-risk hazard, a comprehensive early warning mechanism is immediately activated. Warnings are sent to park staff simultaneously via SMS, app push notifications, audio and visual alarms, and campus broadcasts. On the smart fire monitoring platform, a striking flashing red icon highlights the hazard's location, type, and risk level, allowing management personnel to quickly locate and understand the situation. Upon receiving the warning, relevant personnel responded swiftly. The park's fire management department immediately reviewed surveillance video to further confirm the hazard and coordinated resources. Property maintenance personnel, equipped with specialized testing and repair tools, rushed to the scene as quickly as possible to de-energize and repair the hazardous equipment. Park safety management staff also organized on-site personnel to prepare for safety precautions and evacuation. For example, during an equipment overheating warning incident, maintenance personnel arrived and discovered that debris had blocked the equipment's cooling ducts, resulting in excessive heat. They immediately cleared the ducts and restored normal heat dissipation, rapidly reducing the temperature and resuming normal operation, successfully averting a potential electrical fire. Since the intelligent fire protection electrical fire monitoring system was put into use, it has issued more than 500 effective warnings and successfully handled more than 450 electrical fire hazards of various types, reducing the incidence of electrical fires in the park by 70%, and effectively ensuring the safe production and stable operation of the industrial park.

[0070] In summary, this embodiment builds an efficient prevention and control system through five key phases: The data acquisition and preprocessing phase utilizes intelligent sensors and high-definition cameras to frequently collect electrical parameters and equipment images, and then applies filtering, grayscaling, and other algorithms to improve data quality. The fire hazard identification model phase utilizes the LSTM algorithm to process electrical parameter time series data and the CNN algorithm to analyze image data. Through extensive sample training, the model achieves a high level of recognition accuracy. The hazard type identification and risk quantification assessment phase accurately determines hazard types based on model results, and quantitatively classifies them using a risk assessment matrix based on multiple factors. The early warning and response phase provides timely warnings through multiple channels, enabling relevant personnel to quickly respond and address hazards. Since the monitoring system was put into operation, it has issued over 500 effective warnings, addressed over 450 hazards, and reduced the incidence of electrical fires by 70%. This demonstrates its ability to integrate multi-source data for early hazard identification, accurate assessment, and timely response, providing replicable experience for other scenarios and significant value in ensuring public safety.

[0071] Next, a smart fire protection electrical fire monitoring device proposed according to an embodiment of the present application is described with reference to the accompanying drawings.

[0072] Figure 7 It is a block diagram of an intelligent fire protection electrical fire monitoring device according to an embodiment of the present application.

[0073] like Figure 7As shown, the smart fire protection electrical fire monitoring device 10 includes: an acquisition module 100, a construction module 200, an identification module 300, an evaluation module 400 and a feedback module 500.

[0074] Among them, the acquisition module 100 is used to obtain electrical parameters; the construction module 200 is used to establish a fire hazard identification model using the LSTM algorithm and the CNN algorithm; the identification module 300 is used to predict the operating status of the electrical equipment according to the fire hazard identification model, obtain the operating status trend prediction results, and at the same time, identify the potential fire hazard types; the evaluation module 400 is used to quantitatively evaluate the fire risk based on the fire hazard type identification results, combined with the use environment of the electrical equipment and the threshold judgment algorithm, to obtain the corresponding risk level; the feedback module 500 is used to automatically start the linkage control mechanism when the risk level exceeds the set multi-layer threshold. Among them, when the risk level is low and exceeds the first-level threshold, the local early warning is triggered and the light flashes. When it is medium and exceeds the second-level threshold, the remote alarm is triggered and a text message is pushed to the mobile phone. When it is high and exceeds the third-level threshold, the fire extinguishing device is triggered and the power is automatically cut off.

[0075] It should be noted that the above explanation of an embodiment of a smart fire electrical fire monitoring method is also applicable to an smart fire electrical fire monitoring device of this embodiment, and will not be repeated here.

[0076] According to an intelligent fire protection electrical fire monitoring device proposed in an embodiment of the present application, the LSTM algorithm is used to capture the time-series dependency characteristics of electrical parameters and the CNN algorithm is used to extract spatial correlation and frequency domain anomaly characteristics to construct a multimodal fire hazard identification model. Combined with real-time data preprocessing on the edge and continuous learning mechanism on the cloud, the problems of weak intermittent fault feature extraction and delayed warning in traditional solutions are solved. SMOTE sample balancing, five-fold cross-validation and other technologies are used to improve the model's recognition accuracy for minority hidden dangers, and Z-score normalization and early stopping method are used to prevent overfitting, thereby enhancing the early warning capability in complex scenarios. A nonlinear risk model is constructed through the hierarchical analysis method, and environmental parameters are dynamically integrated to correct the hidden danger weights. Combined with a three-level threshold hierarchical linkage mechanism, the entire process from risk identification to disposal is automated to avoid response delays caused by manual intervention, reduce labor costs, and effectively solve the core problems of high labor dependence, high intermittent fault missed detection rate and insufficient warning capability in related technologies.

[0077] Figure 8 This is a schematic diagram of the structure of an electronic device provided in an embodiment of the present application. The electronic device may include: A memory 801 , a processor 802 , and a computer program stored in the memory 801 and executable on the processor 802 .

[0078] When the processor 802 executes the program, the intelligent fire protection electrical fire monitoring method provided in the above embodiment is implemented.

[0079] Furthermore, the electronic device further includes: The communication interface 803 is used for communication between the memory 801 and the processor 802 .

[0080] The memory 801 is used to store computer programs that can be run on the processor 802.

[0081] The memory 801 may include a high-speed RAM (Random Access Memory) memory, and may also include a non-volatile memory, such as at least one disk memory.

[0082] If the memory 801, processor 802, and communication interface 803 are implemented independently, the communication interface 803, memory 801, and processor 802 can be connected to each other via a bus and communicate with each other. The bus can be an ISA (Industry Standard Architecture) bus, a PCI (Peripheral Component Interconnect) bus, or an EISA (Extended Industry Standard Architecture) bus. The bus can be divided into an address bus, a data bus, a control bus, etc. For ease of representation, Figure 8 Only one thick line is used in the diagram, but this does not mean that there is only one bus or one type of bus.

[0083] Optionally, in a specific implementation, if the memory 801, the processor 802 and the communication interface 803 are integrated on a chip, the memory 801, the processor 802 and the communication interface 803 can communicate with each other through an internal interface.

[0084] The processor 802 may be a CPU (Central Processing Unit), or an ASIC (Application Specific Integrated Circuit), or one or more integrated circuits configured to implement the embodiments of the present application.

[0085] An embodiment of the present application also provides a computer-readable storage medium on which a computer program or instruction is stored. When the computer program or instruction is executed by a processor, it implements the above-mentioned intelligent fire protection electrical fire monitoring method.

[0086] An embodiment of the present application also provides a computer program product, including a computer program or instructions, characterized in that when the computer program or instructions are executed, they implement the above-mentioned intelligent fire protection electrical fire monitoring method.

[0087] In the description of this specification, the description with reference to the terms "one embodiment", "some embodiments", "example", "specific example", or "some examples" means that the specific features, structures, materials or characteristics described in conjunction with the embodiment or example are included in at least one embodiment or example of the present application. In this specification, the schematic representations of the above terms do not necessarily refer to the same embodiment or example. Moreover, the specific features, structures, materials or characteristics described can be combined in any one or N embodiments or examples in a suitable manner. In addition, those skilled in the art can combine and combine different embodiments or examples described in this specification and features of different embodiments or examples without contradiction.

[0088] Furthermore, the terms "first" and "second" are used for descriptive purposes only and should not be understood to indicate or imply relative importance or implicitly specify the number of technical features indicated. Thus, a feature specified as "first" or "second" may explicitly or implicitly include at least one such feature. In the description of this application, "N" means at least two, for example, two, three, etc., unless otherwise specifically defined.

[0089] Any process or method description in a flowchart or otherwise described herein may be understood to represent a module, fragment or portion of code comprising one or N executable instructions for implementing a custom logical function or process step, and the scope of the preferred embodiments of the present application includes alternative implementations in which functions may be performed in a different order than shown or discussed, including performing functions in a substantially simultaneous manner or in a reverse order depending on the functions involved, which should be understood by those skilled in the art to which the embodiments of the present application pertain.

[0090] It should be understood that various parts of this application can be implemented using hardware, software, firmware, or a combination thereof. In the above embodiment, the N steps or methods can be implemented using software or firmware stored in a memory and executed by a suitable instruction execution system. For example, if implemented using hardware, as in another embodiment, it can be implemented using any one or a combination of the following technologies known in the art: discrete logic circuits having logic gate circuits for implementing logical functions on data signals, application-specific integrated circuits having suitable combinational logic gate circuits, programmable gate arrays (PGAs), field-programmable gate arrays (FPGAs), etc.

[0091] Those skilled in the art will appreciate that all or part of the steps in the method for implementing the above-mentioned embodiment can be completed by instructing related hardware through a program, and the program can be stored in a computer-readable storage medium. When the program is executed, it includes one or a combination of the steps of the method embodiment.

[0092] Although the embodiments of the present application have been shown and described above, it can be understood that the above embodiments are exemplary and cannot be understood as limitations on the present application. Ordinary technicians in this field can change, modify, replace and modify the above embodiments within the scope of the present application.

Claims

1. A smart fire protection electrical fire monitoring method, characterized in that: include: Get electrical parameters; Based on the electrical parameters, a fire hazard identification model is established using the LSTM algorithm and the CNN algorithm. The operating status of the electrical equipment is predicted based on the fire hazard identification model to obtain an operating status trend prediction result, and at the same time, the potential fire hazard type is identified; According to the fire hazard type identification results, combined with the use environment of the electrical equipment and the threshold judgment algorithm, the fire risk is quantitatively assessed to obtain the corresponding risk level. When the risk level exceeds the set multi-layer threshold, the linkage control mechanism is automatically activated. Among them, when the risk level is low and exceeds the first-level threshold, a local early warning is triggered and the light flashes. When it is medium and exceeds the second-level threshold, a remote alarm is triggered and a text message is pushed to the mobile phone. When it is high and exceeds the third-level threshold, the fire extinguishing device is triggered and the power is automatically cut off.

2. A smart fire protection electrical fire monitoring method according to claim 1, characterized in that: Based on the electrical parameters, a fire hazard identification model is established using the LSTM algorithm and the CNN algorithm, including: Build a training dataset; Based on the training data set, a single-device anomaly detection and regional risk association model is constructed using the LSTM algorithm and the CNN algorithm; Based on the single-device anomaly detection and regional risk association model, the SMOTE algorithm is used for sample balancing through data annotation methods, and the model is trained after five-fold cross-validation and confusion matrix evaluation; According to the trained model, a fire hazard identification model is formed through optimization through a multimodal data fusion algorithm combined with edge-cloud collaborative deployment and a continuous learning mechanism.

3. The intelligent fire protection electrical fire monitoring method according to claim 1, characterized in that: The LSTM formula is: ； in, Indicates the long-term and short-term timing dependency information of the electrical parameters stored at the current moment. Represents historical temporal memory, For the Gate of Forgetfulness, is the input gate, is the hyperbolic tangent activation function, For input to the weight matrix of the candidate state space, For input to the candidate state space to the weight matrix, is the temporal abstract feature after processing at the previous moment, The electrical parameter timing data collected at this moment, is an independent bias term.

4. The intelligent fire protection electrical fire monitoring method according to claim 1, characterized in that: The CNN algorithm formula is: ； in, represents the extracted electrical fire related features, To input the spatiotemporal data of electrical monitoring, is the row position of the output feature map, is the column position of the output feature map, is the offset of the convolution kernel in the row direction, is the offset of the convolution kernel in the column direction, Indicates that the convolution kernel is at position The weight value at , M is the number of rows of the convolution kernel, N is the number of columns of the convolution kernel, and b is the bias term.

5. The intelligent fire protection electrical fire monitoring method according to claim 1, characterized in that: The operating status of the electrical equipment is predicted based on the fire hazard identification model to obtain an operating status trend prediction result, including: Obtaining electrical equipment operating status data; Based on the fire hazard identification model, before training and prediction, the operating status data of the electrical equipment is subjected to Z-score normalization processing. During training, the early stopping method is used to prevent overfitting. At the same time, data enhancement is used to expand samples and optimize parameters. The predicted value and change of the fire hazard point are output, and the equipment trend label is generated according to the risk change rate. Through threshold comparison, historical pattern matching and risk classification mapping, the device trend label is converted into a visual risk classification signal, and the risk classification signal is comprehensively judged to generate a development trend prediction result.

6. The intelligent fire protection electrical fire monitoring method according to claim 1, characterized in that: Based on the fire hazard type identification results, combined with the use environment of the electrical equipment and the threshold determination algorithm, a quantitative assessment of the fire risk is performed, including: Obtain fire hazard type data; Based on the fire hazard type data, combined with the hazard-causing characteristics, historical data and environmental parameters, the basic risk weight of the hazard type is determined through the hierarchical analysis method; Based on the basic risk weight of the hidden danger type, a nonlinear mapping function is used to characterize the nonlinear coupling relationship between the basic risk of the hidden danger and the ambient temperature correction factor, and a nonlinear risk model is constructed; According to the nonlinear risk model, the risk index is calculated in real time, and the risk index is graded through a threshold determination algorithm, so as to quantitatively evaluate and grade the fire risk.

7. The intelligent fire protection electrical fire monitoring method according to claim 1, characterized in that: The fire risk is quantitatively assessed, wherein the formula for the quantitative assessment is: ； Among them, R is the comprehensive risk value, is the number of electrical parameters, is the total number of electrical parameters, For the The risk factor function of parameters, For the The weight of the electrical parameters, is the trend persistence coefficient, is the confidence coefficient.

8. A smart fire protection electrical fire monitoring device, characterized in that: include: Acquisition module, used to obtain electrical parameters; Building a module for building a fire hazard identification model using LSTM and CNN algorithms; an identification module for predicting the operating status of electrical equipment according to the fire hazard identification model, obtaining an operating status trend prediction result, and identifying potential fire hazard types; An assessment module is used to quantitatively assess the fire risk based on the fire hazard type identification result, combined with the use environment of the electrical equipment and the threshold judgment algorithm, to obtain a corresponding risk level; The feedback module is used to automatically start the linkage control mechanism when the risk level exceeds the set multi-layer threshold. When the risk level is low and exceeds the first-level threshold, a local early warning is triggered and the light flashes. When the risk level is medium and exceeds the second-level threshold, a remote alarm is triggered and a text message is pushed to the mobile phone. When the risk level is high and exceeds the third-level threshold, the fire extinguishing device is triggered and the power supply is automatically cut off.

9. An electronic device, characterized in that: include: A memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the program to implement a smart fire protection electrical fire monitoring method as described in any one of claims 1 to 7.

10. A computer-readable storage medium having a computer program or instruction stored thereon, characterized in that: When the computer program or instruction is executed, it implements the intelligent fire protection electrical fire monitoring method described in any one of claims 1-7.

Images