Charging pile fire early warning method and system

Abstract

The invention provides a charging pile fire early warning method and system, and the method comprises the steps: collecting the multi-dimensional monitoring data of a charging pile in real time, and carrying out the preprocessing of the monitoring data; carrying out fast Fourier transform on voltage and current data in the monitoring data to extract frequency characteristics, calculating a current fluctuation coefficient to extract time sequence characteristics, and calculating a temperature change rate; constructing a data set including the time-frequency domain characteristics and the temperature change rate, and outputting fire risk hidden variables by using a deep learning model; and dynamically calculating a fire risk posterior probability based on a Bayesian network, and generating an early warning level in combination with the hidden variables and the time-frequency domain features. Based on the method, the invention further provides a charging pile fire early warning system. According to the invention, the monitoring data of the charging pile and the surrounding environment thereof are collected in real time by using the Internet of Things technology, rapid local data processing is carried out through edge calculation, and the sensitivity of the early warning model is dynamically adjusted by using the adaptive Bayesian algorithm, so that the accuracy and response speed of fire early warning are improved.

Description

Technical Field

[0001] The present invention belongs to the technical field of line live detection, and in particular relates to a charging pile fire early warning method and system. Background Art

[0002] Charging pile fire warning systems are technologies that monitor and warn of potential fire risks in electric vehicle charging facilities. With the global energy transition and rising environmental awareness, the use of electric vehicles (EVs) as a clean energy vehicle has increased dramatically. At the same time, charging piles, as the infrastructure for EV charging, are facing increasing safety concerns, particularly regarding fire risk prevention and control. Charging pile fires not only cause significant property damage but can also endanger personal safety. Therefore, the development of an efficient and accurate fire warning system is of paramount importance.

[0003] Existing charging pile fire warning technologies rely on centralized servers to process collected data, which can lead to delays in data processing and warning issuance. Furthermore, insufficient accuracy in fire warning algorithms or sensors can lead to false or missed alerts, impacting the reliability of warnings and user trust. Warnings employ fixed thresholds to determine fire risk, lacking the ability to dynamically adjust to environmental changes. Fire warning product maintenance requires specialized knowledge, and timely troubleshooting and repairs can be costly and complex in the event of a malfunction. Summary of the Invention

[0004] In order to solve the above technical problems, the present invention proposes a charging pile fire warning method and system, which can effectively improve the charging safety of electric vehicles, reduce fire risks, and is of great significance to the safe management of charging infrastructure.

[0005] To achieve the above object, the present invention adopts the following technical solutions: A charging pile fire warning method includes the following steps: Real-time collection of multi-dimensional monitoring data of charging piles and pre-processing of monitoring data; The voltage and current data in the monitoring data are all subjected to fast Fourier transform to extract frequency characteristics, calculate the current fluctuation coefficient to extract time series characteristics, and calculate the temperature change rate; Construct a data set including time-frequency domain features and temperature change rate, and use a deep learning model to output fire risk latent variables; The fire risk posterior probability is dynamically calculated based on the Bayesian network, and the warning level is generated by combining the latent variables and time-frequency domain features.

[0006] Furthermore, the process of collecting multi-dimensional monitoring data of charging piles in real time includes: Use the voltage sensor to obtain the voltage data of the charging pile; Use current sensors to obtain current data of charging piles; Use smoke sensors to obtain smoke data around charging piles; The temperature sensor obtains temperature data around the charging pile.

[0007] Furthermore, the voltage sensor adopts a through-hole design and is connected in series with the main charging circuit; The current sensor adopts a through-hole design and is connected in series with the main charging circuit; The smoke sensor is installed on the top of the charging pile cavity to preferentially capture rising smoke particles; The temperature sensor is mounted on the power module and cable connector inside the charging pile and other places prone to heat.

[0008] Furthermore, the process of preprocessing the monitoring data includes using the Kalman filter algorithm to perform denoising on the monitoring data. The prediction equation for denoising is:

[0009] The observation update equation is:

[0010] For the The prior state estimate at time t; in, For the The posterior state estimate at the moment, that is, the data after denoising; For the The observation value at the moment, that is, the data before denoising; is the observation matrix; is the Kalman gain matrix; is the state transfer matrix; is the control input matrix; Normalize the denoised data: ; in, is the normalized data point; is the original data point; is the mean; is the standard deviation.

[0011] Furthermore, the process of performing fast Fourier transform on the voltage and current data in the monitoring data to extract frequency features includes: Perform Fourier transform on the current and voltage signals to obtain the spectrum, calculate the power spectrum density, and extract the frequency eigenvector; The power spectral density is expressed as: ; in, The frequency is The power spectral density at ; Frequency domain data points after discrete Fourier transform; is the number of FFT points.

[0012] Furthermore, the process of calculating the current fluctuation coefficient and extracting the timing characteristics includes: ; in, is the current fluctuation coefficient; is the current sequence within the sliding window.

[0013] Furthermore, the process of constructing a dataset including time-frequency domain features and temperature change rate and using a deep learning model to output the fire risk latent variables includes: The data set received by the input layer is represented as: ; in, For the The input feature vector at time t; is the normalized temperature value; is the normalized current value; is the normalized voltage value; for The temperature change rate at each moment; for Current fluctuation coefficient at each moment; Through the Gate of Oblivion ; Input Gate ; and output gate ; in, For the The output vector of the forget gate at time t; is the output vector of the input gate; For the The output vector of the output gate at time t; is the Sigmoid activation function; is the first weight matrix from the input layer to the gating unit; is the second weight matrix from the input layer to the gating unit; is the third weight matrix from the input layer to the gating unit; is the first weight matrix from the hidden layer to the gating unit; is the second weight matrix from the hidden layer to the gating unit; is the third weight matrix from the hidden layer to the gating unit; is the first bias vector of the gating unit; is the second bias vector of the gate control unit; is the third bias vector of the gate control unit; for The hidden layer state vector at time t; The output layer outputs the fire risk probability distribution through the fully connected layer and the softmax function:

[0014] in, For the Fire risk probability distribution vector predicted by the moment model; is the weight matrix from the hidden layer to the output layer; is the bias vector of the output layer.

[0015] Furthermore, the process of dynamically calculating the posterior probability of fire risk based on the Bayesian network includes: Defining a node collection They represent temperature anomaly, current anomaly, smoke density and fire risk respectively; Constructing a conditional probability table , by Bayes' theorem Perform reasoning; wherein: is the posterior probability of fire occurring under known temperature, current, and smoke conditions; is the joint conditional probability of observing a specific temperature, current, and smoke state when a fire occurs; is the prior probability of fire occurrence; is the joint probability of observing a specific temperature, current, and smoke state; Use Naive Bayes assumption to simplify calculations And through maximum a posteriori estimation Determine the risk level; including: is the conditional probability that the temperature is in a specific state when the fire occurs; is the conditional probability that the current is in a specific state when the fire occurs; is the conditional probability that smoke is in a specific state when a fire occurs; is the maximum probability fire risk level predicted based on observation data.

[0016] Furthermore, the process of generating a warning level by combining the latent variables and the time-frequency domain features includes: Output the deep learning model and the Bayesian network output Perform weighted fusion, and the calculation formula is: ; in: The integrated fire risk score after fusion; Fire risk score output by the deep learning model; Fire risk score output by the Bayesian network; is the weight coefficient, and its value range is , used to adjust the contribution of the two models; According to the fusion results With preset threshold The fire risk level is judged to be low ,middle or high .

[0017] The present invention also proposes a charging pile fire warning system, comprising: a data acquisition module, a feature extraction module, a prediction module and a warning module; The data acquisition module is used to collect multi-dimensional monitoring data of the charging pile in real time and pre-process the monitoring data; The feature extraction module is used to perform fast Fourier transform on the voltage and current data in the monitoring data to extract frequency features, calculate the current fluctuation coefficient to extract time series features, and calculate the temperature change rate; The prediction module is used to construct a data set including time-frequency domain features and temperature change rate, and output fire risk latent variables using a deep learning model; The early warning module is used to dynamically calculate the posterior probability of fire risk based on the Bayesian network, and generate an early warning level by combining the latent variables and time-frequency domain features.

[0018] The effects provided in the summary of the invention are only the effects of the embodiments, not all the effects of the invention. One of the above technical solutions has the following advantages or beneficial effects: The present invention proposes a charging pile fire warning method and system, which includes the following steps: real-time collection of multi-dimensional monitoring data of the charging pile, and pre-processing of the monitoring data; fast Fourier transform of the voltage and current data in the monitoring data to extract frequency characteristics, calculate the current fluctuation coefficient to extract time series characteristics, and calculate the temperature change rate; construct a data set including time-frequency domain characteristics and temperature change rate, and use a deep learning model to output the fire risk latent variable; dynamically calculate the posterior probability of the fire risk based on the Bayesian network, and generate the warning level by combining the latent variable and time-frequency domain characteristics. Based on the charging pile fire warning method, a charging pile fire warning system is also proposed. The present invention can effectively improve the charging safety of electric vehicles, reduce fire risks, and is of great significance to the safety management of charging infrastructure.

[0019] The present invention uses Internet of Things technology to collect real-time monitoring data of charging piles and their surrounding environment, performs rapid local data processing through edge computing, and adopts an adaptive Bayesian algorithm to dynamically adjust the sensitivity of the early warning model to improve the accuracy and response speed of fire warnings. BRIEF DESCRIPTION OF THE DRAWINGS

[0020] Figure 1 This is a flow chart of a charging pile fire warning method proposed in Example 1 of the present invention; Figure 2 This is a schematic diagram of a charging pile fire warning system proposed in Example 2 of the present invention. DETAILED DESCRIPTION

[0021] In order to clearly illustrate the technical features of this solution, the present invention is described in detail below through specific implementation methods and in conjunction with the accompanying drawings. The disclosure below provides many different embodiments or examples for realizing different structures of the present invention. In order to simplify the disclosure of the present invention, the components and settings of specific examples are described below. In addition, the present invention may repeat reference numbers and / or letters in different examples. This repetition is for the purpose of simplicity and clarity and does not itself indicate the relationship between the various embodiments and / or settings discussed. It should be noted that the components illustrated in the accompanying drawings are not necessarily drawn to scale. The present invention omits descriptions of well-known components and processing technologies and processes to avoid unnecessary limitations on the present invention.

[0022] Example 1 Embodiment 1 of the present invention proposes a charging pile fire warning method, which is used to solve the technical problems existing in the charging pile fire warning technology in the prior art.

[0023] The specific process of a charging pile fire warning method proposed in Example 1 of the present invention includes: Data preprocessing: Kalman filtering is used to denoise the raw sensor data, and the data estimate is iteratively optimized through the state prediction equation and observation update equation. Z-score normalization is then performed to eliminate the dimensional impact of different sensor data.

[0024] Feature extraction: Perform fast Fourier transform (FFT) on current and voltage data, calculate power spectrum density, and extract frequency domain features; simultaneously calculate temperature change rate and current fluctuation coefficient to extract timing features.

[0025] Deep learning model: An LSTM network is used to analyze the preprocessed multidimensional feature vector, and the unit state is updated through the forget gate, input gate, and output gate to output the fire risk probability distribution.

[0026] Bayesian network: Construct a Bayesian network that includes temperature anomalies, current anomalies, smoke concentration, and fire risk. Use Bayesian theorem for reasoning and adopt the Naive Bayes hypothesis to simplify calculations and determine the risk level.

[0027] Fusion decision-making: The deep learning model output and the Bayesian network output are weightedly fused, and the fire risk level is determined and the corresponding warning is triggered based on the comparison of the fusion result with the preset threshold.

[0028] Figure 1 This is a flow chart of a charging pile fire warning method proposed in Example 1 of the present invention.

[0029] In step S100, multi-dimensional monitoring data of the charging pile is collected in real time, and the monitoring data is pre-processed; The voltage sensor is used to obtain the voltage data of the charging pile; the current sensor is used to obtain the current data of the charging pile; the smoke sensor is used to obtain the smoke data around the charging pile; and the temperature sensor is used to obtain the temperature data around the charging pile.

[0030] The voltage sensor adopts a through-hole design and is connected in series with the main charging circuit; the current sensor adopts a through-hole design and is connected in series with the main charging circuit; the smoke sensor is installed on the top of the charging pile cavity to preferentially capture rising smoke particles; the temperature sensor is mounted on the power module and cable connector inside the charging pile and other heat-prone parts.

[0031] The process of preprocessing the monitoring data includes using the Kalman filter algorithm to denoise the monitoring data. The prediction equation for denoising is:

[0032] The observation update equation is:

[0033] For the The prior state estimate at time t; in, For the The posterior state estimate at the moment, that is, the data after denoising; For the The observation value at the moment, that is, the data before denoising; is the observation matrix; is the Kalman gain matrix; is the state transfer matrix; is the control input matrix; Normalize the denoised data: ; in, is the normalized data point; is the original data point; is the mean; is the standard deviation.

[0034] The present invention realizes comprehensive monitoring of the operating status of the charging pile and improves the accuracy of early warning by fusing multiple sensors such as temperature, smoke, current and voltage.

[0035] In step S200, the voltage and current data in the monitoring data are subjected to fast Fourier transform to extract frequency characteristics, the current fluctuation coefficient is calculated to extract time series characteristics, and the temperature change rate is calculated; The process of performing fast Fourier transform on the voltage and current data in the monitoring data to extract the frequency characteristics includes: performing Fourier transform on the current and voltage signals to obtain the spectrum, calculating the power spectrum density, and extracting the frequency characteristic vector; The power spectral density is expressed as: ; in, The frequency is The power spectral density at ; Frequency domain data points after discrete Fourier transform; is the number of FFT points.

[0036] The process of calculating the current fluctuation coefficient and extracting the timing characteristics includes: ; in, is the current fluctuation coefficient; is the current sequence within the sliding window.

[0037] In step S300, a data set including time-frequency domain features and temperature change rate is constructed, and a fire risk latent variable is output using a deep learning model; The process of constructing a dataset including time-frequency domain features and temperature change rate and outputting fire risk latent variables using a deep learning model includes: The data set received by the input layer is represented as: ; in, For the The input feature vector at time t; is the normalized temperature value; is the normalized current value; is the normalized voltage value; for The temperature change rate at each moment; for Current fluctuation coefficient at each moment; Through the Gate of Oblivion ; Input Gate ; and output gate ; in, For the The output vector of the forget gate at time t; is the output vector of the input gate; For the The output vector of the output gate at time t; is the Sigmoid activation function; is the first weight matrix from the input layer to the gating unit; is the second weight matrix from the input layer to the gating unit; is the third weight matrix from the input layer to the gating unit; is the first weight matrix from the hidden layer to the gating unit; is the second weight matrix from the hidden layer to the gating unit; is the third weight matrix from the hidden layer to the gating unit; is the first bias vector of the gating unit; is the second bias vector of the gate control unit; is the third bias vector of the gate control unit; for The hidden layer state vector at time t; The output layer outputs the fire risk probability distribution through the fully connected layer and the softmax function:

[0038] in, For the Fire risk probability distribution vector predicted by the moment model; is the weight matrix from the hidden layer to the output layer; is the bias vector of the output layer.

[0039] The deep learning model is trained using the Adam optimization algorithm, and the parameter update formula is: ; in, For the Model parameters after iteration update; For the Model parameters after iteration update; is the learning rate, which controls the parameter update step size; For the First-order moment estimate of the gradient at that moment (after bias correction); For the Second moment estimate of the moment gradient (after bias correction); To prevent the denominator from being zero, a very small constant and ; in, For the First moment estimate of the gradient at that moment (uncorrected); For the Second moment estimate of the moment gradient (uncorrected); For the The gradient vector calculated at the moment; is the exponential decay rate of the first-order moment estimate, is the exponential decay rate of the second-order moment estimate (usually , ).

[0040] The loss function uses cross entropy loss:

[0041] is the loss function value; Total number of training samples; The total number of categories used to classify fire risk; For samples Belong to category The true label (one-hot encoding); Predict samples for the model Belong to category probability.

[0042] In step S400, the fire risk posterior probability is dynamically calculated based on the Bayesian network, and the warning level is generated by combining the latent variables and time-frequency domain features.

[0043] The process of dynamically calculating the posterior probability of fire risk based on Bayesian networks includes: Defining a node collection They represent temperature anomaly, current anomaly, smoke density and fire risk respectively; Constructing a conditional probability table , by Bayes' theorem Perform reasoning; wherein: is the posterior probability of fire occurring under known temperature, current, and smoke conditions; is the joint conditional probability of observing a specific temperature, current, and smoke state when a fire occurs; is the prior probability of fire occurrence; is the joint probability of observing a specific temperature, current, and smoke state; Use Naive Bayes assumption to simplify calculations And through maximum a posteriori estimation Determine the risk level; including: is the conditional probability that the temperature is in a specific state when the fire occurs; is the conditional probability that the current is in a specific state when the fire occurs; is the conditional probability that smoke is in a specific state when a fire occurs; is the maximum probability fire risk level predicted based on observation data.

[0044] The parameters of the Bayesian network are learned by maximum likelihood estimation, and the calculation formula is: ; in, For the parent node The value is Under the condition of The probability estimate of ; is the parent node in the training data The value is And the child node value is The number of samples; is the Dirichlet prior parameter, used for smoothing probability estimation; For all possible values of child nodes Perform a sum operation.

[0045] The process of generating the warning level by combining the latent variables and the time-frequency domain features includes: Output the deep learning model and the Bayesian network output Perform weighted fusion, and the calculation formula is: ; in: The integrated fire risk score after fusion; Fire risk score output by the deep learning model; Fire risk score output by the Bayesian network; is the weight coefficient, and its value range is , which is used to adjust the contribution of the two models.

[0046] According to the fusion results With preset threshold The fire risk level is judged to be low ,middle or high .

[0047] The first embodiment of the present invention proposes a charging pile fire warning method, employing an architecture comprising a sensor layer, a data transmission layer, a data processing layer, and a warning decision layer. The sensor layer collects data such as the charging pile's temperature, smoke, current, and voltage; the data transmission layer transmits the sensor data to the data processing layer; the data processing layer performs data preprocessing, feature extraction, and risk assessment; and the warning decision layer issues corresponding warning information based on the assessment results.

[0048] A charging pile fire warning method proposed in Example 1 of the present invention, combined with a deep learning model and a Bayesian network, can automatically learn normal and abnormal patterns during the charging process and achieve accurate assessment of fire risks.

[0049] A charging pile fire warning method proposed in Example 1 of the present invention can issue a timely warning in the early stages of a fire (such as overheating, electric arcing, etc.) through real-time analysis and feature extraction of sensor data, thereby buying time for taking measures.

[0050] The charging pile fire warning method proposed in Example 1 of the present invention can continuously optimize model parameters according to actual operation data, adapt to the characteristics of different charging piles and environmental changes, and improve the adaptability and reliability of the warning system.

[0051] To illustrate the implementation of the above method, a charging station was used as an example. During normal charging, the temperature sensor detected a slow rise in the power module temperature from 25°C to 60°C. Simultaneously, the current sensor detected an increase in the current fluctuation coefficient, and FFT analysis showed an increase in high-frequency components. The deep learning model predicted a fire risk probability of 0.35, while the Bayesian network assessed the risk level as moderate. After fusion decision-making, the fire risk index reached the warning threshold, and the system issued a medium-level warning. Maintenance personnel discovered a fault in the power module's cooling fan, which they promptly replaced to prevent further escalation.

[0052] At another charging station, a smoke sensor detected slight smoke during charging, the temperature rapidly rose to 85°C, and the current exhibited abnormal fluctuations. A deep learning model predicted a fire risk probability of 0.92, while a Bayesian network assessed the risk level as high. After integrating the decision-making process, the system immediately issued a high-level warning and automatically disconnected the charging circuit. On-site inspection revealed an internal short circuit in the charging cable, generating an arc and causing the smoke. Thanks to the timely warning, no serious consequences were reported.

[0053] Example 2 Based on the charging pile fire warning method proposed in Example 1 of the present invention, Example 2 of the present invention further proposes a charging pile fire warning system. Figure 2 This is a schematic diagram of a charging pile fire warning system proposed in Example 2 of the present invention, which includes: a data acquisition module, a feature extraction module, a prediction module, and an early warning module; The data acquisition module is used to collect multi-dimensional monitoring data of charging piles in real time and pre-process the monitoring data; The feature extraction module is used to perform fast Fourier transform on the voltage and current data in the monitoring data to extract frequency features, calculate the current fluctuation coefficient to extract time series features, and calculate the temperature change rate; The prediction module is used to construct a data set including time-frequency domain features and temperature change rate, and output the fire risk latent variable using a deep learning model; The early warning module is used to dynamically calculate the posterior probability of fire risk based on the Bayesian network, and generate an early warning level by combining the latent variables and time-frequency domain features.

[0054] The process of implementing the data acquisition module includes: using the voltage sensor to obtain the voltage data of the charging pile; The current sensor is used to obtain the current data of the charging pile; the smoke sensor is used to obtain the smoke data around the charging pile; and the temperature sensor is used to obtain the temperature data around the charging pile.

[0055] The voltage sensor adopts a through-hole design and is connected in series with the main charging circuit; the current sensor adopts a through-hole design and is connected in series with the main charging circuit; the smoke sensor is installed on the top of the charging pile cavity to preferentially capture rising smoke particles; the temperature sensor is mounted on the power module and cable connector inside the charging pile and other heat-prone parts.

[0056] The process of preprocessing the monitoring data includes using the Kalman filter algorithm to denoise the monitoring data. The prediction equation for denoising is:

[0057] The observation update equation is:

[0058] For the The prior state estimate at time t; in, For the The posterior state estimate at the moment, that is, the data after denoising; For the The observation value at the moment, that is, the data before denoising; is the observation matrix; is the Kalman gain matrix; is the state transfer matrix; is the control input matrix; Normalize the denoised data: ; in, is the normalized data point; is the original data point; is the mean; is the standard deviation The process of implementing the feature extraction module includes: Perform Fourier transform on the current and voltage signals to obtain the spectrum, calculate the power spectrum density, and extract the frequency eigenvector; The power spectral density is expressed as: ; in, The frequency is The power spectral density at ; Frequency domain data points after discrete Fourier transform; is the number of FFT points.

[0059] The process of calculating the current fluctuation coefficient and extracting the timing characteristics includes: ; in, is the current fluctuation coefficient; is the current sequence within the sliding window.

[0060] The process of implementing the prediction module includes: The process of constructing a dataset including time-frequency domain features and temperature change rate and outputting fire risk latent variables using a deep learning model includes: The data set received by the input layer is represented as: ; in, For the The input feature vector at time t; is the normalized temperature value; is the normalized current value; is the normalized voltage value; for The temperature change rate at each moment; for Current fluctuation coefficient at each moment; Through the Gate of Oblivion ; Input Gate ; and output gate ; in, For the The output vector of the forget gate at time t; is the output vector of the input gate; For the The output vector of the output gate at time t; is the Sigmoid activation function; is the first weight matrix from the input layer to the gating unit; is the second weight matrix from the input layer to the gating unit; is the third weight matrix from the input layer to the gating unit; is the first weight matrix from the hidden layer to the gating unit; is the second weight matrix from the hidden layer to the gating unit; is the third weight matrix from the hidden layer to the gating unit; is the first bias vector of the gating unit; is the second bias vector of the gate control unit; is the third bias vector of the gate control unit; for The hidden layer state vector at time t; The output layer outputs the fire risk probability distribution through the fully connected layer and the softmax function:

[0061] in, For the Fire risk probability distribution vector predicted by the moment model; is the weight matrix from the hidden layer to the output layer; is the bias vector of the output layer.

[0062] The process of implementing the early warning module includes: Defining a node collection They represent temperature anomaly, current anomaly, smoke density and fire risk respectively; Constructing a conditional probability table , by Bayes' theorem Perform reasoning; wherein: is the posterior probability of fire occurring under known temperature, current, and smoke conditions; is the joint conditional probability of observing a specific temperature, current, and smoke state when a fire occurs; is the prior probability of fire occurrence; is the joint probability of observing a specific temperature, current, and smoke state; Use Naive Bayes assumption to simplify calculations And through maximum a posteriori estimation Determine the risk level; including: is the conditional probability that the temperature is in a specific state when the fire occurs; is the conditional probability that the current is in a specific state when the fire occurs; is the conditional probability that smoke is in a specific state when a fire occurs; is the maximum probability fire risk level predicted based on observation data.

[0063] The process of generating the warning level by combining the latent variables and the time-frequency domain features includes: outputting the deep learning model and the Bayesian network output Perform weighted fusion, and the calculation formula is: ; in: The integrated fire risk score after fusion; Fire risk score output by the deep learning model; Fire risk score output by the Bayesian network; is the weight coefficient, and its value range is , which is used to adjust the contribution of the two models.

[0064] According to the fusion results With preset threshold The fire risk level is judged to be low ,middle or high .

[0065] A charging pile fire warning system proposed in Example 2 of the present invention, combined with a deep learning model and a Bayesian network, can automatically learn normal and abnormal patterns during the charging process and achieve accurate assessment of fire risks.

[0066] A charging pile fire warning system proposed in Example 1 of the present invention can issue a timely warning in the early stages of a fire (such as overheating, electric arcing, etc.) through real-time analysis and feature extraction of sensor data, thereby buying time for taking measures.

[0067] The charging pile fire warning system proposed in Example 1 of the present invention can continuously optimize model parameters according to actual operation data, adapt to the characteristics of different charging piles and environmental changes, and improve the adaptability and reliability of the warning system.

[0068] For the description of the relevant parts of a charging pile fire warning system provided in Example 2 of the present application, please refer to the detailed description of the corresponding parts in a charging pile fire warning method provided in Example 1 of the present application, and will not be repeated here.

[0069] It should be noted that, in this article, relational terms such as first and second, etc. are only used to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply the existence of any such actual relationship or order between these entities or operations. Moreover, the terms "include", "comprise" or any other variants thereof are intended to cover non-exclusive inclusion, so that the process, method, article or device comprising a series of elements are inherent to the elements. In the absence of further restrictions, the elements limited by the statement "comprise one..." do not exclude the presence of other identical elements in the process, method, article or device comprising the elements. In addition, the above-mentioned technical solutions provided in the embodiments of the present application are not described in detail in accordance with the corresponding technical solutions in the prior art to achieve the same principle, so as to avoid excessive elaboration.

[0070] Although the above description is of specific embodiments of the present invention in conjunction with the accompanying drawings, it does not limit the scope of protection of the present invention. For those skilled in the art, other different forms of modifications or variations can be made based on the above description. It is not necessary and impossible to list all embodiments here. Based on the technical solution of the present invention, various modifications or variations that can be made by those skilled in the art without expending creative effort are still within the scope of protection of the present invention.

Claims

1. A charging pile fire warning method, characterized in that: The following steps are involved: Real-time collection of multi-dimensional monitoring data of charging piles and pre-processing of monitoring data; The voltage and current data in the monitoring data are all subjected to fast Fourier transform to extract frequency characteristics, calculate the current fluctuation coefficient to extract time series characteristics, and calculate the temperature change rate; Construct a data set including time-frequency domain features and temperature change rate, and use a deep learning model to output fire risk latent variables; The fire risk posterior probability is dynamically calculated based on the Bayesian network, and the warning level is generated by combining the latent variables and time-frequency domain features.

2. A charging pile fire warning method according to claim 1, characterized in that: The process of real-time collection of multi-dimensional monitoring data of charging piles includes: Use the voltage sensor to obtain the voltage data of the charging pile; Use current sensors to obtain current data of charging piles; Use smoke sensors to obtain smoke data around charging piles; The temperature sensor obtains temperature data around the charging pile.

3. A charging pile fire warning method according to claim 2, characterized in that: The voltage sensor adopts a through-hole design and is connected in series with the main charging circuit; The current sensor adopts a through-hole design and is connected in series with the main charging circuit; The smoke sensor is installed on the top of the charging pile cavity to preferentially capture rising smoke particles; The temperature sensor is mounted on the power module and cable connector inside the charging pile and other places prone to heat.

4. A charging pile fire warning method according to claim 3, characterized in that: The process of preprocessing the monitoring data includes using the Kalman filter algorithm to denoise the monitoring data. The prediction equation for denoising is: The observation update equation is: For the The prior state estimate at time t; in, For the The posterior state estimate at the moment, that is, the data after denoising; For the The observation value at the moment, that is, the data before denoising; is the observation matrix; is the Kalman gain matrix; is the state transfer matrix; is the control input matrix; Normalize the denoised data: ； in, is the normalized data point; is the original data point; is the mean; is the standard deviation.

5. A charging pile fire warning method according to claim 4, characterized in that: The process of extracting frequency features by performing fast Fourier transform on the voltage and current data in the monitoring data includes: Perform Fourier transform on the current and voltage signals to obtain the spectrum, calculate the power spectrum density, and extract the frequency eigenvector; The power spectral density is expressed as: ; in, The frequency is The power spectral density at ; Frequency domain data points after discrete Fourier transform; is the number of FFT points.

6. A charging pile fire warning method according to claim 5, characterized in that: The process of calculating the current fluctuation coefficient and extracting the timing characteristics includes: ； in, is the current fluctuation coefficient; is the current sequence within the sliding window.

7. A charging pile fire warning method according to claim 6, characterized in that: The process of constructing a dataset including time-frequency domain features and temperature change rate and outputting fire risk latent variables using a deep learning model includes: The data set received by the input layer is represented as: ； in, For the The input feature vector at time t; is the normalized temperature value; is the normalized current value; is the normalized voltage value; for The temperature change rate at each moment; for Current fluctuation coefficient at each moment; Through the Gate of Oblivion ; Input Gate ; and output gate ; in, For the The output vector of the forget gate at time t; is the output vector of the input gate; For the The output vector of the output gate at time t; is the Sigmoid activation function; is the first weight matrix from the input layer to the gating unit; is the second weight matrix from the input layer to the gating unit; is the third weight matrix from the input layer to the gating unit; is the first weight matrix from the hidden layer to the gating unit; is the second weight matrix from the hidden layer to the gating unit; is the third weight matrix from the hidden layer to the gating unit; is the first bias vector of the gating unit; is the second bias vector of the gate control unit; is the third bias vector of the gate control unit; for The hidden layer state vector at time t; The output layer outputs the fire risk probability distribution through the fully connected layer and the softmax function: in, For the Fire risk probability distribution vector predicted by the moment model; is the weight matrix from the hidden layer to the output layer; is the bias vector of the output layer.

8. A charging pile fire warning method according to claim 7, characterized in that: The process of dynamically calculating the posterior probability of fire risk based on Bayesian networks includes: Defining a node collection They represent temperature anomaly, current anomaly, smoke density and fire risk respectively; Constructing a conditional probability table , by Bayes' theorem Perform reasoning; wherein: is the posterior probability of fire occurring under known temperature, current, and smoke conditions; is the joint conditional probability of observing a specific temperature, current, and smoke state when a fire occurs; is the prior probability of fire occurrence; is the joint probability of observing a specific temperature, current, and smoke state; Use Naive Bayes assumption to simplify calculations And through maximum a posteriori estimation Determine the risk level; including: is the conditional probability that the temperature is in a specific state when the fire occurs; is the conditional probability that the current is in a specific state when the fire occurs; is the conditional probability that smoke is in a specific state when a fire occurs; is the maximum probability fire risk level predicted based on observation data.

9. A charging pile fire warning method according to claim 8, characterized in that: The process of generating the warning level by combining the latent variables and the time-frequency domain features includes: Output the deep learning model and the Bayesian network output Perform weighted fusion, and the calculation formula is: ; in: The integrated fire risk score after fusion; Fire risk score output by the deep learning model; Fire risk score output by the Bayesian network; is the weight coefficient, and its value range is , used to adjust the contribution of the two models; According to the fusion results With preset threshold The fire risk level is judged to be low ,middle or high .

10. A charging pile fire warning system, characterized in that: include: Data acquisition module, feature extraction module, prediction module and early warning module; The data acquisition module is used to collect multi-dimensional monitoring data of the charging pile in real time and pre-process the monitoring data; The feature extraction module is used to perform fast Fourier transform on the voltage and current data in the monitoring data to extract frequency features, calculate the current fluctuation coefficient to extract time series features, and calculate the temperature change rate; The prediction module is used to construct a data set including time-frequency domain features and temperature change rate, and output fire risk latent variables using a deep learning model; The early warning module is used to dynamically calculate the posterior probability of fire risk based on the Bayesian network, and generate an early warning level by combining the latent variables and time-frequency domain features.

Images