Raspberry Pi-based fault arc positioning detection and electrical fire early warning method

Abstract

The invention relates to a Raspberry Pi-based fault arc positioning detection and electrical fire early warning method, and belongs to the field of fault arc fire. According to the method, Raspberry Pi is used as a core processing unit, multiple types of sensors are integrated, and multiple kinds of alternating current and direct current load data are collected to establish a typical multi-dimensional data waveform library; deep processing is carried out on multi-time-frequency features and wavelet transform features of the signals in combination with a residual neural network, multi-dimensional features of a fault arc are comprehensively captured, and alternating-current and direct-current universal fault arc detection is achieved; multi-dimensional data are analyzed, and the arc danger level is evaluated through the time difference of fault arcs detected by different sensors, so that graded alarm is realized. Meanwhile, the Bayesian network is used for reasoning and judgment, and accurate positioning of the fault arc is achieved. Through cooperative work with the cloud host, a remote monitoring and management platform is established. The invention provides an electrical fire early warning solution which is intelligent, real-time and high in reliability.

Description

Technical Field

[0001] The present invention belongs to the field of faulty arc fires, and particularly relates to a method for detecting and locating faulty arcs and warning of electrical fires based on Raspberry Pi. Background Art

[0002] With the development of modern society and the wide application of electrical equipment, electrical fires have become one of the main types of fires. Arcs, short circuits, overvoltage, and overload are important causes of electrical fires. Existing protection devices such as circuit breakers and fuses can detect short circuits, overvoltage, and overload. However, due to the small current, high temperature, and short duration of faulty arcs, traditional protection devices are difficult to detect in a timely manner. The non-linear and intermittent characteristics of faulty arcs make detection even more difficult. With the rapid development of renewable energy systems, there is an urgent need in the market for a faulty arc detection device that is compatible with both AC and DC. However, most of the existing faulty arc detection devices on the market are usually designed specifically for AC or DC systems, rather than being compatible with both. Moreover, the detection methods of faulty arcs mostly rely on time-domain or frequency-domain analysis of a single signal, usually using current or voltage signals to judge the occurrence of faulty arcs, which is prone to false judgment due to the diversity of loads. At the same time, these methods can usually only detect the presence of faulty arcs, but cannot accurately locate the fault position, which affects subsequent fault handling and maintenance. In addition, most of the existing systems are statically set. Once the thresholds and detection models are configured, it is difficult to dynamically adjust and adapt to changes in different environments or loads, which makes the performance of the detection system decline gradually during long-term operation. Summary of the Invention

[0003] The purpose of the present invention is to provide a method for detecting and locating faulty arcs and warning of electrical fires based on Raspberry Pi. This method gives full play to the advantages of Raspberry Pi, such as high flexibility, rich interfaces, easy hardware expansion, and support for remote management, and provides an intelligent, real-time, and highly reliable solution for warning of electrical fires.

[0004] To achieve the above purpose, the technical solution of the present invention is: A method for detecting and locating faulty arcs and warning of electrical fires based on Raspberry Pi, including:

[0005] Taking Raspberry Pi as the core processing unit, integrating multiple types of sensors, collecting various AC and DC load data, and establishing a typical multi-dimensional fusion faulty arc database;

[0006] Using the obtained data as input, combining with a residual neural network to deeply analyze the multi-time-frequency characteristics and wavelet transform characteristics of the signal, comprehensively capturing the characteristic information of faulty arcs, and realizing the detection of faulty arcs that is common to both AC and DC;

[0007] The time difference of the faulty arc detected by multiple sensors is used to evaluate the risk level of the faulty arc for hierarchical alarm. The Bayesian network is used to reason and judge different fault modes to achieve the precise positioning of the faulty arc;

[0008] Through the collaborative work with the cloud host, a remote monitoring and management platform is built, and self-learning and model updating are regularly carried out using the historical data stored locally.

[0009] In an embodiment of the present invention, the Raspberry Pi is used as the core processing unit, integrating multiple types of sensors to collect various AC and DC load data, and a typical multi-dimensional data waveform library is established. Specifically, the Raspberry Pi is used as the core processing unit, and multiple types of sensors are carried to generate arcs through the faulty arc characteristic test system device to obtain current, voltage, smoke, and temperature waveform data of various AC and DC loads. Signal preprocessing including noise filtering and normalization is performed on the collected waveform data to establish a multi-dimensional fusion faulty arc database.

[0010] In an embodiment of the present invention, taking the acquired data as input, the residual neural network is combined to deeply analyze the multi-time-frequency characteristics and wavelet transform characteristics of the signal, comprehensively capture the characteristic information of the faulty arc, and achieve the detection of faulty arcs common to AC and DC. The specific implementation is as follows:

[0011] Apply the fast Fourier transform FFT to the signal in the multi-dimensional fusion faulty arc database to convert the signal to the frequency domain, analyze its frequency components to identify the periodic changes unique to AC arcs, and then obtain the power distribution of the signal by calculating the power spectral density PSD; based on the FFT and PSD results, extract the features helpful for distinguishing DC and AC arcs, including the main frequency components, energy peaks, and spectral entropy;

[0012] Use wavelet transform for time-frequency analysis to provide the frequency and energy changes of the signal at different time points;

[0013] Use the processed time-frequency domain features to train the residual neural network to detect and classify the faulty arc signals. The residual neural network can effectively extract the features in complex signals through multiple layers of convolution and residual connections, and connect the attention mechanism channels, further improving the accuracy of identifying faulty arcs;

[0014] Deploy the trained residual neural network to the Raspberry Pi to complete the hardware module, automatically distinguish whether the detected arc is DC or AC, and take corresponding protection measures.

[0015] In an embodiment of the present invention, the time difference of the faulty arc detected by multiple sensors is used to evaluate the risk level of the faulty arc for hierarchical alarm. The specific implementation is as follows:

[0016] According to the danger level of arc faults, current sensors and voltage sensors are installed between the key access points of the circuit and the load. Multiple sensors are installed in the areas most prone to arc faults for multi-dimensional detection.

[0017] The trained residual neural network is transplanted to the Raspberry Pi. Based on historical arc fault data and real-time environmental data, the Raspberry Pi is equipped with multiple sensors to receive and analyze signals, and multi-level alarms are performed through the time difference between different sensors detecting the fault arc. When the current sensor or voltage sensor first detects an abnormality, and the temperature sensor, smoke sensor, and light sensor do not report an abnormality, a primary alarm will be triggered. If the temperature sensor also reports a significant increase after the current or voltage abnormality, this indicates that there is an overheating condition, and the alarm will be upgraded to a secondary level, requiring more urgent technical intervention to prevent the problem from worsening. When the light sensor detects strong light emission and the smoke sensor also begins to report smoke accumulation, it indicates that a fire may have broken out or is in progress. In this case, a high-level alarm is triggered, the automatic fire extinguishing system is activated, and the fire department and emergency response team are notified to intervene.

[0018] In one embodiment of the present invention, a Bayesian network is used to infer and judge different fault modes to achieve accurate positioning of the fault arc, which is specifically implemented as follows:

[0019] After learning the features of each sensor data based on the residual neural network, an initial weight is set for each sensor as the prior probability P(θ) of the Bayesian network, where θ represents the occurrence of a fault arc; the multi-dimensional observation data set of the sensor is set by real-time detection of the sensor data as follows: D = {I(t), V(t), T(t), S(t), G(t)}, where I is current, V is voltage, T is temperature, S is smoke, G is light, t is time, and D is a waveform data set collected by a current sensor, a voltage sensor, a temperature sensor, a smoke sensor, and a light sensor for time series analysis; on the basis of the multi-dimensional observation data set A conditional probability distribution table CPT is established for each sensor node to describe the probability distribution P(Dθ) of the sensor observation data in the fault state. When some sensors detect related events, the Bayesian network uses the known observation data to infer the probability distribution of other position-related variables that are not directly observed. As the collected signals are continuously updated, the Bayesian formula is used to update the posterior probability of each sensor, and the area with the largest posterior probability is selected as the area where the fault arc is most likely to occur. The weight of the sensor is subsequently adjusted according to user feedback and the actual situation of the alarm. If an alarm is a false alarm, the weight of the corresponding sensor is reduced; if the alarm is valid, the weight of its sensor is increased.

[0020] In one embodiment of the present invention, the formula for updating the posterior probability of each sensor using the Bayesian formula is expressed as follows:

[0021]

[0022] Where P(θD) is the posterior probability, P(D|θ) is the likelihood function, and P(D) is the normalization term.

[0023] In an embodiment of the present invention, during real-time monitoring, the Raspberry Pi stores information such as sensor data, alarm records, fault characteristics, and model output results collected in a local SQLite database; the Raspberry Pi regularly sorts and filters the locally stored data, and filters out representative fault data for model training and optimization; the system periodically clears expired and no-longer-useful data according to the set policy to save storage space and maintain the efficient operation of the database.

[0024] In an embodiment of the present invention, after the model is updated, the updated model may have a problem of decreasing accuracy due to new data coming in. To ensure that the new model is always better than the old model, the model adaptively adjusts the model structure according to the actual performance, dynamically controls the region complexity through a formula, and prevents overfitting:

[0025]

[0026] Where α is an adjustment factor for controlling the model complexity, representing the structural complexity of the model; θ new represents the updated model parameters. During the model optimization process, the parameters θ are adjusted to balance the performance and complexity of the model; model refers to the machine learning model; complexity is a quantitative index of the model complexity, used to control overfitting; is the partial derivative of the loss function with respect to the model complexity, used to dynamically adjust the complexity.

[0027] In an embodiment of the present invention, after the model is trained or updated, model verification and deployment are required. The validation set D val is used to calculate the accuracy Accuracy of the model:

[0028]

[0029] Where θ is the model function, f ResNet (X; θ) is the output of the residual neural network, X is the input feature vector, Y is the true label, 1[·] is the indicator function, which is 1 when the predicted value is equal to the true value, otherwise 0; when the new model passes the performance evaluation in the validation set and all indicators are better than the old model, safety and stability tests are carried out to ensure the reliability of the model under various boundary conditions; after the test passes, the new model is deployed into the system, and θ deployed = θ newAs the model parameters for the final application, the new model will be applied to the subsequent fault detection process; at the same time, the old model will be retained as a backup so that it can be quickly rolled back in case of problems with the new model, ensuring the overall stability and reliability of the system; through model updates, the system can continuously improve the accuracy of fault arc detection and the reliability of electrical fire warning, avoiding false alarms or missed alarms caused by environmental changes.

[0030] The present invention also provides a computer-readable storage medium, on which computer program instructions capable of being run by a processor are stored. When the processor runs the computer program instructions, the method steps as described in any one of the above can be implemented.

[0031] Compared with the prior art, the present invention has the following beneficial effects:

[0032] 1. The Raspberry Pi is used as the core processing unit in the field of fault arc detection, and a fault arc detection system that is universal for AC and DC is invented. The system is equipped with sensors that can detect AC and DC fault arcs simultaneously, and integrates functions such as real-time data acquisition, fault analysis, warning issuance, and remote monitoring and management.

[0033] 2. A fault arc detection method is proposed, which combines the time difference grading alarm of sensors and the positioning technology of Bayesian networks. There is a time difference in the detection of fault arcs by different sensors. The risk level of the fault arc is evaluated according to the time difference for grading alarms; the Bayesian network processes the uncertainty of data through probability reasoning, and can provide a transparent decision-making process explanation compared with the commonly used neural network method, and is more suitable for dealing with the dependency relationship of multi-dimensional data. The Bayesian method is used to dynamically adjust the alarm weights of different sensors to generate posterior probabilities, and the association between the data of each sensor is analyzed through the conditional probability table (CPT) to achieve the purpose of locating the fault arc.

[0034] 3. Optimize the machine learning model according to historical data. This mechanism enables the system to continuously adapt to the changes in the electrical environment over time, automatically adjust the alarm threshold and optimize the detection model, and adaptively adjust the model structure according to the actual performance to ensure that the accuracy of the new model is always better than that of the old model. After synchronously uploading the data to the upper computer, the historical data stored locally by the Raspberry Pi is used for regular self-learning and model optimization to ensure timely warning of potential electrical fire risks and maintain high-efficiency and accurate detection performance. BRIEF DESCRIPTION OF THE DRAWINGS

[0035] Figure 1 This is the fault arc characteristic test system of the present invention.

[0036] Figure 2 This is the schematic flow chart of the fault arc positioning detection and electrical fire warning method based on the Raspberry Pi of the present invention. DETAILED DESCRIPTION OF THE INVENTION

[0037] The technical solution of the present invention will be specifically described below in conjunction with the accompanying drawings.

[0038] The present invention provides a method for fault arc location detection and electrical fire warning based on Raspberry Pi, including:

[0039] Using Raspberry Pi as the core processing unit, integrating multiple types of sensors, collecting various AC and DC load data, and establishing a typical multi-dimensional fusion fault arc database;

[0040] Taking the acquired data as input, deeply analyzing the multi-time-frequency characteristics and wavelet transform characteristics of the signal in combination with a residual neural network, comprehensively capturing the characteristic information of the fault arc, and realizing universal fault arc detection for AC and DC;

[0041] Using the time difference of the fault arc detected by multiple sensors to evaluate the risk level of the fault arc for hierarchical alarm, and using a Bayesian network to reason and judge different fault modes to achieve precise positioning of the fault arc;

[0042] By working in coordination with the cloud host, building a remote monitoring and management platform, and regularly performing self-learning and model updating using the historical data stored locally.

[0043] The following is the specific implementation process of the present invention.

[0044] (1) Build an AC / DC fault arc detection system based on Raspberry Pi

[0045] With the progress of technology and the rapid development of renewable energy systems, more and more systems involve the mixed use of AC and DC, which promotes the increasing demand in the market for universal AC / DC fault arc detection devices. There are significant differences between DC arcs and AC arcs in electrical systems. AC arcs have zero-crossing points and are relatively easy to extinguish, while DC arcs are more persistent and stable and are more likely to cause more serious consequences. To enable Raspberry Pi to better distinguish the waveform characteristics during AC / DC mixing, first, multiple types of sensors are installed to obtain current, voltage, smoke, and temperature waveform data of various AC and DC loads through a fault arc characteristic test system device (as Figure 1 shown), perform signal preprocessing such as noise filtering and normalization on the collected waveform data, and establish a multi-dimensional fusion fault arc database.

[0046] Apply the fast Fourier transform (FFT) to the preprocessed signal to convert the signal into the frequency domain, analyze its frequency components to identify the periodic changes characteristic of AC arcs, and then obtain the power distribution of the signal by calculating the power spectral density (PSD), which helps to identify the energy characteristics of the arc. Based on the FFT and PSD results, extract features such as the main frequency components, energy peaks, spectral entropy, etc. that are helpful for distinguishing DC and AC arcs. These features are crucial for subsequent pattern recognition and signal classification. In addition, use wavelet transform for time-frequency analysis to provide the frequency and energy changes of the signal at different time points, further enhancing the dimension and accuracy of feature extraction.

[0047] Use the processed time-frequency domain features to train a residual neural network machine learning model to detect and classify fault arc signals. The residual neural network can effectively extract features from complex signals through multiple layers of convolution and residual connections, and is connected with an attention mechanism channel, further improving the accuracy of identifying fault arcs. The trained model is deployed to the Raspberry Pi to complete the hardware module, automatically distinguishing whether the detected arc is DC or AC, and thus taking appropriate protection measures accordingly.

[0048] (2) Time difference grading alarm and Bayesian network inference positioning (as Figure 2 shown)

[0049] Arrange points according to the risk level of the occurrence of fault arcs. Install current sensors and voltage sensors between the key access points and loads of the circuit, and install multiple sensors in the areas where fault arcs are most likely to occur for multi-dimensional detection, and reasonably allocate to improve the detection accuracy of the system. When a fault arc occurs, it will be accompanied by a series of physical phenomena such as smoke, strong light, flame, and even sometimes noise. The appearance order and characteristics of these phenomena can provide important information for the detection and treatment of arcs. The fault arc will first generate strong light radiation, and as the arc continues, the high temperature will cause surrounding materials (such as wire insulation layers, plastic parts, etc.) to start burning, generating smoke and flame. The smoke is generated due to the burning materials, while the flame marks the intensification of the combustion process.

[0050] The trained residual neural network model is transplanted onto a Raspberry Pi. Based on historical arc fault data and real-time environmental data, the Raspberry Pi is equipped with multiple sensors to receive and analyze signals, and multi-level alarms are carried out through the time difference of fault arcs detected by different sensors. When the current or voltage sensor first detects an anomaly, such as overload or short circuit, while the temperature, smoke, and light sensors do not report anomalies, the system will trigger a primary alarm. This level usually means potential electrical problems that require immediate inspection and maintenance, but the fire risk has not yet emerged; if the temperature sensor also reports a significant increase after the current or voltage anomaly, this indicates overheating, which may develop into more serious problems. At this time, the system will upgrade the alarm to secondary, requiring more urgent technical intervention to prevent the problem from worsening; when the light sensor detects strong light emission and the smoke sensor also starts to report smoke accumulation, it indicates that a fire may have broken out or is in progress. In this case, the system will trigger a high-level alarm, activate the automatic fire extinguishing system, and notify the fire department and emergency response team for intervention.

[0051] After feature learning for each sensor data based on the residual neural network training model, an initial weight is set for each sensor as the prior probability P(θ) of the Bayesian network, where θ represents the occurrence of a fault arc. The system sets the multi-dimensional observation dataset of the sensor by real-time detecting sensor data as: D = {I(t), V(t), T(t), S(t), G(t)}, where I is the current, V is the voltage, T is the temperature, S is the smoke, G is the light, and t is the time. D is the waveform dataset collected by sensors such as current, voltage, temperature, smoke, and light for time series analysis. Based on the multi-dimensional observation dataset, a conditional probability distribution table (CPT) is established for each sensor node to describe the probability distribution P(D|θ) of the sensor observation data in the fault state. When some sensors detect relevant events, the Bayesian network infers the probability distribution of other location-related variables that are not directly observed using the known observation data. As the system continuously updates the collected signals, the posterior probability of each sensor is updated using Bayes' formula:

[0052]

[0053] where P(θ|D) is the posterior probability, P(D|θ) is the likelihood function, and P(D) is the normalization term. The region with the maximum posterior probability is selected as the region where the fault arc is most likely to occur. Subsequently, the weights of the sensors are adjusted according to user feedback and the actual situation of the alarm. If an alarm is a false alarm, the weight of the corresponding sensor is reduced; if the alarm is valid, the weight of its sensor is increased.

[0054] Fault location achieved by using sensor time difference hierarchical alarm and Bayesian network ensures the flexible response of the system to electrical faults, helps to detect the signs of electrical fires at an early stage and quickly take measures to repair the faults to prevent fires from occurring.

[0055] (3) Regular optimization of the model

[0056] Using Raspberry Pi as the control unit, combined with multiple sensors, visualizes all data and alarm information on the host computer. At the same time, this platform can communicate with the electrical fire warning device to determine the location of the abnormality and send warning information to relevant personnel in the form of information push, reminding relevant personnel to take corresponding measures in time.

[0057] During the real-time monitoring process, Raspberry Pi stores information such as sensor data, alarm records, fault characteristics, and model output results collected in the local SQLite database. Raspberry Pi regularly sorts and filters the locally stored data, and filters out representative fault data for model training and optimization. The system cleans up expired and no longer useful data regularly according to the set strategy to save storage space and keep the database running efficiently.

[0058] The updated model may have a problem of decreasing accuracy due to new data coming in. To ensure that the new model is always better than the old model, the model adaptively adjusts the model structure according to the actual performance, dynamically controls the regional complexity through a formula, and prevents overfitting:

[0059]

[0060] where α is an adjustment factor that controls the model complexity, representing the structural complexity of the model (such as the number of layers and the number of parameters); θ new represents the updated model parameters. During the model optimization process, the parameters θ are adjusted to balance the performance and complexity of the model; model refers to the machine learning model; complexity is a quantitative index of the model complexity, used to control overfitting; is the partial derivative of the loss function with respect to the model complexity, used to dynamically adjust the complexity.

[0061] Finally, model verification and deployment are carried out, and the validation set D val is used to calculate the accuracy Accuracy of the model:

[0062]

[0063] where θ is the model function, f ResNet(X; θ) Output of the residual neural network, where X is the input feature vector, Y is the true label, and 1[·] is the indicator function, which is 1 when the predicted value is equal to the true value and 0 otherwise. When the new model passes the performance evaluation in the validation set and all indicators are better than the old model, safety and stability tests are conducted to ensure the reliability of the model under various boundary conditions. After passing the tests, the new model is deployed into the system, using θ deployed = θ new As the model parameters for the final application, the new model will be applied to the subsequent fault detection process. At the same time, the old model will be retained as a backup so that it can be quickly rolled back in case of problems with the new model, ensuring the overall stability and reliability of the system. Through model update, the system can continuously improve the accuracy of fault arc detection and the reliability of electrical fire warning, avoiding false alarms or missed alarms caused by environmental changes.

[0064] The present invention also provides a computer-readable storage medium, on which computer program instructions capable of being run by a processor are stored. When the processor runs the computer program instructions, the method steps as described in any one of the above can be implemented.

[0065] The above are the preferred embodiments of the present invention. All changes made according to the technical solutions of the present invention, when the functions and effects produced do not exceed the scope of the technical solutions of the present invention, fall within the protection scope of the present invention.

Claims

1. A fault arc location detection and electrical fire warning method based on Raspberry Pi, characterized in that, Including: Using Raspberry Pi as the core processing unit, integrating multiple types of sensors, collecting various AC and DC load data, and establishing a typical multi-dimensional fusion fault arc database; Taking the acquired data as input, combining with the residual neural network to deeply analyze the multi-time-frequency characteristics and wavelet transform characteristics of the signal, comprehensively capturing the characteristic information of the fault arc, and realizing the fault arc detection that is common for both AC and DC; Using the time difference detected by multiple sensors for the fault arc to evaluate the danger level of the fault arc for hierarchical alarm, and using the Bayesian network to reason and judge different fault modes to achieve the precise positioning of the fault arc; Through collaborative work with the cloud host, building a remote monitoring and management platform, and regularly using the historical data stored locally for self-learning and model update.

2. The method for fault arc location detection and electrical fire warning based on Raspberry Pi according to claim 1, characterized in that, Using Raspberry Pi as the core processing unit, integrating multiple types of sensors, collecting various AC and DC load data, and establishing a typical multi-dimensional data waveform library. Specifically, Raspberry Pi is used as the core processing unit, equipped with multiple types of sensors to generate arcs through the fault arc characteristic test system device to obtain the current, voltage, smoke, and temperature waveform data of various AC and DC loads, and performing signal preprocessing including noise filtering and normalization on the collected waveform data to establish a multi-dimensional fusion fault arc database.

3. A method for fault arc location detection and electrical fire warning based on Raspberry Pi according to claim 1, characterized in that, Taking the acquired data as input, combining with the residual neural network to deeply analyze the multi-time-frequency characteristics and wavelet transform characteristics of the signal, comprehensively capturing the characteristic information of the fault arc, and realizing the fault arc detection that is common for both AC and DC. The specific implementation is as follows: Applying the fast Fourier transform (FFT) to the signals in the multi-dimensional fusion fault arc database to convert the signals to the frequency domain, analyzing its frequency components to identify the periodic changes unique to AC arcs, and then obtaining the power distribution of the signals by calculating the power spectral density (PSD); based on the FFT and PSD results, extracting the features that help distinguish DC and AC arcs, including the main frequency components, energy peaks, and spectral entropy; Using wavelet transform for time-frequency analysis to provide the frequency and energy changes of the signal at different time points; Using the processed time-frequency domain features to train the residual neural network to detect and classify the fault arc signals. The residual neural network can effectively extract the features in complex signals through multi-layer convolution and residual connection, and connecting the attention mechanism channels, further improving the accuracy of identifying fault arcs; Deploying the trained residual neural network into Raspberry Pi to complete the hardware module, automatically distinguishing whether the detected arc is DC or AC to take corresponding protection measures.

4. A method for fault arc location detection and electrical fire warning based on Raspberry Pi according to claim 1, characterized in that, Using the time difference detected by multiple sensors for the fault arc to evaluate the danger level of the fault arc for hierarchical alarm. The specific implementation is as follows: Arranging points according to the danger degree of the occurrence of the fault arc, installing current sensors and voltage sensors between the key access points and loads of the circuit, and installing multiple sensors in the area where the fault arc is most likely to occur for multi-dimensional detection; The trained residual neural network is transplanted onto the Raspberry Pi. The Raspberry Pi, based on historical arc fault data and real-time environmental data, is equipped with multiple sensors to receive and analyze signals, and conducts multi-level alarms based on the time difference of fault arcs detected by different sensors. When the current sensor or voltage sensor first detects an anomaly while the temperature sensor, smoke sensor, and light sensor do not report any anomalies, a primary alarm will be triggered. If, after the current or voltage anomaly, the temperature sensor also reports a significant increase, indicating overheating, the alarm will be upgraded to a secondary level, requiring more urgent technical intervention to prevent the problem from worsening. When the light sensor detects strong light emission and the smoke sensor also starts reporting smoke accumulation, it indicates that a fire may have broken out or is in progress. In this case, a high-level alarm is triggered, the automatic fire extinguishing system is activated, and the fire department and emergency response team are notified for intervention.

5. A fault arc location detection and electrical fire warning method based on Raspberry Pi according to claim 1, characterized in that, Use the Bayesian network to reason and judge different fault modes to achieve precise location of the fault arc. The specific implementation is as follows: After feature learning of each sensor data based on the residual neural network, an initial weight is set for each sensor as the prior probability P(θ) of the Bayesian network, where θ represents the occurrence of the fault arc. By real-time detecting the sensor data, the multi-dimensional observation data set of the sensor is set as: D = {I(t), V(t), T(t), S(t), G(t)}, where I is the current, V is the voltage, T is the temperature, S is the smoke, G is the light, and t is the time. D is the waveform data set collected by the current sensor, voltage sensor, temperature sensor, smoke sensor, and light sensor for time series analysis. Based on the multi-dimensional observation data set, a conditional probability distribution table CPT is established for each sensor node to describe the probability distribution P(D|θ) of the sensor observation data in the fault state. When some sensors detect relevant events, the Bayesian network uses the known observation data to infer the probability distribution of other location-related variables that are not directly observed. As the collected signals are continuously updated, the posterior probability of each sensor is updated using Bayes' formula, and the region with the maximum posterior probability is selected as the region where the fault arc is most likely to occur. Subsequently, according to user feedback and the actual situation of the alarm, the weights of the sensors are adjusted. If an alarm is a false alarm, the weight of the corresponding sensor is decreased; if the alarm is valid, the weight of its sensor is increased.

6. The method for detecting and locating faulty electric arcs and warning of electrical fires based on Raspberry Pi according to claim 5, characterized in that, The formula for updating the posterior probability of each sensor using Bayes' formula is as follows: Where P(θ|D) is the posterior probability, P(D|θ) is the likelihood function, and P(D) is the normalization term.

7. A method for fault arc location detection and electrical fire warning based on Raspberry Pi according to claim 1, characterized in that, During the real-time monitoring process, the Raspberry Pi stores information such as the collected sensor data, alarm records, fault characteristics, and model output results in the local SQLite database. The Raspberry Pi regularly organizes and filters the locally stored data, and selects representative fault data for model training and optimization. The system regularly clears expired and no longer useful data according to the set strategy to save storage space and maintain the efficient operation of the database.

8. A method for fault arc location detection and electrical fire warning based on Raspberry Pi according to claim 1, characterized in that, After the model is updated, the updated model may have a problem of decreased accuracy due to new data coming in. To ensure that the new model is always better than the old model, the model adaptively adjusts the model structure according to the actual performance, dynamically controls the regional complexity through formulas, and prevents overfitting: where α is an adjustment factor that controls the model complexity, representing the structural complexity of the model; θ new represents the updated model parameters. During the model optimization process, the parameters θ are adjusted to balance the performance and complexity of the model; model refers to the machine learning model; Complexity is a quantitative metric for model complexity, used to control overfitting; is the partial derivative of the loss function with respect to model complexity, used to dynamically adjust the complexity.

9. A method for fault arc location detection and electrical fire warning based on Raspberry Pi according to claim 1, characterized in that, After model training or update, model validation and deployment are required, using the validation set D val Calculate the accuracy Accuracy of the model: where θ is the model function, f ResNet (X; θ) is the output of the residual neural network, X is the input feature vector, Y is the true label, 1[·] is the indicator function which is 1 when the predicted value is equal to the true value and 0 otherwise; when the new model passes the performance evaluation in the validation set and all indicators are better than the old model, safety and stability tests are carried out to ensure the reliability of the model under various boundary conditions; after passing the tests, the new model is deployed into the system, using θ deployed = θ new as the model parameters for the final application, the new model will be applied to the subsequent fault detection process; at the same time, the old model will be retained as a backup so that it can be quickly rolled back in case of problems with the new model, ensuring the overall stability and reliability of the system; through model update, the system can continuously improve the accuracy of fault arc detection and the reliability of electrical fire warning, avoiding false alarms or missed alarms caused by environmental changes.

10. A computer-readable storage medium storing computer program instructions executable by a processor, and when the processor runs the computer program instructions, the method steps as described in any one of claims 1-9 can be implemented.

Images