CN117942517A - Intelligent fire control monitored control system of filling electric pile - Google Patents
Intelligent fire control monitored control system of filling electric pile Download PDFInfo
- Publication number
- CN117942517A CN117942517A CN202410144973.6A CN202410144973A CN117942517A CN 117942517 A CN117942517 A CN 117942517A CN 202410144973 A CN202410144973 A CN 202410144973A CN 117942517 A CN117942517 A CN 117942517A
- Authority
- CN
- China
- Prior art keywords
- data
- monitoring
- fault
- charging pile
- model
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Pending
Links
- 238000011049 filling Methods 0.000 title description 3
- 238000012544 monitoring process Methods 0.000 claims abstract description 125
- 230000002159 abnormal effect Effects 0.000 claims abstract description 71
- 238000013527 convolutional neural network Methods 0.000 claims abstract description 43
- 238000013136 deep learning model Methods 0.000 claims abstract description 35
- 238000012545 processing Methods 0.000 claims abstract description 23
- 238000000605 extraction Methods 0.000 claims abstract description 11
- 238000007781 pre-processing Methods 0.000 claims abstract description 7
- 230000006870 function Effects 0.000 claims description 51
- 238000000034 method Methods 0.000 claims description 21
- 238000000354 decomposition reaction Methods 0.000 claims description 19
- 238000012549 training Methods 0.000 claims description 15
- 238000004422 calculation algorithm Methods 0.000 claims description 13
- 230000004913 activation Effects 0.000 claims description 9
- 238000004458 analytical method Methods 0.000 claims description 9
- 230000003993 interaction Effects 0.000 claims description 7
- 230000008569 process Effects 0.000 claims description 7
- ORILYTVJVMAKLC-UHFFFAOYSA-N Adamantane Natural products C1C(C2)CC3CC1CC2C3 ORILYTVJVMAKLC-UHFFFAOYSA-N 0.000 claims description 6
- 230000032683 aging Effects 0.000 claims description 6
- 238000005457 optimization Methods 0.000 claims description 5
- 230000017525 heat dissipation Effects 0.000 claims description 4
- 108010076504 Protein Sorting Signals Proteins 0.000 claims description 3
- 230000005856 abnormality Effects 0.000 claims description 3
- 238000009499 grossing Methods 0.000 claims description 3
- 238000011176 pooling Methods 0.000 claims description 3
- 238000005303 weighing Methods 0.000 claims description 3
- 238000002156 mixing Methods 0.000 claims description 2
- 238000012935 Averaging Methods 0.000 claims 1
- 206010000369 Accident Diseases 0.000 abstract description 6
- 238000012423 maintenance Methods 0.000 description 6
- 230000000694 effects Effects 0.000 description 4
- 239000011159 matrix material Substances 0.000 description 4
- 238000013461 design Methods 0.000 description 3
- 238000010891 electric arc Methods 0.000 description 3
- 239000000284 extract Substances 0.000 description 3
- 238000011161 development Methods 0.000 description 2
- 230000018109 developmental process Effects 0.000 description 2
- 238000003745 diagnosis Methods 0.000 description 2
- 238000010586 diagram Methods 0.000 description 2
- 210000000056 organ Anatomy 0.000 description 2
- 238000000746 purification Methods 0.000 description 2
- 230000001105 regulatory effect Effects 0.000 description 2
- 230000006978 adaptation Effects 0.000 description 1
- 230000003044 adaptive effect Effects 0.000 description 1
- 230000009286 beneficial effect Effects 0.000 description 1
- 230000033228 biological regulation Effects 0.000 description 1
- 230000005540 biological transmission Effects 0.000 description 1
- 230000008859 change Effects 0.000 description 1
- 238000010276 construction Methods 0.000 description 1
- 230000007547 defect Effects 0.000 description 1
- 238000003912 environmental pollution Methods 0.000 description 1
- 238000004880 explosion Methods 0.000 description 1
- 238000010304 firing Methods 0.000 description 1
- 230000020169 heat generation Effects 0.000 description 1
- 230000006872 improvement Effects 0.000 description 1
- 238000013507 mapping Methods 0.000 description 1
- 238000012821 model calculation Methods 0.000 description 1
- 238000012986 modification Methods 0.000 description 1
- 230000004048 modification Effects 0.000 description 1
- 238000010606 normalization Methods 0.000 description 1
- 239000002243 precursor Substances 0.000 description 1
- 238000011160 research Methods 0.000 description 1
- 239000007787 solid Substances 0.000 description 1
Abstract
The invention discloses an intelligent fire-fighting monitoring system of a charging pile, which relates to the fire-fighting field of the charging pile, and the technical scheme is characterized in that the intelligent fire-fighting monitoring system comprises: the data acquisition unit is used for continuously acquiring monitoring signals generated by the charging pile in the period time, the data processing module is used for preprocessing the data transmitted by the data acquisition module to acquire real-time monitoring data, and the EEMD model is used for decomposing the real-time monitoring data to acquire a first data set; importing the first data set into a trained LSTM deep learning model, identifying data with abnormal characteristics, and recording the data as abnormal monitoring data; the abnormal monitoring data is imported into the G-L convolutional neural network model, the fault type of the abnormal monitoring data is obtained, the extraction unit marks the position 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 occurrence probability of fire accidents can be effectively reduced.
Description
Technical Field
The invention relates to the field of fire protection of charging piles, in particular to an intelligent fire protection monitoring system of a charging pile.
Background
With the increasing global energy demand and the aggravation of environmental pollution problems, various industries implement sustainable development environment protection concepts through innovative research, electric vehicles are widely focused and popularized as vehicles using clean energy in order to reduce negative effects on the environment, meanwhile, the rapid development of the electric vehicles also challenges the improvement of charging infrastructure, and a charging pile station plays an important role in popularization of new energy vehicles.
But in the safety supervision process of the charging pile station, fire safety is an important hidden danger, especially for a large-scale charging pile station, the number of charging piles is extremely large and the charging tasks are numerous, so that the charging piles are in a charging working state for a long time, fire safety problems such as fire and explosion are extremely easy to occur, therefore, early warning and monitoring are extremely important before the occurrence of the safety accidents, the fire of the charging piles or the occurrence of the fire-fighting accidents can be caused by various factors, for example, the electric elements in the charging piles are failed, the long-time high-power charging can cause excessive heat in the charging piles, and monitoring and detecting the current changes are key steps for preventing fires in advance.
Disclosure of Invention
Aiming at the problems existing in the prior art, the invention aims to provide an intelligent fire-fighting monitoring system for a charging pile, which realizes high-efficiency and accurate monitoring and early warning of the fire fighting of the charging pile.
In order to achieve the above purpose, the present invention provides the following technical solutions: an intelligent fire monitoring system of charging pile, the intelligent fire monitoring system of charging pile includes:
and a data acquisition module: the device comprises a data acquisition unit and a marking unit, wherein the data acquisition unit is used for continuously acquiring monitoring signals generated by a charging pile in a period time, and the monitoring signals comprise voltage signals, current signals and temperature signals; the marking unit is used for adding corresponding charging pile information into the data acquired by the acquisition unit;
and a storage module: for storage of data;
and a data processing module: preprocessing data transmitted by a data acquisition module, recording the preprocessed data as real-time monitoring data, decomposing voltage signals, current signals and temperature signals of the real-time monitoring data through an EEMD model, acquiring a real-time input sequence, and recording the real-time input sequence as a first data set;
An anomaly identification module: the method comprises the steps of identifying abnormal data, including an LSTM deep learning model, importing a first data set into the trained LSTM deep learning model, identifying data with abnormal characteristics, and recording the data as abnormal monitoring data;
And a fault analysis module: the system is used for analyzing fault types and positioning fault positions and comprises a G-L convolutional neural network model and an extraction unit, wherein abnormal monitoring data are imported into the G-L convolutional neural network model, fault types of the abnormal monitoring data are obtained, and the extraction unit calibrates the positions of faults through charging pile information corresponding to the abnormal monitoring data;
and a display module: the method comprises a man-machine interaction interface for checking the running state of the charging pile.
Preferably, the EEMD model processing includes:
Setting a monitoring signal of real-time monitoring data as x (t), wherein x (t) is used for replacing a voltage signal, a current signal and a temperature signal, white noise n i (t) is added in the decomposition process of the monitoring signal x (t), and a new signal sequence is formed:
Xi(t)＝x(t)+ni(t)；
wherein i=1, … …, N is the average number of times, i is the index;
EEMD decomposition is performed for each X i (t) to obtain:
Calculating a correlation coefficient for each IMF component of the decomposition:
where c j (t) represents the jth IMF component of EEMD decomposition; res (t) represents EEMD decomposition monitoring The variance of c j (t) is shown.
Preferably, the EEMD model further comprises:
setting a threshold value a;
if the correlation coefficient r j is more than or equal to a, the correlation between the IMF component and the monitoring signal reaches the expected requirement, and the correlation coefficient can be used as a basis for judging the characteristics of the monitoring signal;
If the image definition r j is smaller than a, the correlation between the IMF component and the monitoring signal does not meet the expected requirement, and the correlation cannot be used as a basis for judging the characteristics of the monitoring signal.
Preferably, the LSTM deep learning model includes:
Constructing a basic framework of the LSTM deep learning model, wherein the basic framework comprises an input layer, a hidden layer, a full-connection layer and an output layer, and initializing parameters to obtain an initialized LSTM deep learning model;
Setting an LSTM deep learning model, selecting an Adam optimizer, setting an Adam algorithm to adaptively adjust the learning rate based on optimizing random gradient descent;
Acquiring gg historical data sets, wherein each set of historical data sets comprises qq characteristic data sets formed by processing historical monitoring signals through an EEMD model and abnormal data in the characteristic data sets, and the gg historical data sets form a training set;
Inputting the training set into the LSTM deep learning model in batches, and outputting a corresponding identification result of each batch; by sequentially calculating SoftMarginLoss loss function values between each batch of identification results and the abnormal data in the characteristic dataset; and updating parameters in each layer of LSTM deep learning model by using a back propagation algorithm to obtain a trained convolutional neural network deep model.
Preferably, the loss function of the LSTM deep learning model is selected from SoftMarginLoss, softMarginLoss loss function formulas as follows:
Where V denotes the number of samples, y v is the actual label (-1 or 1), ε is the decision function of the model, λ is the regularization parameter, ω 2 is the L2 norm of the weight.
Preferably, the G-L convolutional neural network model includes:
constructing a basic framework of a G-L convolutional neural network model, setting an optimization target of the convolutional neural network as a fault type of abnormal monitoring data, wherein the fault type comprises an input layer, a convolutional layer, a pooling layer, an activation layer and a full-connection layer, the activation layer uses a ReLU activation function, and setting a learning rate as self-adaptive adjustment;
The G-L convolutional neural network model adopts the combination of a mean square error Loss MSELoss and a structural similarity Loss function L-Loss as a mixed Loss function;
Acquiring mm groups of historical fault abnormal data, wherein each group of historical fault abnormal data comprises one historical abnormality monitoring data and a corresponding real fault type, and the mm groups of historical fault abnormal data form a fault set;
the input end of the G-L convolutional neural network model is historical abnormal monitoring data, and the output end is a predicted fault type;
And importing the historical anomaly monitoring data in the fault set into a G-L convolutional neural network model, and obtaining a trained G-L convolutional neural network model by using the function value of the mixed loss function as a supervisory signal.
Preferably, the learning rate α of the G-L convolutional neural network model is adaptively updated by using an improved AdaBelief algorithm, and the next iteration number q+1 model parameter value θ q obtained by using the improved AdaBelief algorithm is:
Where m q represents a first moment estimate of the current iteration number q, k q represents a second moment estimate of the current iteration number q, θ q represents a model parameter value of the current iteration number q, α represents a learning rate, ε represents a smoothing term, and η represents an adjustment parameter.
Preferably, the mixing loss function of the G-L convolutional neural network model is that,
TotalLoss＝φMSELoss+(1-φ)L-Loss；
Wherein, φ is a weighing coefficient, L-Loss is a structural similarity Loss function, MSELoss represents a mean square error Loss, and the mean square error Loss function is as follows:
Wherein Yb is the real fault type corresponding to the fault abnormal data in the fault set, Is the model generation predicted fault type, and L is the number of fault anomaly data.
Preferably, the structural similarity Loss function L-Loss is defined as:
B is the number of true fault type signal features.
Preferably, the fault categories include short circuit, overload, cable aging or insufficient heat dissipation, among others;
the charging pile information includes a position number of the charging pile and equipment line information.
Compared with the prior art, the invention has the following beneficial effects:
1. According to the invention, the EEMD model is used for carrying out a self-adaptive decomposition method on the signals, the features with lower relevance in the monitoring signals are removed, the input sequence of the LSTM model is obtained, the purification of the monitoring signals is realized, the workload of data processing is reduced, the data processing time is shortened, the data processing efficiency is improved, the convergence efficiency and quality of the model are accelerated through the setting of the LSTM deep learning model optimizer and the loss function, the recognition quality and speed of abnormal feature data in a large number of charging pile signal data are improved, the automatic diagnosis of the charging pile fire-fighting fault data is realized, the timely early warning is convenient, the normal operation of the charging pile is ensured, and the fire hazard is reduced.
2. According to the invention, the fault type of the abnormal monitoring data is obtained through the G-L convolutional neural network model, so that the fault type of the charging pile running is obtained, the workload of maintenance personnel for checking is reduced, the extraction unit extracts the charging pile information corresponding to the abnormal monitoring data, the position and other related information of the charging pile with the fault are calibrated, the maintenance time is reduced, the maintenance efficiency is improved, the intelligent, efficient and refined management of the fire control monitoring of the charging pile is realized through the automatic identification of the fault type and the position, the level and effect of the fire control supervision management are improved, and the occurrence probability and the loss degree of fire accidents are reduced.
Drawings
Fig. 1 is a schematic structural diagram of a vending machine management system according to the present invention;
FIG. 2 is a schematic diagram of the method steps of the present invention;
FIG. 3 is a schematic flow chart of a data processing module and an anomaly identification module in the present invention.
Detailed Description
Hereinafter, exemplary embodiments according to the present application will be described in detail with reference to the accompanying drawings. It should be apparent that the described embodiments are only some embodiments of the present application and not all embodiments of the present application, and it should be understood that the present application is not limited by the example embodiments described herein.
Example 1
Referring to fig. 1, fig. 2 and fig. 3, an embodiment of an intelligent fire-fighting monitoring system for a charging pile according to the present invention is further described.
The firing of the charging pile or the occurrence of a fire accident may be caused by various factors including: faults of electrical elements in the charging pile, such as short circuit, overload, cable aging and the like, can cause arc discharge or overheat of electrical equipment, thereby causing fire disaster; the long-time high-power charging may cause excessive heat generation inside the charging pile, and if a heat dissipation system is insufficient or there is a design defect, the charging pile may be overheated, thereby causing a fire disaster; electrical faults typically cause abnormal changes in current which may be one of the precursors to a fire, including: short circuit can lead to sudden rise of current, because current bypasses the normal path and directly flows through the short circuit point, such abnormal current can cause overload of electric elements, risk of overheat and fire, overload can cause current to exceed rated current of equipment design, cause electric equipment to overheat, if overload condition exists continuously, fire can be caused, cable aging can cause resistance to increase, thus current rise is caused, cable aging can also cause disconnection or short circuit, all of which are potential fire risks, faults of electric equipment can cause arc discharge, high temperature and flame are generated, and arc discharge is usually accompanied with fluctuation of current.
The monitoring and detecting of these current, voltage and temperature changes are key steps for preventing fire in advance, the invention continuously detects parameters of current, voltage and temperature of each line and charging pile through the data acquisition module,
And a data acquisition module: the device comprises a data acquisition unit and a marking unit, wherein the data acquisition unit is used for continuously acquiring an original signal generated by a charging pile in a period time, and the original signal comprises a voltage signal, a current signal and a temperature signal; the marking unit is used for adding corresponding charging pile information into the data acquired by the acquisition unit to acquire the original data of current, voltage and temperature in the charging pile. This may be accomplished by sensors, data recording devices, or other data acquisition systems, ensuring that the acquired data contains sufficient time series information for subsequent analysis and model training.
And a storage module: the system is used for storing data, and the storage module is used for storing parameters of the charging pile, data acquired by the sensor, fault information, alarm information and the like, and the system adopts the three-star solid state disk K9F1G for storage. Communicates with STM32 via data lines and is connected to CPU and 25Q32, 25Q32 being a serial memory.
And a data processing module: preprocessing data transmitted by a data acquisition module, recording the preprocessed data as real-time monitoring data, decomposing voltage signals, current signals and temperature signals of the real-time monitoring data through an EEMD model, acquiring a real-time input sequence, and recording the real-time input sequence as a first data set;
EEMD model processing (EEMD Preprocessing) is an adaptive decomposition method of signals to decompose the signals into a plurality of eigenmode functions (IMFs) by applying EEMD to each of the voltage signal, current signal, and temperature signal in real-time monitoring data and obtaining its eigenmode functions IMFs.
EEMD model processing includes:
Setting a monitoring signal of real-time monitoring data as x (t), wherein x (t) is used for replacing a voltage signal, a current signal and a temperature signal, namely, the voltage signal, the current signal and the temperature signal can be expressed by x (t), but x (t) can only express the voltage signal, the current signal or the temperature signal at each time and can not express the voltage signal, or the current signal or the temperature signal at the same time, white noise n i (t) is added in the decomposition process of the monitoring signal x (t), the white noise n i (t) is introduced to enable the result of each EMD (Empirical Mode Decomposition) to be more stable and consistent, thereby improving the reliability of EEMD, and in each EMD process, the white noise component is treated as an additional modal function to form a new signal sequence:
Xi(t)＝x(t)+ni(t)；
wherein i=1, … …, N is the average number of times, i is the index;
the IMF components of each X i (t) decomposition are overall averaged to obtain the result of the EEMD decomposition.
Where c j (t) represents the jth IMF component of EEMD decomposition; c i,j (t) represents the jth IMF component after the ith average, which is decomposed by EMD; res (t) represents the monitor signal residual component decomposed by EEMD; res i (t) represents the residual component after the ith average that was decomposed by EMD
The monitoring signal may be expressed as:
Calculating a correlation coefficient for each IMF component of the decomposition:
Setting a threshold value a, if the correlation coefficient r j is larger than or equal to a, indicating that the correlation between the IMF component and the monitoring signal meets the expected requirement, and taking the IMF component and the monitoring signal as a basis for judging the characteristics of the monitoring signal, if the image definition r j is smaller than a, indicating that the correlation between the IMF component and the monitoring signal does not meet the expected requirement, and taking the correlation between the IMF component and the monitoring signal as a basis for judging the characteristics of the monitoring signal.
The input sequence is constructed and the eigenmode functions IMFs of the monitor signal x (t) after EEMD processing are combined into the input sequence, each IMFs can be regarded as a new feature, so that the final input sequence will contain a plurality of features and the input sequence will be recorded as a first dataset.
The following is an example of a procedure for decomposing real-time monitoring data by an EEMD model;
The current_ signal, voltage _signal and the temperature_signal represent the voltage signal, the current signal and the temperature signal, respectively, and the input_sequence, i.e. the first data, is integrated into the input sequence of the LSTM model.
An anomaly identification module: the method comprises the steps of identifying abnormal data, including an LSTM deep learning model, importing a first data set into the trained LSTM deep learning model, identifying data with abnormal characteristics, and recording the data as abnormal monitoring data;
In the LSTM deep learning model, the information of the first data set is subjected to anomaly analysis by using a long-short-time memory network LSTM, the first data set with abnormal data characteristics is selected to form anomaly monitoring data, and fault category analysis is performed.
The LSTM deep learning model comprises:
Constructing a basic framework of the LSTM deep learning model, wherein the basic framework comprises an input layer, a hidden layer, a full-connection layer and an output layer, and initializing parameters to obtain an initialized LSTM deep learning model;
Acquiring gg historical data sets, wherein each set of historical data set comprises qq characteristic data sets formed by processing historical monitoring signals through an EEMD model and abnormal data in the characteristic data sets, the gg historical data sets form a training set, and the training set is used for training an LSTM deep learning model;
The input layer is set as a characteristic data set in the training set, each hidden layer is composed of a plurality of cell units, and the hidden layer units of the LSTM design 3 gate structures, namely a forgetting gate, an input gate and an output gate through the concept of leading cell states.
The forgetting gate controls the forgetting degree of the information in the previous memory unit and determines the information of the unit layer at the last moment to be remained in the unit layer at the current moment, so,
fe＝σ(Wf·[he-1,se]+bf)；
Wherein f e E [0,1] represents the weight of the information reservation of the previous layer; sigma represents the number of active rain; w f represents a forgetting gate weight matrix; h e-1 represents the output of the hidden layer at the previous time e-1; s e represents the person transmission at the current time e; b f denotes a forget gate bias term.
The input gate calculates and updates the cell layer state of the current person to be input through the output of the cell layer at the previous moment and the input of the cell layer at the current moment:
ze＝σ(Wz·[he-1,se]+bz)；
z e represents the state at the time of input gate e; w z represents an input gate weight matrix; w c represents a unit layer state weight matrix; b z represents a door bias term, b c represents a unit layer state bias term: Representing the state of the input unit layer at the current time e; c e represents the updated state of the unit layer at the current time e; c e-1 represents the state of the cell layer at the previous time e-1.
The output gate calculates the state of the cell to be output from the state of the current cell layer,
oe＝σ(Wo·[he-1,se]+bo)；
he＝oe·tanh(Ce)；
Wherein o e represents the state of the output gate at the time e; w o represents an output gate weight matrix; b o denotes an output gate bias term; h e denotes the output gate hidden layer output.
The role of the optimizer is an algorithm for adjusting the model parameters to minimize the loss function. In the training process, the optimizer updates the weight and bias of the model according to the information of the gradient by calculating the gradient of the loss function, so that the loss function reaches the minimum value. The loss function measures the difference between the model predicted output and the actual label. In training, the model learns the mapping from input to output by minimizing the loss function so that it achieves good performance on a given task.
The LSTM deep learning model selects an Adam optimizer, and sets the Adam algorithm to adaptively adjust the learning rate based on the random gradient descent optimization;
The loss function is chosen by SoftMarginLoss, softMarginLoss as follows:
Where V denotes the number of samples, y v is the actual label (-1 or 1), ε is the decision function of the model, λ is the regularization parameter, ω 2 is the L2 norm of the weight.
The input of the full-connection layer is the output of the last hidden layer of the LSTM, each input node of the full-connection layer is connected with all output nodes of the upper layer, the output characteristic representation of the LSTM network layer is mapped to the sample mark space in a synthesized mode, the initial result of the normalization processing classification is finally output through the output layer, and the result of the characteristic data set identification is finally output.
Inputting the training set into the LSTM deep learning model in batches, and outputting a corresponding identification result of each batch; by sequentially calculating SoftMarginLoss loss function values between each batch of identification results and the abnormal data in the characteristic dataset; and returning the loss error along the reverse direction of the model calculation graph by using a reverse propagation algorithm, updating parameters in each layer of LSTM deep learning model, and when SoftMarginLoss loss function values are not reduced any more, training is expected and the training is finished, so that a trained convolutional neural network deep model is obtained.
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 are transmitted to a fault analysis module, and the fault analysis module analyzes the reasons and fault types generated by the abnormal monitoring data. Fault categories include short circuits, overload, cable aging or insufficient heat dissipation, among others.
In the embodiment, the EEMD model is used for carrying out a self-adaptive decomposition method on the signals, eliminating the characteristics with lower relevance in the monitoring signals, acquiring the input sequence of the LSTM model, realizing the purification of the monitoring signals, reducing the workload of data processing, reducing the data processing time, improving the data processing efficiency, accelerating the model convergence efficiency and quality through the setting of an LSTM deep learning model optimizer and a loss function, improving the recognition quality and speed of abnormal characteristic data in a large number of charging pile signal data, realizing the automatic diagnosis of the charging pile fire-fighting fault data, facilitating the timely early warning, ensuring the normal operation of the charging piles and reducing the fire hazard.
Example two
Referring to fig. 1 and fig. 2, a second embodiment of the present invention provides an intelligent fire-fighting monitoring system and system for a charging pile.
An intelligent fire control monitoring system of charging pile, a fault analysis module: the method is used for analyzing the fault type and positioning the fault position and comprises a G-L convolutional neural network model and an extraction unit, wherein abnormal monitoring data are imported into the G-L convolutional neural network model, the fault type of the abnormal monitoring data is obtained, the extraction unit extracts charging pile information corresponding to the abnormal monitoring data, and the position where the fault occurs is calibrated.
The construction of the G-L convolutional neural network model comprises the following steps:
Constructing a basic framework of a G-L convolutional neural network model, setting an optimization target of the convolutional neural network as a fault type of abnormal monitoring data, wherein the basic framework comprises an input layer, a convolutional layer, a pooling layer, an activation layer and a full-connection layer, the activation layer uses a ReLU activation function, and sets a learning rate as self-adaptive adjustment, and then the learning rate alpha is adaptively updated by adopting an improved AdaBelief algorithm, wherein the improved AdaBelief algorithm is as follows:
Wherein, the self-adaptive first moment estimation attenuation rate beta 1 and the self-adaptive second moment estimation attenuation rate beta 2,β1,q represent the first moment estimation attenuation rate represented by the current iteration number q, the second moment estimation attenuation rate represented by the current iteration number q is represented by beta 2,q, Represents the self-adaptive first moment estimation attenuation rate represented by the current iteration number q,/>The self-adaptive second moment estimation attenuation rate represented by the current iteration number q is represented, u q represents the gradient of the time step of the current iteration number q, m q、mq-1 represents the first moment estimation of the current iteration number q and the last iteration number q-1 respectively, k q、kq-1 represents the second moment estimation of the current iteration number q and the last iteration number q-1 respectively, θ q、θq+1 represents the model parameter value of the current iteration number q and the next iteration number q+1 respectively, alpha represents the learning rate, epsilon represents the smoothing term, and the model parameter value is proportionally regulated by introducing the regulating parameter eta, so that the self-adaptive regulation optimization effect of the learning rate is better.
The G-L convolutional neural network model adopts the combination of the mean square error Loss MSELoss and the structural similarity Loss function L-los to define the mixed Loss function of the G-L convolutional neural network model as,
TotalLoss＝φMSELoss+(1-φ)L-Loss；
Wherein phi is a weighing coefficient, the weight proportion of the two losses in the total Loss is controlled, the value range of phi is between [0,1], and when lambda=0, the model is more focused on the structural similarity Loss function L-Loss and more focused on the characteristic similarity of the fault type; when λ=1, the model is more focused on the mean square error loss MSELoss, more content consistency of the failure type.
The mean square error loss function is as follows:
Wherein Yb is the real fault type corresponding to the fault abnormal data in the fault set, Is the model generation predicted fault type, and L is the number of fault anomaly data.
The structural similarity Loss function L-Loss is defined as:
B is the number of true fault type signal features.
And acquiring mm groups of historical fault abnormal data, wherein each group of historical fault abnormal data comprises one historical abnormality monitoring data and a corresponding real fault type, and the mm groups of historical fault abnormal data form a fault set which is used for training a G-L convolutional neural network model.
The method comprises the steps of importing historical anomaly monitoring data in a fault set into a G-L convolutional neural network model, wherein the input end of the G-L convolutional neural network model is the historical anomaly monitoring data, and the output end of the G-L convolutional neural network model is the predicted fault type; and using the function value of the mixed loss function as a supervision signal, updating parameters in the frame network through back propagation, and obtaining a trained G-L convolutional neural network model when the function value of the mixed loss function is minimized and does not change.
In this embodiment, the fault type to which the abnormal monitoring data belongs is obtained through the G-L convolutional neural network model, so as to obtain the fault type of the operation occurrence of the charging pile, reduce the workload of the maintenance personnel for checking, the extraction unit extracts the charging pile information corresponding to the abnormal monitoring data, calibrate the position and other related information of the charging pile with the fault, reduce the maintenance time, improve the maintenance efficiency, realize the intelligent, efficient and refined management of the fire control monitoring of the charging pile through the automatic identification of the fault type and the position, improve the level and effect of the fire control supervision management, and reduce the occurrence probability and the loss degree of fire accidents.
Example III
Referring to fig. 1 and fig. 2, a third embodiment of the present invention provides an intelligent fire-fighting monitoring system and system for a charging pile.
An intelligent fire control monitoring system of a charging pile, the operation of the intelligent fire control monitoring system of the charging pile comprises the following steps:
step S1: the data acquisition unit continuously acquires an original signal generated by the charging pile in the period time, marks the original signal by the marking unit and transmits the marked original signal to the data processing module;
step S2: the data processing module is used for preprocessing the original signals transmitted by the data acquisition module to obtain real-time monitoring data, and the data processing module is used for obtaining a first data set after analyzing by the EEMD model decomposition and judgment unit;
Step S3: importing the first data set into a trained LSTM deep learning model to obtain anomaly monitoring data;
Step S4: and importing the abnormal monitoring data into a G-L convolutional neural network model to obtain the fault type of the abnormal monitoring data.
Step S5: and alarming the abnormal monitoring data and the fault type of the abnormal monitoring data and the fault occurrence position through a display module.
The intelligent fire control monitored control system of filling electric pile includes:
and a data acquisition module: the device comprises a data acquisition unit and a marking unit, wherein the data acquisition unit is used for continuously acquiring monitoring signals generated by a charging pile in a period time, and the monitoring signals comprise voltage signals, current signals and temperature signals; the marking unit is used for adding corresponding charging pile information into the data acquired by the acquisition unit;
and a storage module: for storage of data;
and a data processing module: preprocessing data transmitted by a data acquisition module, recording the preprocessed data as real-time monitoring data, decomposing voltage signals, current signals and temperature signals of the real-time monitoring data through an EEMD model, acquiring a real-time input sequence, and recording the real-time input sequence as a first data set;
An anomaly identification module: the method comprises the steps of identifying abnormal data, including an LSTM deep learning model, importing a first data set into the trained LSTM deep learning model, identifying data with abnormal characteristics, and recording the data as abnormal monitoring data;
And a fault analysis module: the system is used for analyzing fault types and positioning fault positions and comprises a G-L convolutional neural network model and an extraction unit, wherein abnormal monitoring data are imported into the G-L convolutional neural network model, fault types of the abnormal monitoring data are obtained, and the extraction unit calibrates the positions of faults through charging pile information corresponding to the abnormal monitoring data;
and a display module: the method comprises a man-machine interaction interface for checking the running state of the charging pile.
And a display module: the system comprises a man-machine interaction interface, a display module and a control module, wherein the man-machine interaction interface is used for checking the running state of the charging pile, and a manager needs to check the running state of the charging pile through the man-machine interaction interface (a display or a management system of a computer end). The man-machine interaction interface comprises the monitored working parameters of each charging pile, the fault parts, the fault types, the specific events and the like.
In the charging pile of this system overall arrangement, still include running light and bee calling organ, when the normal operation of charging pile work, show with green light source, when the abnormal operation of charging pile work, then use red light source to report to the police, and carry out audible warning through bee calling organ.
Through functions such as real-time monitoring, intelligent early warning, remote control, etc. provide powerful technical support for the fire control safety of charging stake, when the abnormal conditions is monitored, the system can send early warning signal immediately, arouses the attention of relevant personnel to take measures in advance and prevent the emergence of fire accident, real-time monitoring and early warning based on internet of things are more efficient, accurate, can effectively reduce the probability of occurrence of fire accident.
The above description is only a preferred embodiment of the present invention, and the protection scope of the present invention is not limited to the above examples, and all technical solutions belonging to the concept of the present invention belong to the protection scope of the present invention. It should be noted that modifications and adaptations to those skilled in the art without departing from the principles of the present invention are intended to be comprehended within the scope of the present invention.
Claims (10)
1. Fill intelligent fire control monitored control system of electric pile, its characterized in that, fill intelligent fire control monitored control system of electric pile includes:
and a data acquisition module: the device comprises a data acquisition unit and a marking unit, wherein the data acquisition unit is used for continuously acquiring monitoring signals generated by a charging pile in a period time, and the monitoring signals comprise voltage signals, current signals and temperature signals; the marking unit is used for adding corresponding charging pile information into the data acquired by the acquisition unit;
and a storage module: for storage of data;
and a data processing module: preprocessing data transmitted by a data acquisition module, recording the preprocessed data as real-time monitoring data, decomposing voltage signals, current signals and temperature signals of the real-time monitoring data through an EEMD model, acquiring a real-time input sequence, and recording the real-time input sequence as a first data set;
An anomaly identification module: the method comprises the steps of identifying abnormal data, including an LSTM deep learning model, importing a first data set into the trained LSTM deep learning model, identifying data with abnormal characteristics, and recording the data as abnormal monitoring data;
And a fault analysis module: the system is used for analyzing fault types and positioning fault positions and comprises a G-L convolutional neural network model and an extraction unit, wherein abnormal monitoring data are imported into the G-L convolutional neural network model, fault types of the abnormal monitoring data are obtained, and the extraction unit calibrates the positions of faults through charging pile information corresponding to the abnormal monitoring data;
and a display module: the method comprises a man-machine interaction interface for checking the running state of the charging pile.
2. The intelligent fire monitoring system of a charging pile of claim 1, wherein the EEMD model process comprises:
Setting a monitoring signal of real-time monitoring data as x (t), wherein x (t) is used for replacing a voltage signal, a current signal and a temperature signal, white noise n i (t) is added in the decomposition process of the monitoring signal x (t), and a new signal sequence is formed:
Xi(t)＝x(t)+ni(t)；
wherein i=1, … …, N is the average number of times, i is the index;
EEMD decomposition is performed for each X i (t) to obtain:
Calculating a correlation coefficient for each IMF component of the decomposition:
Where c j (t) represents the jth IMF component of EEMD decomposition; res (t) represents the monitor signal residual component decomposed by EEMD; res i (t) represents the residual component after the ith averaging that was decomposed by EMD, cov (x (t), c j (t)) represents the covariance of x (t) and c j (t), var [ x (t) ] represents the variance of x (t), C j (variance of t).
3. The intelligent fire monitoring system of a charging pile of claim 2, wherein the EEMD model further comprises:
setting a threshold value a;
if the correlation coefficient r j is more than or equal to a, the correlation between the IMF component and the monitoring signal reaches the expected requirement, and the correlation coefficient can be used as a basis for judging the characteristics of the monitoring signal;
If the image definition r j is smaller than a, the correlation between the IMF component and the monitoring signal does not meet the expected requirement, and the correlation cannot be used as a basis for judging the characteristics of the monitoring signal.
4. The intelligent fire monitoring system of a charging pile of claim 1, wherein the LSTM deep learning model comprises:
Constructing a basic framework of the LSTM deep learning model, wherein the basic framework comprises an input layer, a hidden layer, a full-connection layer and an output layer, and initializing parameters to obtain an initialized LSTM deep learning model;
Setting an LSTM deep learning model, selecting an Adam optimizer, setting an Adam algorithm to adaptively adjust the learning rate based on optimizing random gradient descent;
Acquiring gg historical data sets, wherein each set of historical data sets comprises qq characteristic data sets formed by processing historical monitoring signals through an EEMD model and abnormal data in the characteristic data sets, and the gg historical data sets form a training set;
Inputting the training set into the LSTM deep learning model in batches, and outputting a corresponding identification result of each batch; by sequentially calculating SoftMarginLoss loss function values between each batch of identification results and the abnormal data in the characteristic dataset; and updating parameters in each layer of LSTM deep learning model by using a back propagation algorithm to obtain a trained convolutional neural network deep model.
5. The intelligent fire control monitoring system of a charging pile according to claim 4, wherein the LSTM deep learning model has a loss function formula SoftMarginLoss, softMarginLoss as follows:
Where V denotes the number of samples, y v is the actual label (-1 or 1), ε is the decision function of the model, λ is the regularization parameter, ω 2 is the L2 norm of the weight.
6. The intelligent fire monitoring system of a charging pile of claim 1, wherein the G-L convolutional neural network model comprises:
constructing a basic framework of a G-L convolutional neural network model, setting an optimization target of the convolutional neural network as a fault type of abnormal monitoring data, wherein the fault type comprises an input layer, a convolutional layer, a pooling layer, an activation layer and a full-connection layer, the activation layer uses a ReLU activation function, and setting a learning rate as self-adaptive adjustment;
The G-L convolutional neural network model adopts the combination of a mean square error Loss MSELoss and a structural similarity Loss function L-Loss as a mixed Loss function;
Acquiring mm groups of historical fault abnormal data, wherein each group of historical fault abnormal data comprises one historical abnormality monitoring data and a corresponding real fault type, and the mm groups of historical fault abnormal data form a fault set;
the input end of the G-L convolutional neural network model is historical abnormal monitoring data, and the output end is a predicted fault type;
And importing the historical anomaly monitoring data in the fault set into a G-L convolutional neural network model, and obtaining a trained G-L convolutional neural network model by using the function value of the mixed loss function as a supervisory signal.
7. The intelligent fire control monitoring system of a charging pile according to claim 6, wherein the learning rate α of the G-L convolutional neural network model is adaptively updated by using an improved AdaBelief algorithm, and the next iteration number q+1 model parameter value θ q is obtained by using an improved AdaBelief algorithm:
Where m q represents a first moment estimate of the current iteration number q, k q represents a second moment estimate of the current iteration number q, θ q represents a model parameter value of the current iteration number q, α represents a learning rate, ε represents a smoothing term, and η represents an adjustment parameter.
8. The intelligent fire control monitoring system of a charging pile according to claim 6, wherein the mixing loss function of the G-L convolutional neural network model is that,
TotalLoss＝φMSELoss+(1-φ)L-Loss；
Wherein, φ is a weighing coefficient, L-Loss is a structural similarity Loss function, MSELoss represents a mean square error Loss, and the mean square error Loss function is as follows:
Wherein Yb is the real fault type corresponding to the fault abnormal data in the fault set, Is the model generation predicted fault type, and L is the number of fault anomaly data.
9. The vending machine management system of claim 8, wherein the structural similarity Loss function L-Loss is defined as:
Wherein, Is ce and/>Var (ce) represents the variance of ce,/>Representation/>Ce represents the characteristics of the true fault type signal in the fault set,/>Representing the characteristics of the model generation predictions, B is the number of true fault type signal characteristics.
10. The vending machine management system of claim 1, wherein the fault categories include short circuits, overload, cable aging, or insufficient heat dissipation, among others;
the charging pile information includes a position number of the charging pile and equipment line information.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202410144973.6A CN117942517A (en) | 2024-02-01 | 2024-02-01 | Intelligent fire control monitored control system of filling electric pile |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202410144973.6A CN117942517A (en) | 2024-02-01 | 2024-02-01 | Intelligent fire control monitored control system of filling electric pile |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN117942517A true CN117942517A (en) | 2024-04-30 |
Family
ID=90794219
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202410144973.6A Pending CN117942517A (en) | 2024-02-01 | 2024-02-01 | Intelligent fire control monitored control system of filling electric pile |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN117942517A (en) |
-
2024
- 2024-02-01 CN CN202410144973.6A patent/CN117942517A/en active Pending
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN108418841B (en) | Next-generation key message infrastructure network Security Situation Awareness Systems based on AI | |
| CN111339712B (en) | Proton exchange membrane fuel cell residual life prediction method | |
| CN110807550B (en) | Distribution transformer overload recognition and early warning method based on neural network and terminal equipment | |
| CN110188015B (en) | Host access relation abnormal behavior self-adaptive detection device and monitoring method thereof | |
| CN113420509A (en) | Wind turbine state evaluation method and device and storage medium | |
| CN116308304B (en) | New energy intelligent operation and maintenance method and system based on meta learning concept drift detection | |
| CN116090821A (en) | Power distribution network line security risk assessment method considering extreme weather | |
| CN117354171B (en) | Platform health condition early warning method and system based on Internet of things platform | |
| CN108446202A (en) | A kind of judgment method of the safe condition of calculator room equipment | |
| CN113487086A (en) | Method and device for predicting remaining service life of equipment, computer equipment and medium | |
| CN113612625A (en) | Network fault positioning method and device | |
| CN117332215A (en) | High-low voltage power distribution cabinet abnormal fault information remote monitoring system | |
| CN112836843B (en) | Base station out-of-service alarm prediction method and device | |
| CN112132195A (en) | Method for analyzing and predicting machine room fault by using Markov model | |
| CN117942517A (en) | Intelligent fire control monitored control system of filling electric pile | |
| ul Hassan et al. | Online static security assessment for cascading failure using stacked De-noising Auto-encoder | |
| CN116151799A (en) | BP neural network-based distribution line multi-working-condition fault rate rapid assessment method | |
| CN115409100A (en) | Method, device, medium and equipment for generating power accident emergency scheme | |
| CN114708718A (en) | Wind generating set temperature cluster control method, device, equipment and medium | |
| CN114970950A (en) | Fan fault alarm method and device, storage medium and electronic equipment | |
| CN114049014A (en) | Method, device and system for evaluating operation state of offshore wind turbine generator | |
| CN114091750A (en) | Power grid load abnormity prediction method, system and storage medium | |
| CN110826690A (en) | Equipment state identification method and system and computer readable storage medium | |
| CN110987062A (en) | Intelligent metering box early warning method and system based on data fusion | |
| CN114393994B (en) | Multi-target collaborative health management method and system for motorized chassis |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination |