An intelligent fire monitoring system for charging piles

Abstract

The invention discloses an intelligent fire monitoring system for a charging pile, and relates to the fire protection field of the charging pile. The technical scheme thereof includes: a data acquisition unit is used for continuously acquiring a monitoring signal generated by the charging pile within a cycle time; a data processing module pre-processes data transmitted by the data acquisition module to acquire real-time monitoring data; an EEMD model decomposes the real-time monitoring data to acquire a first data set; the first data set is imported into a trained LSTM deep learning model, data with abnormal characteristics are identified, and recorded as abnormal monitoring data; the abnormal monitoring data is imported into a G‑L convolutional neural network model, and a fault type to which the abnormal monitoring data belongs is acquired; an extraction unit calibrates the location of the fault through the charging pile information corresponding to the abnormal monitoring data, so that the fire monitoring and early warning of the charging pile are more efficient and accurate, and the probability of fire accidents can be effectively reduced.

Description

Technical Field

[0001] The present invention relates to the field of fire protection of charging piles, and more specifically, to an intelligent fire protection monitoring system for charging piles. Background Art

[0002] With the growing global demand for energy and the intensification of environmental pollution, all walks of life are implementing the environmental protection concept of sustainable development through innovative research. In order to reduce the negative impact on the environment, electric vehicles, as a means of transportation using clean energy, have received widespread attention and promotion. At the same time, the rapid development of electric vehicles has also posed challenges to the improvement of charging infrastructure. Charging stations play an important role in the promotion of new energy vehicles.

[0003] However, in the process of safety supervision of charging pile stations, fire safety is an important hidden danger, especially for large charging pile stations, where there are a large number of charging piles and a large number of charging tasks, which causes the charging piles to be in a charging working state for a long time, and fire safety problems are very likely to occur, such as fire, explosion and other safety accidents. Therefore, it is extremely important to carry out early warning and monitoring before a safety accident occurs. A fire in a charging pile or a fire accident may be caused by a variety of factors. For example, a fault occurs in the internal electrical components of the charging pile, and long-term high-power charging may cause excessive heat to be generated inside the charging pile. Monitoring and detecting these current changes are key steps to prevent fires in advance. Due to the large number of components and circuits inside the charging pile, if manual inspection and monitoring are carried out, not only will the labor cost increase, but there will also be errors in the inspection, and the work efficiency will be low. Moreover, the analysis of the cause and type of the fault also requires extremely high professional skills of the operator, further leading to an increase in labor costs. In addition, manual inspection is also difficult to accurately warn before an accident occurs, and the warning effect in advance is minimal. Therefore, a fire monitoring system based on the Internet of Things technology can effectively solve the limitations of traditional fire monitoring methods. Therefore, the present invention is innovatively designed to address the above problems. Summary of the invention

[0004] In view of the problems existing in the prior art, the object of the present invention is to provide an intelligent fire monitoring system for charging piles, so as to achieve efficient and accurate monitoring and early warning of charging pile fire.

[0005] To achieve the above object, the present invention provides the following technical solution: an intelligent fire monitoring system for a charging pile, the intelligent fire monitoring system for a charging pile comprising:

[0006] Data acquisition module: including a data acquisition unit and a marking unit. The data acquisition unit is used to continuously acquire the original signal generated by the charging pile within the cycle time. The monitoring signal includes a voltage signal, a current signal and a temperature signal. The marking unit is used to add the corresponding charging pile information to the data acquired by the acquisition unit.

[0007] Storage module: used for data storage;

[0008] Data processing module: pre-processes the data transmitted by the data acquisition module, and records the pre-processed data as real-time monitoring data. The voltage signal, current signal and temperature signal of the real-time monitoring data are decomposed by the EEMD model to obtain a real-time input sequence, which is recorded as the first data set;

[0009] Anomaly identification module: used to identify abnormal data, including LSTM deep learning model, importing the first data set into the trained LSTM deep learning model, identifying data with abnormal characteristics, and recording them as abnormal monitoring data;

[0010] Fault analysis module: used for fault type analysis and fault location location, including GL convolutional neural network model and extraction unit. The abnormal monitoring data is imported into the GL convolutional neural network model to obtain the fault type to which the abnormal monitoring data belongs. The extraction unit calibrates the location of the fault through the charging pile information corresponding to the abnormal monitoring data.

[0011] Display module: includes a human-computer interaction interface for viewing the operating status of the charging pile.

[0012] Preferably, the EEMD model processing includes:

[0013] Set the monitoring signal of real-time monitoring data to ,in, Used to replace voltage signals, current signals and temperature signals, and monitor signals Add white noise during decomposition , forming a new signal sequence:

[0014] ;

[0015] in, =1,……, , is the average number of times, is the label;

[0016] For each Perform EEMD decomposition to obtain: ;

[0017] Calculate the correlation coefficient for each decomposed IMF component:

[0018] ;

[0019] In the formula, represents the EEMD decomposition Order IMF component; represents the residual component of the monitoring signal decomposed by EEMD; Indicates The residual component after EMD decomposition after averaging is express and The covariance of express The variance of express The variance of .

[0020] Preferably, the EEMD model further includes:

[0021] Set a threshold ;

[0022] If the correlation coefficient , it means that the correlation between the IMF component and the monitoring signal meets the expected requirements and can be used as a basis for judging the characteristics of the monitoring signal;

[0023] If the image clarity , it means that the correlation between the IMF component and the monitoring signal does not meet the expected requirements and cannot be used as a basis for judging the characteristics of the monitoring signal.

[0024] Preferably, the LSTM deep learning model includes:

[0025] Build the basic framework of the LSTM deep learning model, which includes the input layer, hidden layer, fully connected layer and output layer, and initialize the parameters to obtain the initialized LSTM deep learning model;

[0026] Set the LSTM deep learning model to use the Adam optimizer, and set the Adam algorithm to adaptively adjust the learning rate based on optimized stochastic gradient descent;

[0027] Obtain gg groups of historical data sets, each of which includes qq feature data sets formed by historical monitoring signals processed by the EEMD model and abnormal data in the feature data sets. The gg groups of historical data sets form a training set;

[0028] The training set is input into the LSTM deep learning model in batches, and the recognition results corresponding to each batch are output; the SoftMarginLoss loss function value between each batch of recognition results and the abnormal data in the feature data set is calculated in turn; the back propagation algorithm is used to update the parameters in each layer of the LSTM deep learning model to obtain a trained convolutional neural network deep model.

[0029] Preferably, the loss function of the LSTM deep learning model is SoftMarginLoss, and the SoftMarginLoss loss function formula is as follows:

[0030] ;

[0031] in, It represents the number of samples: is the actual label, is the decision function of the model, is the regularization parameter, is the L2 norm of the weights.

[0032] Preferably, the GL convolutional neural network model includes:

[0033] Construct the basic framework of the GL convolutional neural network model, set the optimization target of the convolutional neural network to the fault type of abnormal monitoring data, including input layer, convolution layer, pooling layer, activation layer and fully connected layer. The activation layer uses the ReLU activation function and sets the learning rate to adaptive adjustment.

[0034] The GL convolutional neural network model uses a combination of mean square error loss MSELoss and structural similarity loss function L-Loss as a hybrid loss function;

[0035] Obtain mm groups of historical fault anomaly data, each group of historical fault anomaly data includes a historical anomaly monitoring data and a corresponding real fault type, and mm groups of historical fault anomaly data form a fault set;

[0036] That is, the input of the GL convolutional neural network model is the historical abnormal monitoring data, and the output is the predicted fault type;

[0037] The historical abnormal monitoring data in the fault set is imported into the GL convolutional neural network model, and the function value of the mixed loss function is used as the supervision signal to obtain the trained GL convolutional neural network model.

[0038] Preferably, the learning rate of the GL convolutional neural network model is Adopt the improved AdaBelief algorithm for adaptive update and obtain the next iteration number through the improved AdaBelief algorithm Model parameter values for:

[0039] ;

[0040] in, Indicates the current iteration number The first moment estimate of Indicates the current iteration number The second moment estimate of Indicates the current iteration number Model parameter values, represents the learning rate, represents the smoothing term, Indicates adjustment parameters.

[0041] Preferably, the mixed loss function of the GL convolutional neural network model is,

[0042] ;

[0043] in, is the trade-off coefficient, L-Loss is the structural similarity loss function, MSELoss represents the mean square error loss, and the mean square error loss function is as follows:

[0044] ;

[0045] in, is the actual fault type corresponding to the fault anomaly data in the fault set, is the type of fault for which the model generates predictions, is the number of fault anomaly data.

[0046] Preferably, the structural similarity loss function L-Loss is defined as: ;

[0047] in, yes and The covariance of express The variance of express The variance of The characteristics of the real fault type signal in the fault set, represents the features of the model to generate predictions, is the number of true fault type signal features.

[0048] Preferably, the fault types include short circuit, overload, cable aging or insufficient heat dissipation and others;

[0049] The charging pile information includes the location number and equipment line information of the charging pile.

[0050] Compared with the prior art, the present invention has the following beneficial effects:

[0051] 1. In the present invention, an adaptive decomposition method is performed on the signal through the EEMD model, and the features with low correlation in the monitoring signal are eliminated, and the input sequence of the LSTM model is obtained to realize the purification of the monitoring signal, reduce the workload of data processing, reduce the data processing time, and improve the data processing efficiency. Through the setting of the LSTM deep learning model optimizer and the loss function, the model convergence efficiency and quality are accelerated, and the recognition quality and speed of abnormal feature data in a large amount of charging pile signal data are improved, and the automatic diagnosis of charging pile fire fault data is realized, which is convenient for timely warning, ensures the normal operation of the charging pile, and reduces the fire hazard.

[0052] 2. In the present invention, the fault type to which the abnormal monitoring data belongs is obtained through the GL convolutional neural network model, and then the fault type occurring during the operation of the charging pile is obtained, thereby reducing the workload of maintenance personnel for troubleshooting. The extraction unit extracts the charging pile information corresponding to the abnormal monitoring data, calibrates the location of the faulty charging pile and other relevant information, reduces maintenance time, and improves maintenance efficiency. Through automatic identification of fault type and location, intelligent, efficient, and refined management of charging pile fire monitoring is achieved, thereby improving the level and effect of fire supervision management and reducing the probability of fire accidents and the degree of loss. BRIEF DESCRIPTION OF THE DRAWINGS

[0053] Figure 1 The present invention provides a structural schematic diagram of an intelligent fire monitoring system for a charging pile;

[0054] Figure 2 A schematic diagram of the method steps in the present invention;

[0055] Figure 3 It is a flow chart of the data processing module and the abnormality identification module in the present invention. DETAILED DESCRIPTION

[0056] Below, the exemplary embodiments according to the present application will be described in detail with reference to the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present application, rather than all the embodiments of the present application, and it should be understood that the present application is not limited to the exemplary embodiments described here.

[0057] Embodiment 1

[0058] Reference Figure 1 , Figure 2 and Figure 3 Embodiment 1 further illustrates the intelligent fire monitoring system for a charging pile proposed by the present invention.

[0059] A fire in a charging pile or a fire accident may be caused by a variety of factors, including: failure of electrical components inside the charging pile, such as short circuit, overload, cable aging, etc., which may cause arc discharge or overheating of electrical equipment, causing a fire; long-term high-power charging may cause excessive heat to be generated inside the charging pile. If the heat dissipation system is insufficient or there are design defects, the charging pile may overheat, causing a fire; electrical failures usually cause abnormal changes in current, and these changes may be one of the precursors to fire. Current changes include: short circuits can cause a sudden increase in current because the current bypasses the normal path and flows directly through the short circuit point. This abnormal current may cause electrical components to overload, creating the risk of overheating and fire. Overloads can cause current to exceed the rated current of the equipment design, causing electrical equipment to overheat. If the overload persists, it may cause a fire. Cable aging may cause increased resistance, causing an increase in current. Cable aging may also cause open circuits or short circuits, which are all potential fire risks. Failures in electrical equipment may cause arc discharges, generating high temperatures and flames. Arc discharges are usually accompanied by current fluctuations.

[0060] Monitoring and detecting these current, voltage and temperature changes is a key step in preventing fires in advance. The present invention continuously detects the current, voltage, temperature and other parameters of each line and charging pile through the data acquisition module.

[0061] Data acquisition module: including data acquisition unit and marking unit. The data acquisition unit is used to continuously acquire the original signals generated by the charging pile during the cycle time. The original signals include voltage signals, current signals and temperature signals. The marking unit is used to add the corresponding charging pile information to the data acquired by the acquisition unit to obtain the original data of current, voltage and temperature in the charging pile. This can be achieved through sensors, data recording devices or other data acquisition systems to ensure that the collected data contains sufficient time series information for subsequent analysis and model training.

[0062] Storage module: used for data storage. The storage module is used to store the parameters of the charging pile itself, the data collected by the sensor, fault information, alarm information, etc. This system uses Samsung's solid-state hard drive K9F1G for storage. It communicates with STM32 through a data line and is connected to the CPU and 25Q32, which is a serial memory.

[0063] Data processing module: pre-processes the data transmitted by the data acquisition module, and records the pre-processed data as real-time monitoring data. The voltage signal, current signal and temperature signal of the real-time monitoring data are decomposed by the EEMD model to obtain a real-time input sequence, which is recorded as the first data set;

[0064] EEMD model processing (EEMD Preprocessing) is an adaptive decomposition method of signals, which is used to decompose signals into multiple intrinsic mode functions (IMFs). EEMD is applied to each voltage signal, current signal and temperature signal in the real-time monitoring data to obtain its intrinsic mode function IMFs.

[0065] EEMD model processing includes:

[0066] in, Used to replace voltage signals, current signals and temperature signals, that is, voltage signals, current signals and temperature signals can be used To indicate, but Each time can only display either voltage signal, current signal, or temperature signal, not at the same time, monitoring signal Add white noise during decomposition , white noise The introduction of is to make the results of each EMD (Empirical Mode Decomposition) more stable and consistent, thereby improving the reliability of EEMD. In each EMD process, the white noise component is treated as an additional mode function to form a new signal sequence:

[0067] ;

[0068] in, =1,……, , is the average number of times, is the label;

[0069] For each The overall average of the decomposed IMF components is calculated to obtain the result of EEMD decomposition.

[0070] ;

[0071] ;

[0072] In the formula, represents the EEMD decomposition Order IMF component; Indicates After averaging, the EMD decomposition IMF components; represents the residual component of the monitoring signal decomposed by EEMD; Indicates The residual component after EMD decomposition after sub-average

[0073] The monitoring signal can be expressed as: ;

[0074] Calculate the correlation coefficient for each decomposed IMF component: ;

[0075] In the formula, express and The covariance of express The variance of express The variance of

[0076] Set a threshold , if the correlation coefficient , it means that the correlation between the IMF component and the monitoring signal meets the expected requirements and can be used as a basis for judging the characteristics of the monitoring signal. , it means that the correlation between the IMF component and the monitoring signal does not meet the expected requirements, the correlation with the monitoring signal is not high, and it cannot be used as a basis for judging the characteristics of the monitoring signal.

[0077] Construct the input sequence and monitor the signal The intrinsic mode functions IMFs after EEMD processing are combined into an input sequence. Each IMFs can be regarded as a new feature, so the final input sequence will contain multiple features. The input sequence is recorded as the first data set.

[0078] The following is an example of the procedure for decomposing real-time monitoring data using the EEMD model;

[0079] from pyhht.emd import EEMD

[0080] # Create EEMD model

[0081] eemd=EEMD()

[0082] # Input the signal into the model to get IMFs

[0083] IMFs = eemd.eemd(signal)

[0084] # Analyze and reconstruct the signal

[0085] # ...

[0086] # The signal in the example can be replaced by each voltage signal, current signal and temperature signal in the real-time monitoring data)

[0087] import numpy as np

[0088] # Merge IMFs for each signal

[0089] input_sequence=np.vstack((imfs_current,imfs_voltage,imfs_temperature)).T

[0090] Among them, current_signal, voltage_signal and temperature_signal represent voltage signal, current signal and temperature signal respectively, and input_sequence, that is, the first data set, constitutes the input sequence of the LSTM model.

[0091] Anomaly identification module: used to identify abnormal data, including LSTM deep learning model, importing the first data set into the trained LSTM deep learning model, identifying data with abnormal characteristics, and recording them as abnormal monitoring data;

[0092] In the LSTM deep learning model, the long short-term memory network LSTM is used to perform abnormal analysis on the information of the first data set, select the first data set with abnormal data features, form abnormal monitoring data, and perform fault type analysis.

[0093] The LSTM deep learning model includes:

[0094] Build the basic framework of the LSTM deep learning model, which includes the input layer, hidden layer, fully connected layer and output layer, and initialize the parameters to obtain the initialized LSTM deep learning model;

[0095] Get gg groups of historical data sets. Each group of historical data sets includes qq feature data sets formed by historical monitoring signals processed by the EEMD model and abnormal data in the feature data sets. The gg groups of historical data sets form a training set, which is used to train the LSTM deep learning model.

[0096] The input layer is set as the feature data set in the training set. Each hidden layer is composed of multiple cell units. The hidden layer unit of LSTM introduces the concept of cell state and designs three gate structures, namely forget gate, input gate and output gate.

[0097] The forget gate controls the degree to which the information in the previous memory unit is forgotten, and determines how much information in the unit layer at the previous moment is retained in the unit layer at the current moment. Therefore,

[0098] ;

[0099] in, Indicates the weight of the information retained in the previous layer; Indicates the number of activated rain; represents the forget gate weight matrix; Indicates the last moment The output of the hidden layer; Indicates the current time of input; Represents the forget gate bias term.

[0100] The input gate calculates the current input unit layer state and updates it by the output of the unit layer at the previous moment and the input of the unit layer at the current moment:

[0101] ;

[0102] ;

[0103] ;

[0104] Represents the input gate The state of the moment; represents the input gate weight matrix; Represents the unit layer state weight matrix; Represents the input gate bias term: Represents the unit layer state bias term: Indicates the current time The state of the input unit layer; Indicates the current time The updated state of the unit layer; Indicates the last moment The status of the unit layer.

[0105] The output gate calculates the cell state to be output through the state of the current cell layer.

[0106] ;

[0107] ;

[0108] in, Output gate The state of the moment; represents the output gate weight matrix; represents the output gate bias term; Represents the output gate hidden layer output.

[0109] The role of the optimizer is to adjust the model parameters to minimize the loss function. During the training process, by calculating the gradient of the loss function, the optimizer updates the weights and biases of the model based on the information of the gradient so that the loss function is minimized. The loss function measures the difference between the model's predicted output and the actual label. During training, the model learns the mapping from input to output by minimizing the loss function so that it can achieve good performance on a given task.

[0110] The LSTM deep learning model uses the Adam optimizer, and sets the Adam algorithm to adaptively adjust the learning rate based on optimized stochastic gradient descent;

[0111] The loss function is SoftMarginLoss, and the loss function formula of SoftMarginLoss is as follows:

[0112] ;

[0113] in, It represents the number of samples: is the actual label (-1 or 1), is the decision function of the model, is the regularization parameter, L2 norm of the weights.

[0114] The input of the fully connected layer is the output of the last hidden layer of LSTM. Each input node of the fully connected layer is connected to all the output nodes of the previous layer. The output feature representation of the LSTM network layer is integrated and mapped to the sample label space. The initial classification results are normalized, and finally the results identified in the feature data set are output through the output layer.

[0115] The training set is input into the LSTM deep learning model in batches, and the recognition results corresponding to each batch are output; the SoftMarginLoss loss function value between each batch of recognition results and the abnormal data in the feature data set is calculated in turn; the back propagation algorithm is used to propagate the loss error in the reverse direction of the model calculation graph, and the parameters in each layer of the LSTM deep learning model are updated. When the SoftMarginLoss loss function value no longer decreases, the training reaches the expected level and the training ends, and a trained convolutional neural network deep model is obtained.

[0116] The first data set is imported into the trained LSTM deep learning model, and data with abnormal characteristics is identified and recorded as abnormal monitoring data. The abnormal monitoring data is transmitted to the fault analysis module, and the fault analysis module analyzes the cause and fault type of the abnormal monitoring data. Fault types include short circuit, overload, cable aging or insufficient heat dissipation, and others.

[0117] In this embodiment, the signal is adaptively decomposed by the EEMD model to remove features with low correlation in the monitoring signal, obtain the input sequence of the LSTM model, realize the purification of the monitoring signal, reduce the workload of data processing, reduce the data processing time, and improve the data processing efficiency. By setting the LSTM deep learning model optimizer and the loss function, the model convergence efficiency and quality are accelerated, the recognition quality and speed of abnormal feature data in a large amount of charging pile signal data are improved, and the automatic diagnosis of charging pile fire fault data is realized, which facilitates timely warning, ensures the normal operation of the charging pile, and reduces the fire hazard.

[0118] Embodiment 2

[0119] Reference Figure 1 and Figure 2 , Example 2 further illustrates the intelligent fire monitoring system and system of a charging pile proposed by the present invention.

[0120] An intelligent fire monitoring system for a charging pile, a fault analysis module: used for fault type analysis and fault location positioning, including a GL convolutional neural network model and an extraction unit, the abnormal monitoring data is imported into the GL convolutional neural network model, the fault type to which the abnormal monitoring data belongs is obtained, the extraction unit extracts the charging pile information corresponding to the abnormal monitoring data, and calibrates the location where the fault occurs.

[0121] The construction of the GL convolutional neural network model includes:

[0122] Construct the basic framework of the GL convolutional neural network model, set the optimization target of the convolutional neural network to the fault type of abnormal monitoring data, including input layer, convolution layer, pooling layer, activation layer and full connection layer. The activation layer uses the ReLU activation function and sets the learning rate to adaptive adjustment. The learning rate The improved AdaBelief algorithm is used for adaptive updating. The improved AdaBelief algorithm is:

[0123] ;

[0124] ;

[0125] ;

[0126] ;

[0127] ;

[0128] Among them, the adaptive first-order moment estimation attenuation rate is introduced and the second-order moment estimate decay rate , Indicates the current iteration number represents the first-order moment estimate of the decay rate, Indicates the current iteration number Represents the decay rate of the second-order moment estimate, Indicates the current iteration number represents the adaptive first-order moment estimate decay rate, Indicates the current iteration number represents the decay rate of the adaptive second-order moment estimate, Indicates the current iteration number The gradient at a time step, , Respectively represent the current iteration number , the number of previous iterations The first moment estimate of , Respectively represent the current iteration number , the number of previous iterations The second moment estimate of , Respectively represent the current iteration number , the next iteration number Model parameter values, represents the learning rate, Represents the smoothing term and introduces the adjustment parameter The model parameter values ​​are proportionally adjusted to make the learning rate adaptively adjusted for better optimization effect.

[0129] The GL convolutional neural network model uses a combination of mean square error loss MSELoss and structural similarity loss function L-Loss. The hybrid loss function of the GL convolutional neural network model is defined as:

[0130] ;

[0131] in, is the trade-off coefficient, which controls the weight ratio of the two losses in the total loss. The value range is between, when When , the model pays more attention to the structural similarity loss function L-Loss, and emphasizes the feature similarity of the fault type; when When , the model pays more attention to the mean square error loss MSELoss and the content consistency of the fault type.

[0132] The mean square error loss function is as follows:

[0133] ;

[0134] in, is the actual fault type corresponding to the fault anomaly data in the fault set, is the type of fault for which the model generates predictions, is the number of fault anomaly data.

[0135] The structural similarity loss function L-Loss is defined as: ;

[0136] in, yes and The covariance of express The variance of express The variance of The characteristics of the real fault type signal in the fault set, represents the features of the model to generate predictions, is the number of true fault type signal features.

[0137] Obtain mm groups of historical fault anomaly data. Each group of historical fault anomaly data includes a historical anomaly monitoring data and the corresponding real fault type. The mm groups of historical fault anomaly data form a fault set, which is used for training the GL convolutional neural network model.

[0138] The historical abnormal monitoring data in the fault set is imported into the GL convolutional neural network model, that is, the input of the GL convolutional neural network model is the historical abnormal monitoring data, and the output is the predicted fault type; the function value of the mixed loss function is used as the supervision signal, and the parameters in the framework network are updated through back propagation. When the function value of the mixed loss function is minimized and does not change, the trained GL convolutional neural network model is obtained.

[0139] In this embodiment, the fault type to which the abnormal monitoring data belongs is obtained through the GL convolutional neural network model, and then the fault type occurring during the operation of the charging pile is obtained, thereby reducing the workload of maintenance personnel for troubleshooting. The extraction unit extracts the charging pile information corresponding to the abnormal monitoring data, calibrates the location of the faulty charging pile and other relevant information, reduces maintenance time, and improves maintenance efficiency. Through automatic identification of fault types and locations, intelligent, efficient, and refined management of charging pile fire monitoring is achieved, thereby improving the level and effect of fire supervision management and reducing the probability of fire accidents and the degree of loss.

[0140] Embodiment 3

[0141] Reference Figure 1 and Figure 2 , Example 3 further illustrates the intelligent fire monitoring system and system of a charging pile proposed by the present invention.

[0142] An intelligent fire monitoring system for a charging pile, the operation of the intelligent fire monitoring system for the charging pile comprises the following steps:

[0143] Step S1: The data acquisition unit continuously acquires the original signal generated by the charging pile within the cycle time, and transmits it to the data processing module after being marked by the marking unit;

[0144] Step S2: the data processing module pre-processes the original signal transmitted by the data acquisition module to obtain real-time monitoring data, and obtains the first data set after EEMD model decomposition and judgment unit analysis;

[0145] Step S3: importing the first data set into the trained LSTM deep learning model to obtain abnormal monitoring data;

[0146] Step S4: Import the abnormal monitoring data into the GL convolutional neural network model to obtain the fault type to which the abnormal monitoring data belongs.

[0147] Step S5: The abnormal monitoring data, the fault type to which it belongs, and the location where the fault occurs are alarmed through the display module.

[0148] The intelligent fire monitoring system of the charging pile includes:

[0149] Data acquisition module: including a data acquisition unit and a marking unit. The data acquisition unit is used to continuously acquire the monitoring signals generated by the charging pile within the cycle time. The monitoring signals include voltage signals, current signals and temperature signals. The marking unit is used to add corresponding charging pile information to the data acquired by the acquisition unit.

[0150] Storage module: used for data storage;

[0151] Data processing module: pre-processes the data transmitted by the data acquisition module, and records the pre-processed data as real-time monitoring data. The voltage signal, current signal and temperature signal of the real-time monitoring data are decomposed by the EEMD model to obtain a real-time input sequence, which is recorded as the first data set;

[0152] Anomaly identification module: used to identify abnormal data, including LSTM deep learning model, importing the first data set into the trained LSTM deep learning model, identifying data with abnormal characteristics, and recording them as abnormal monitoring data;

[0153] Fault analysis module: used for fault type analysis and fault location location, including GL convolutional neural network model and extraction unit. The abnormal monitoring data is imported into the GL convolutional neural network model to obtain the fault type to which the abnormal monitoring data belongs. The extraction unit calibrates the location of the fault through the charging pile information corresponding to the abnormal monitoring data.

[0154] Display module: includes a human-computer interaction interface for viewing the operating status of the charging pile.

[0155] Display module: includes a human-computer interaction interface, which is used to view the operating status of the charging pile. In the display module, managers need to view the operating status of the charging pile through the human-computer interaction interface (display or computer management system). The human-computer interaction interface includes the working parameters of each monitored charging pile, the location and type of fault, and specific events.

[0156] The charging piles in the layout of this system also include running lights and buzzers. When the charging piles are working normally, a green light source is used for display. When the charging piles are working abnormally, a red light source is used for alarm and a buzzer is used for sound warning.

[0157] Through real-time monitoring, intelligent early warning, remote control and other functions, it provides strong technical support for the fire safety of charging piles. When an abnormal situation is detected, the system will immediately issue a warning signal to attract the attention of relevant personnel, so that measures can be taken in advance to prevent fire accidents. Real-time monitoring and early warning based on Internet of Things technology are more efficient and accurate, and can effectively reduce the probability of fire accidents.

[0158] The above is only a preferred embodiment of the present invention, and the protection scope of the present invention is not limited to the above embodiments. All technical solutions under the concept of the present invention belong to the protection scope of the present invention. It should be pointed out that for ordinary technical users in this technical field, some improvements and modifications without departing from the principle of the present invention should also be regarded as the protection scope of the present invention.

Claims

1. An intelligent fire monitoring system for a charging pile, characterized in that: The intelligent fire monitoring system of the charging pile includes: Data acquisition module: including a data acquisition unit and a marking unit. The data acquisition unit is used to continuously acquire the original signal generated by the charging pile within the cycle time. The monitoring signal includes a voltage signal, a current signal and a temperature signal. The marking unit is used to add the corresponding charging pile information to the data acquired by the acquisition unit. Storage module: used for data storage; Data processing module: pre-processes the data transmitted by the data acquisition module, and records the pre-processed data as real-time monitoring data. The voltage signal, current signal and temperature signal of the real-time monitoring data are decomposed by the EEMD model to obtain a real-time input sequence, which is recorded as the first data set; The EEMD model processing includes: Set the monitoring signal of real-time monitoring data to ,in, Used to replace voltage signals, current signals and temperature signals, and monitor signals Add white noise during decomposition , forming a new signal sequence: ;in, =1,……, , is the average number of times, is the label; For each, perform EEMD decomposition to obtain: ; Calculate the correlation coefficient for each decomposed IMF component: ; In the formula, represents the EEMD decomposition Order IMF component; represents the residual component of the monitoring signal decomposed by EEMD; Indicates The residual component after EMD decomposition after averaging is express and The covariance of express The variance of express The variance of Anomaly identification module: used to identify abnormal data, including LSTM deep learning model, importing the first data set into the trained LSTM deep learning model, identifying data with abnormal characteristics, and recording them as abnormal monitoring data; Fault analysis module: used for fault type analysis and fault location location, including GL convolutional neural network model and extraction unit. The abnormal monitoring data is imported into the GL convolutional neural network model to obtain the fault type to which the abnormal monitoring data belongs. The extraction unit calibrates the location of the fault through the charging pile information corresponding to the abnormal monitoring data. Display module: includes a human-computer interaction interface for viewing the operating status of the charging pile.

2. The intelligent fire monitoring system for a charging pile according to claim 1, characterized in that: The EEMD model also includes: Set a threshold ; If the correlation coefficient , it means that the correlation between the IMF component and the monitoring signal meets the expected requirements and can be used as a basis for judging the characteristics of the monitoring signal; If the image clarity , it means that the correlation between the IMF component and the monitoring signal does not meet the expected requirements and cannot be used as a basis for judging the characteristics of the monitoring signal.

3. The intelligent fire monitoring system for a charging pile according to claim 2 is characterized in that: The LSTM deep learning model includes: Build the basic framework of the LSTM deep learning model, which includes the input layer, hidden layer, fully connected layer and output layer, and initialize the parameters to obtain the initialized LSTM deep learning model; Set the LSTM deep learning model to use the Adam optimizer, and set the Adam algorithm to adaptively adjust the learning rate based on optimized stochastic gradient descent; Obtain gg groups of historical data sets, each of which includes qq feature data sets formed by historical monitoring signals processed by the EEMD model and abnormal data in the feature data sets. The gg groups of historical data sets form a training set; The training set is input into the LSTM deep learning model in batches, and the recognition results corresponding to each batch are output; the SoftMarginLoss loss function value between each batch of recognition results and the abnormal data in the feature data set is calculated in turn; the back propagation algorithm is used to update the parameters in each layer of the LSTM deep learning model to obtain a trained convolutional neural network deep model.

4. The intelligent fire monitoring system for a charging pile according to claim 3 is characterized in that: The loss function of the LSTM deep learning model uses SoftMarginLoss, and the formula of SoftMarginLoss loss function is as follows: ; in, It represents the number of samples: is the actual label, is the decision function of the model, is the regularization parameter, is the L2 norm of the weights.

5. The intelligent fire monitoring system for a charging pile according to claim 4, characterized in that: The GL convolutional neural network model includes: Construct the basic framework of the GL convolutional neural network model, set the optimization target of the convolutional neural network to the fault type of abnormal monitoring data, including input layer, convolution layer, pooling layer, activation layer and fully connected layer. The activation layer uses the ReLU activation function and sets the learning rate to adaptive adjustment. The GL convolutional neural network model uses a combination of mean square error loss MSELoss and structural similarity loss function L-Loss as a hybrid loss function; Obtain mm groups of historical fault anomaly data, each group of historical fault anomaly data includes a historical anomaly monitoring data and a corresponding real fault type, and mm groups of historical fault anomaly data form a fault set; That is, the input of the GL convolutional neural network model is the historical abnormal monitoring data, and the output is the predicted fault type; The historical abnormal monitoring data in the fault set is imported into the GL convolutional neural network model, and the function value of the mixed loss function is used as the supervision signal to obtain the trained GL convolutional neural network model.

6. The intelligent fire monitoring system for a charging pile according to claim 5, characterized in that: The learning rate of the GL convolutional neural network model Adopt the improved AdaBelief algorithm for adaptive update and obtain the next iteration number through the improved AdaBelief algorithm Model parameter values for: ; in, Indicates the current iteration number The first moment estimate of Indicates the current iteration number The second moment estimate of Indicates the current iteration number Model parameter values, represents the learning rate, represents the smoothing term, Indicates adjustment parameters.

7. The intelligent fire monitoring system for a charging pile according to claim 6, characterized in that: The mixed loss function of the GL convolutional neural network model is, ; in, is the trade-off coefficient, L-Loss is the structural similarity loss function, MSELoss represents the mean square error loss, and the mean square error loss function is as follows: ; in, is the actual fault type corresponding to the fault anomaly data in the fault set, is the type of failure that the model generates predictions for, is the number of fault anomaly data.

8. The intelligent fire monitoring system for a charging pile according to claim 7, characterized in that: The structural similarity loss function L-Loss is defined as: ; in, yes and The covariance of express The variance of express The variance of The characteristics of the real fault type signal in the fault set, represents the features of the model to generate predictions, is the number of true fault type signal features.

9. The intelligent fire monitoring system for a charging pile according to claim 8, characterized in that: The fault types include short circuit, overload, cable aging or insufficient heat dissipation and others; The charging pile information includes the location number and equipment line information of the charging pile.

Images