Electric vehicle charging early warning method and system based on A-LSTM algorithm

Abstract

The present invention discloses an electric vehicle charging warning method and system based on the A‑LSTM algorithm, which belongs to the technical field of safety warning. The method comprises: first, obtaining the data of the normal charging process of the electric vehicle as historical data; second, screening the historical charging data and preprocessing it; then designing an A‑LSTM deep learning model to learn the normal charging data and construct a charging data prediction model; then establishing a dynamic threshold model to determine the warning threshold, and optimizing the threshold in combination with the relevant national charging safety regulations and the historical charging abnormality data of the electric vehicle; finally, applying the trained charging data prediction model and the warning threshold to the real-time charging monitoring of the electric vehicle, realizing the fault warning of the electric vehicle, reducing the charging hidden dangers of the electric vehicle, and improving the charging safety of the electric vehicle.

Description

Technical Field

[0001] The present invention relates to the technical field of safety warning, and more specifically to an electric vehicle charging warning method and system based on an A-LSTM algorithm. Background Art

[0002] Electric vehicles are powered by clean energy, which can alleviate the energy crisis, reduce carbon emissions and protect the environment. They are in line with my country's goal of building a resource-based society, an important theme in the development of automotive technology, and a key development target for governments and companies around the world. As a result, electric vehicles are gradually replacing traditional fuel vehicles and becoming a new type of green transportation tool with their advantages of high efficiency, energy saving and environmental friendliness.

[0003] However, with the continuous growth of electric vehicle ownership, electric vehicle spontaneous combustion and fire accidents continue to occur, causing serious economic losses to car owners and charging facility operators. The issue of charging safety has become a stumbling block to the development of electric vehicles and related industries. At the same time, in terms of the safety of electric vehicles, power batteries and charging equipment, there is no effective safety warning method and evaluation index system, which makes it difficult to provide sufficient and comprehensive protection for the charging safety of electric vehicles. How to significantly reduce the hidden dangers of electric vehicle charging and improve the charging safety of electric vehicles has become an urgent problem to be solved by the entire industry.

[0004] Therefore, how to provide an electric vehicle charging warning method and system based on the A-LSTM algorithm is a problem that technical personnel in this field urgently need to solve. Summary of the invention

[0005] In view of this, the present invention provides an electric vehicle charging warning method and system based on the A-LSTM algorithm. The present invention can obtain a variety of electric vehicle charging data, and realize the fitting of vehicle charging data based on the A-LSTM deep learning network algorithm. By means of real-time analysis and judgment of the vehicle status, the charging safety warning of the electric vehicle can be realized, and the false alarm caused by erroneous charging data can be effectively avoided, which is of great significance to vigorously promoting the development of electric vehicles and related industries.

[0006] In order to achieve the above object, the present invention provides the following technical solutions:

[0007] On the one hand, the present invention provides an electric vehicle charging early warning method based on the A-LSTM algorithm, comprising the following steps:

[0008] S100: Acquire data of a normal charging process of the electric vehicle as historical data;

[0009] S200: screening the historical charging data obtained in step 1, determining the type of electric vehicle, dividing the data into normal charging data and faulty charging data according to whether the charging status is normal or not, and pre-processing the data;

[0010] S300: using the electric vehicle charging voltage, the electric vehicle charging current, the maximum voltage of the battery pack single cell and the maximum temperature of the single cell as the output of the model, respectively, to construct a data fitting deep learning network based on the LSTM algorithm, using the processed charging data as the input of the model, training the LSTM deep learning network, and obtaining a data fitting model of the normal charging state of the electric vehicle;

[0011] S400: Adopt error-correlated linear analysis method to adaptively optimize the output of LSTM network, build A-LSTM network, improve data fitting model, and obtain the predicted value of electric vehicle charging data;

[0012] S500: Establish a dynamic threshold model to determine the warning threshold, optimize the threshold in combination with relevant national charging safety regulations and historical abnormal charging data of electric vehicles, and update the warning threshold in real time according to the difference in the status of electric vehicles;

[0013] S600: Construct a comprehensive early warning model for electric vehicle charging safety, input the real-time charging data of electric vehicles into it, screen the on-board battery type, battery pack initial SOC, battery pack initial temperature, motor initial temperature, vehicle charging type and vehicle charging time of the electric vehicle to determine the initial charging state of the electric vehicle, and monitor the charging state of the electric vehicle in real time based on the comprehensive early warning model for electric vehicle charging safety, the predicted value of electric vehicle charging data and the early warning threshold. When the charging data of the electric vehicle deviates from the set early warning threshold, a charging safety early warning is issued according to the preset early warning rules.

[0014] Preferably, the various parameters for status monitoring of the electric vehicle charging in step S100 include but are not limited to the initial SOC of the vehicle's power battery, the real-time SOC of the vehicle's power battery, the internal temperature of the vehicle's electric vehicle charger, the temperature of the vehicle's electric vehicle charging module, the maximum / minimum / average temperature of the single cells of the vehicle's onboard battery pack, the maximum / minimum / average voltage of the single cells of the vehicle's onboard battery pack, the maximum / minimum voltage / current / temperature allowed for charging of the vehicle's onboard battery pack, and other parameter information.

[0015] Preferably, in step S200, the AC charging data is preprocessed, and the specific operations are as follows:

[0016] (1) Detect outliers in the data and delete particularly abnormal data;

[0017] (2) Use interpolation to fill in missing values ​​in the data;

[0018] (3) The data is normalized using the range standardization method. The specific calculation formula is:

[0019]

[0020] Where data input is the data value after normalization; data i is the original data; data max and data min are the maximum and minimum values ​​in the original data. The processed charging data are all in [-1,1].

[0021] Preferably, the A-LSTM (Long Short Term Memory) deep learning model involved in step S300 has the following specific calculation formula: Long Short Term Memory (LSTM) is a network designed on the basis of recurrent neural network (RNN) to solve problems such as gradient vanishing and explosion. It replaces the hidden layer of the original RNN with LSTM units. In addition to the input and output gates, the LSTM unit also has a built-in forget gate that can control the amount of historical input. The activation functions of the three gates are all sigmoid functions. The value range of the sigmoid function is (0,1). The role of these three gates is equivalent to weighted learning of historical input, current input and historical output, thereby achieving the memory function of historical input and historical output;

[0022] The LSTM unit structure is shown in formula (2) to formula (7): where the candidate LSTM memory unit state value is:

[0023] C (t) =tanh(ω x c x(t) +w hc h (t-1) +b c ) (2)

[0024] Where: x (t) is the input data of the electric vehicle’s historical charging at the current moment, h (t-1) is the LSTM unit output at the previous moment, ω x With w hc Corresponding to input x (t) With the output h (t-1) The connection weight of these two items is is the memory unit reference value, b c is the bias of the network;

[0025] The value of the LSTM network input gate:

[0026] I (t) =sigmoid(ω xi x (t) +ω hi h (t-1) +ω ci C (t-1) +b i ) (3)

[0027] Where: xi ,ω hi With ω ci are the input data of the electric vehicle’s historical charging at the current moment, the LSTM unit output at the previous moment, and the connection weight of the cell unit output to the input gate at the previous moment, b i is the bias of the input gate;

[0028] The value of the forget gate of the LSTM network:

[0029] F (t) =sigmoid(ω xf x (t) +ω hf h (t-1) +ω fi C (t-1) +b f ) (4)

[0030] Where: xf ,ω hf With ω cf are the historical charging input data of the electric vehicle at the current moment, the LSTM unit output at the previous moment, and the connection weight of the cell unit output at the previous moment to the forget gate; b f is the bias of the forget gate;

[0031] Thus, the current LSTM memory cell state value is:

[0032]

[0033] In the formula It represents the multiplication operation of staying in a hotel.

[0034] The value of the LSTM network output gate:

[0035] O (t) =sigmoid(ω xo x (t) +ω ho h (t-1) +ω co C (t-1) +b o ) (6)

[0036] Where:xo ,ω ho With ω co are the connection weights of the current input, the previous LSTM unit output and the previous cell unit output to the output gate, respectively, b o is the bias of the output gate;

[0037] Combining equations (2) to (6), we can conclude that the output of the LSTM memory unit at time t is:

[0038]

[0039] In summary: The working process of LSTM can be simply understood as: given the input value x of the current time step (t) Under the control of the input gate, useful information is screened through candidate memory cells for information update of the current memory cell, while the forget gate controls whether the current memory cell can obtain the information transmitted by the previous unit. The valuable information retained by these two parts, i.e. the updated memory, will be passed to the next LSTM unit module. The output gate controls whether the information in the memory cell is passed to the hidden state for use by the output layer. (t) It will also be connected to the next LSTM unit module. The interaction and control of the three gates realizes the long-term memory of the input information.

[0040] Preferably, step S400 involves adaptive optimization of the LSTM network using an error-related linear analysis method, and its specific calculation formula is: After establishing the LSTM model, in order to further reduce the error of model prediction and improve the prediction accuracy of the algorithm, the present invention adopts an error-related linear analysis method to reduce the error. That is, a relationship (8) is established for the relationship between the historical prediction error and the input:

[0041] e pre =f(x 1 ,...,x n ) (8)

[0042] Where: e pre represents the LSTM historical prediction error; f(x 1 ,…,x n ) is a linear function of the input, (x 1 ,…,x n represents input), and its coefficients are obtained by the least squares method.

[0043] The prediction model after error correction is as follows:

[0044] g'=g(x 1+1 ,...,x n+1 )+f(x 1+1 ,...,x n+1 ) (9)

[0045] Where: g(x 1+1 ,…,x n+1 ) is the established LSTM prediction model; f(x 1+1 ,…,x n+1 ) is the error linear correction function after least squares fitting; g' is the current prediction result of the A-LSTM algorithm.

[0046] The dynamic threshold model involved in step 5 of the present invention has the following specific calculation steps: taking the fitting abnormal data with a time length of l for abnormality detection, and obtaining the error data e as shown in formula (11): (t) :

[0047] e (t) =g' (t) -x (t) (10)

[0048] The error matrix is ​​smoothed based on the SG filtering method of formula (12) to obtain the new error matrix e n

[0049]

[0050] e n =[e n(t-l) ,...,e n(t-1) ,e n(t) ] (12)

[0051] Where e is the original error data; e j is the filtered data; u i is the coefficient when filtering the i-th time series data value; N refers to the number of convolutions; the coefficient j refers to the coefficient of the original time series data set; l is the length of the filter window, which controls the smoothing effect together with the degree of the smoothing polynomial. Observe the smoothed results and set the initial warning threshold k through equations (14) to (17).

[0052] k=μ(e n )+zσ(e n ) (13)

[0053]

[0054] Δμ(e n )=μ(e n )-μ({e n ∈e n ,e n <k}) (15)

[0055] Δσ(e n )=σ(e n) - σ({e n ∈e n , e n <k}) (16)

[0056] e a ={e n ∈e n , e n >k} (17)

[0057] where k is the threshold vector, μ(e n ) is the expectation of e n , σ(e n ) is the standard deviation of e n , e a is the outlier, and E seq is the continuous sequence in e a .

[0058] Thus, the initial threshold k can be determined. If the value in the outlier is not much different from the maximum value of the normal sequence, they may just be normal jitters rather than real outliers. To prevent false outliers, the maximum values of the abnormal data and the normal data are selected and arranged in descending order to form e max , and the abnormal correction is performed using Equation (18) to judge the value of r i to update the threshold (if r i > p, then e (i-1) is still an outlier; if r i < p, then e (i) and subsequent values are reclassified as normal values, and the setting range of p is: 0.2 > p > 0.05).

[0059] r i =(e max(i-1) - e max(i) ) / e max(i-1) (18)

[0060] Preferably, in step S600, a safety warning for the electric vehicle charging process is realized. An integrated safety warning model for electric vehicle charging is constructed and real-time vehicle charging data is input into it. The basic information of the electric vehicle is screened to judge the initial charging state of the electric vehicle. According to the electric vehicle information, charging prediction data and charging warning thresholds are input. Different warning models are selected according to the charging state of the electric vehicle. The charging state of the electric vehicle is monitored in real time. When the electric vehicle charging data deviates from the set warning threshold, a charging safety warning is given according to relevant warning rules, and the power supply of the electric vehicle is cut off when necessary to prevent it from catching fire.

[0061] Preferably, the basic information of the electric vehicle includes: the vehicle's onboard battery type, battery pack initial SOC, battery pack initial temperature, motor initial temperature, vehicle charging type and vehicle charging time, etc.

[0062] On the other hand, the present invention provides an electric vehicle charging early warning system based on the A-LSTM algorithm, comprising:

[0063] An acquisition module is used to acquire data of a normal charging process of an electric vehicle and mark it as historical charging data;

[0064] A preprocessing module, connected to the acquisition module, for preprocessing the historical charging data;

[0065] A construction module, connected to the preprocessing module, is used to construct a data fitting deep learning network model based on the LSTM algorithm using the electric vehicle charging voltage, the electric vehicle charging current, the maximum voltage of the battery pack single cell and the maximum temperature of the single cell as the output of the model, and to train the LSTM deep learning network using the preprocessed historical charging data as the input of the model to obtain a data fitting model of the normal charging state of the electric vehicle;

[0066] A processing module, connected to the construction module, is used to adopt an error-related linear analysis method to perform adaptive optimization on the output of the data fitting model, construct an A-LSTM network model, improve the data fitting model, and obtain a predicted value of electric vehicle charging data;

[0067] An updating module, connected to the processing module, for establishing a dynamic threshold model to determine and optimize the warning threshold, and updating the warning threshold in real time according to the difference in the status of the electric vehicle;

[0068] The early warning module is connected to the processing module and the updating module, and is used to construct a comprehensive early warning model for electric vehicle charging safety, input the real-time charging data of the electric vehicle into the model, screen the on-board battery type, battery pack initial SOC, battery pack initial temperature, motor initial temperature, vehicle charging type and vehicle charging time of the electric vehicle to determine the initial charging state of the electric vehicle, and monitor the charging state of the electric vehicle in real time based on the comprehensive early warning model for electric vehicle charging safety, the predicted value of the electric vehicle charging data and the early warning threshold. When the charging data of the electric vehicle deviates from the set early warning threshold, a charging safety early warning is issued according to a preset early warning rule.

[0069] It can be seen from the above technical solution that compared with the prior art, the present invention discloses an electric vehicle charging warning method and system based on the A-LSTM algorithm, which obtains the electric vehicle charging history data and performs data screening and preprocessing on it, and then designs an A-LSTM deep learning algorithm to perform algorithmic modeling on the electric vehicle charging history data, so as to achieve the fitting of the normal charging state of the electric vehicle to predict the charging data of the vehicle; for the warning threshold, a dynamic threshold optimization method is designed to achieve dynamic update of the vehicle charging state warning threshold, which can effectively warn for different vehicle states, not only can it further enhance the accuracy of the electric vehicle charging safety warning results, but also can eliminate the false warning caused by erroneous data during data transmission. BRIEF DESCRIPTION OF THE DRAWINGS

[0070] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are only embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on the provided drawings without paying creative work.

[0071] Figure 1 It is a flow chart of the electric vehicle charging safety early warning model of the present invention;

[0072] Figure 2 It is a schematic diagram of the LSTM deep network structure of the present invention;

[0073] Figure 3 It is a structural principle diagram of the A-LSTM deep learning algorithm of the present invention;

[0074] Figure 4 It is a schematic diagram of the system structure of the present invention. DETAILED DESCRIPTION

[0075] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0076] See attached Figure 1 As shown, the embodiment of the present invention discloses an electric vehicle charging warning method based on the A-LSTM algorithm, comprising the following steps:

[0077] S100: Collect the charging history data of electric vehicles, including but not limited to the initial SOC of the vehicle's power battery, the real-time SOC of the vehicle's power battery, the internal temperature of the vehicle's charger, the temperature of the vehicle's charging module, the highest / lowest / average temperature of the single cells of the vehicle's onboard battery pack, the highest / lowest / average voltage of the single cells of the vehicle's onboard battery pack, the highest / lowest voltage / current / temperature allowed for charging of the vehicle's onboard battery pack and other parameter information.

[0078] S200: Preprocess the vehicle charging data of the mobile phone according to the process (1)-(3), and the specific operation is as follows:

[0079] (1) Detect outliers in the data and delete particularly abnormal data;

[0080] (2) Use interpolation to fill in missing values ​​in the data;

[0081] (3) The data is normalized using the range standardization method. The specific calculation formula is:

[0082]

[0083] Where data input is the data value after normalization; data i is the original data; data max and data min are the maximum and minimum values ​​in the original data. The processed charging data are all in [-1,1].

[0084] Specifically, the vehicle in this example refers to an electric vehicle.

[0085] In a specific embodiment, S300: using the electric vehicle charging voltage, the electric vehicle charging current, the maximum voltage of the battery pack single cell and the maximum temperature of the single cell as the output of the model, respectively, to construct a data fitting deep learning network model based on the LSTM algorithm, using the preprocessed historical charging data as the input of the model, training the LSTM deep learning network, and obtaining a data fitting model of the normal charging state of the electric vehicle;

[0086] For details, see the attached Figure 2 As shown in the figure, the data fitting deep learning network model based on the LSTM algorithm built by S300 includes: input gate, output gate and a forget gate built into the LSTM unit that can control the amount of historical input;

[0087] Among them, the activation functions of the three gates are all sigmoid functions, and the value range of the sigmoid function is (0,1).

[0088] Specifically, the LSTM (Long Short Term Memory) deep learning model involved in S300 has the following specific calculation formula: Long Short Term Memory (LSTM) is a network designed on the basis of recurrent neural network (RNN) to solve problems such as gradient disappearance and explosion. It replaces the hidden layer of the original RNN with LSTM units. In addition to the input and output gates, the LSTM unit also has a built-in forget gate that can control the amount of historical input. The activation functions of the three gates are all sigmoid functions. The value range of the sigmoid function is (0,1). The role of these three gates is equivalent to weighted learning of historical input, current input and historical output, thereby achieving the memory function of historical input and historical output;

[0089] The LSTM unit structure is shown in formula (2) to formula (7): where the candidate LSTM memory unit state value is:

[0090] C (t) =tanh(ω x c x(t) +w hc h (t-1) +b c ) (2)

[0091] Where: x (t) is the input data of the electric vehicle’s historical charging at the current moment, h (t-1) is the LSTM unit output at the previous moment, ω x With w hc Corresponding to input x (t) With the output h (t-1) The connection weight of these two items is is the memory unit reference value, b c is the bias of the network;

[0092] The value of the LSTM network input gate:

[0093] I (t) =sigmoid(ω xi x (t) +ω hi h (t-1) +ω ci C (t-1) +b i ) (3)

[0094] Where: xi ,ω hi With ω ciare the input data of the electric vehicle’s historical charging at the current moment, the LSTM unit output at the previous moment, and the connection weight of the cell unit output to the input gate at the previous moment, b i is the bias of the input gate;

[0095] The value of the forget gate of the LSTM network:

[0096] F (t) =sigmoid(ω xf x (t) +ω hf h (t-1) +ω fi C (t-1) +b f ) (4)

[0097] Where: xf ,ω hf With ω cf are the historical charging input data of the electric vehicle at the current moment, the LSTM unit output at the previous moment, and the connection weight of the cell unit output at the previous moment to the forget gate; b f is the bias of the forget gate;

[0098] Thus, the current LSTM memory cell state value is:

[0099]

[0100] In the formula It represents the multiplication operation of staying in a hotel.

[0101] The value of the LSTM network output gate:

[0102] O (t) =sigmoid(ω xo x (t) +ω ho h (t-1) +ω co C (t-1) +b o ) (6)

[0103] Where: xo ,ω ho With ω co are the connection weights of the current input, the previous LSTM unit output and the previous cell unit output to the output gate, respectively, b o is the bias of the output gate;

[0104] Combining equations (2) to (6), we can conclude that the output of the LSTM memory unit at time t is:

[0105]

[0106] In summary: The working process of LSTM can be simply understood as: given the input value x of the current time step (t) Under the control of the input gate, useful information is screened through candidate memory cells for information update of the current memory cell, while the forget gate controls whether the current memory cell can obtain the information transmitted by the previous unit. The valuable information retained by these two parts, i.e. the updated memory, will be passed to the next LSTM unit module. The output gate controls whether the information in the memory cell is passed to the hidden state for use by the output layer. (t) It will also be connected to the next LSTM unit module. The interaction and control of the three gates realize the long-term memory of the input information. In this way, the vehicle charging voltage, vehicle charging current, battery pack single cell maximum voltage and single cell maximum temperature are used as the output of the model to build a data fitting deep learning network based on the LSTM algorithm. The processed charging data is used as the input of the model to train the LSTM deep learning network and obtain the data fitting model of the normal charging state of the electric vehicle.

[0107] S400: Adopt the error-correlated linear analysis method to perform adaptive optimization on the output of the LSTM network, build an A-LSTM network, improve the data fitting model, and obtain the predicted value of electric vehicle charging data.

[0108] In a specific embodiment, see the attached Figure 3 As shown in the figure, it is a structural principle diagram of the A-LSTM deep learning algorithm. After the LSTM model is established, in order to further reduce the error of model prediction and improve the prediction accuracy of the algorithm, the present invention adopts the error correlation linear analysis method to reduce the error. That is, the relationship between the historical prediction error and the input is established by equation (8):

[0109] e pre =f(x 1 ,...,x n )(8)

[0110] Where: e pre represents the LSTM historical prediction error; f(x 1 ,…,x n ) is a linear function of the input, (x 1 ,…,x n represents input), and its coefficients are obtained by the least squares method.

[0111] The prediction model after error correction is as follows:

[0112] g'=g(x 1+1 ,...,x n+1 )+f(x 1+1 ,...,x n+1 )(9)

[0113] Where: g(x 1+1 ,…,x n+1 ) is the established LSTM prediction model; f(x 1+1 ,…,x n+1 ) is the error linear correction function after least squares fitting; g' is the current prediction result of the A-LSTM algorithm.

[0114] S500: Establish a dynamic threshold model. The specific calculation steps are: take data with a time length of l to perform anomaly detection and obtain the error data e as shown in formula (11): (t) :

[0115] e (t) =g' (t) -x (t) (10)

[0116] The error matrix is ​​smoothed based on the SG filtering method of formula (12) to obtain the new error matrix e n

[0117]

[0118] e n =[e n(t-l) ,...,e n(t-1) ,e n(t) ] (12)

[0119] Where e is the original error data; e j is the filtered data; u i is the coefficient when filtering the i-th time series data value; N refers to the number of convolutions; the coefficient j refers to the coefficient of the original time series data set; l is the length of the filter window, which controls the smoothing effect together with the degree of the smoothing polynomial. Observe the smoothed results and set the initial warning threshold k through equations (14) to (17).

[0120] k=μ(e n )+zσ(e n ) (13)

[0121]

[0122] Δμ(e n )=μ(e n )-μ({e n ∈e n ,e n <k}) (15)

[0123] Δσ(e n )=σ(e n )-σ({en ∈e n ,e n <k}) (16)

[0124] e a ={e n ∈e n ,e n >k} (17)

[0125] Wherein, k is a threshold vector, μ(e n ) is the expectation of e n , σ(e n ) is the standard deviation of e n , e a is an outlier, and E seq is a continuous sequence in e a .

[0126] In this way, the initial threshold k can be determined. If the value in the outlier is not much different from the maximum value of the normal sequence, they may just be normal jitters rather than real outliers. To prevent false outliers, the maximum values of the abnormal data and the normal data are selected and arranged in descending order to form e max , and the outlier correction is performed using Equation (18) to judge the value of r i to update the threshold (if r i > p, then e (i-1) is still an outlier; if r i < p, then e (i) and subsequent values are reclassified as normal values, and the setting range of p is: 0.2 > p > 0.05).

[0127] r i =(e max(i-1) -e max(i) ) / e max(i-1) (18)

[0128] S600: Safety warning during the electric vehicle charging process. First, construct a comprehensive safety warning model for electric vehicle charging, input the real-time charging data of the vehicle, screen data such as the in-vehicle battery type, initial SOC of the battery pack, battery pack temperature, and charging time of the vehicle to judge the initial charging state of the vehicle, input the charging prediction data and charging warning threshold according to the vehicle information, select different warning models according to the charging state of the vehicle, monitor the charging state of the vehicle in real time. When the electric vehicle charging data deviates from the set warning threshold, conduct a charging safety warning according to relevant warning rules, and cut off the power supply of the electric vehicle when necessary to prevent it from catching fire.

[0129] On the other hand, as shown in the attached Figure 4 , this embodiment also discloses an electric vehicle charging warning system based on the A-LSTM algorithm, including:

[0130] An acquisition module is used to acquire data of a normal charging process of an electric vehicle and mark it as historical charging data;

[0131] A preprocessing module, connected to the acquisition module, for preprocessing the historical charging data;

[0132] A construction module is connected to the preprocessing module, and is used to construct a data fitting deep learning network model based on the LSTM algorithm by taking the electric vehicle charging voltage, the electric vehicle charging current, the maximum voltage of the battery pack single cell and the maximum temperature of the single cell as the output of the model, and to train the LSTM deep learning network by taking the preprocessed historical charging data as the input of the model, so as to obtain a data fitting model of the normal charging state of the electric vehicle;

[0133] The processing module is connected to the construction module and is used to adopt the error-related linear analysis method to perform adaptive optimization on the output of the data fitting model, build an A-LSTM network model, improve the data fitting model, and obtain the predicted value of the electric vehicle charging data;

[0134] An updating module, connected to the processing module, is used to establish a dynamic threshold model to determine and optimize the warning threshold, and to update the warning threshold in real time according to the difference in the status of the electric vehicle;

[0135] The early warning module is connected with the processing module and the updating module, and is used to construct a comprehensive early warning model for electric vehicle charging safety, input the real-time charging data of the electric vehicle, screen the on-board battery type, battery pack initial SOC, battery pack initial temperature, motor initial temperature, vehicle charging type and vehicle charging time of the electric vehicle to judge the initial charging state of the electric vehicle, and monitor the charging state of the electric vehicle in real time based on the comprehensive early warning model for electric vehicle charging safety, the predicted value of the electric vehicle charging data and the early warning threshold. When the charging data of the electric vehicle deviates from the set early warning threshold, a charging safety early warning is issued according to the preset early warning rules.

[0136] The present invention discloses an electric vehicle charging early warning method and system based on the A-LSTM algorithm, which obtains the electric vehicle charging history data and performs data screening and preprocessing on it, and then designs the A-LSTM deep learning algorithm to perform algorithm modeling on the electric vehicle charging history data, so as to achieve the fitting of the normal charging state of the electric vehicle to predict the charging data of the vehicle; designs a dynamic threshold optimization method for the early warning threshold, realizes the dynamic update of the vehicle charging state early warning threshold, and can effectively provide early warning for different vehicle states, which can not only further enhance the accuracy of the electric vehicle charging safety early warning result, but also eliminate the false early warning caused by the erroneous data in the data transmission process.

[0137] In this specification, each embodiment is described in a progressive manner, and each embodiment focuses on the differences from other embodiments. The same or similar parts between the embodiments can be referred to each other. For the device disclosed in the embodiment, since it corresponds to the method disclosed in the embodiment, the description is relatively simple, and the relevant parts can be referred to the method part.

[0138] The above description of the disclosed embodiments enables one skilled in the art to implement or use the present invention. Various modifications to these embodiments will be apparent to one skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of the present invention. Therefore, the present invention will not be limited to the embodiments shown herein, but rather to the widest scope consistent with the principles and novel features disclosed herein.

Claims

1. An electric vehicle charging warning method based on A-LSTM algorithm, It is characterized in that The steps include: S100: Acquire data of a normal charging process of the electric vehicle and mark it as historical charging data; S200: pre-processing the historical charging data; S300: Using the electric vehicle charging voltage, the electric vehicle charging current, the maximum voltage of the battery pack single cell and the maximum temperature of the single cell as the output of the model, respectively, to construct a data fitting deep learning network model based on the LSTM algorithm, using the preprocessed historical charging data as the input of the model, training the LSTM deep learning network, and obtaining a data fitting model of the normal charging state of the electric vehicle; S400: Adopting the error-related linear analysis method to perform adaptive optimization on the output of the data fitting model, constructing an A-LSTM network model, improving the data fitting model, and obtaining the predicted value of the electric vehicle charging data; S500: Establish a dynamic threshold model to determine and optimize the warning threshold, and update the warning threshold in real time according to the difference in the status of the electric vehicle; S600: Construct a comprehensive early warning model for electric vehicle charging safety, input the real-time charging data of electric vehicles into it, screen the on-board battery type, battery pack initial SOC, battery pack initial temperature, motor initial temperature, vehicle charging type and vehicle charging time of the electric vehicle to determine the initial charging state of the electric vehicle, and monitor the charging state of the electric vehicle in real time based on the comprehensive early warning model for electric vehicle charging safety, the predicted value of electric vehicle charging data and the early warning threshold. When the charging data of the electric vehicle deviates from the set early warning threshold, a charging safety early warning is issued according to the preset early warning rules.

2. According to the electric vehicle charging early warning method based on the A-LSTM algorithm according to claim 1, It is characterized in that The data of the normal charging process of the electric vehicle acquired by S100 includes: The initial SOC of the vehicle's power battery, the real-time SOC of the vehicle's power battery, the internal temperature of the vehicle's electric vehicle charger, the temperature of the vehicle's electric vehicle charging module, the maximum / minimum / average temperature of the vehicle's onboard battery pack cells, the maximum / minimum / average voltage of the vehicle's onboard battery pack cells, and the maximum / minimum voltage / current / temperature parameter information allowed for charging the vehicle's onboard battery pack.

3. According to the electric vehicle charging early warning method based on the A-LSTM algorithm according to claim 1, It is characterized in that S200: pre-processing the historical charging data, specifically including: S210: Perform outlier detection on the data and delete particularly abnormal data; S220: Use interpolation to fill missing values ​​in the data; S230: Use the range standardization method to normalize the data. The specific calculation formula is: Where data input is the data value after normalization; data i is the original data; data max and data min are the maximum and minimum values ​​in the original data; the processed charging data is located in [-1,1].

4. According to claim 1, an electric vehicle charging early warning method based on A-LSTM algorithm, It is characterized in that The data fitting deep learning network model based on the LSTM algorithm constructed by the S300 includes: an input gate, an output gate, and a forget gate built into the LSTM unit that can control the historical input amount; Among them, the activation functions of the three gates are all sigmoid functions, and the value range of the sigmoid function is (0, 1).

5. A method for warning of electric vehicle charging based on the A-LSTM algorithm according to claim 4, wherein, the S300 specifically includes: S310: Construct an LSTM unit, wherein the candidate LSTM memory cell state value, the specific formula is: In the formula, x (t) is the input data of the electric vehicle’s historical charging at the current moment, h (t-1) is the LSTM unit output at the previous moment, ω x With w hc Corresponding to input x (t) With the output h (t-1) The connection weight of these two items is is the memory unit reference value, b c is the bias of the network; S320: Calculate the value of the input gate of the LSTM network: I (t) =sigmoid(ω xi x (t) +ω hi h (t-1) +ω ci C (t-1) +b i ) In the formula, ω xi ,ω hi With ω ci are the input data of the electric vehicle’s historical charging at the current moment, the LSTM unit output at the previous moment, and the connection weight of the cell unit output to the input gate at the previous moment, b i is the bias of the input gate; S330: Calculate the value of the forget gate of the LSTM network: F (t) =sigmoid(ω xf x (t) +ω hf h (t-1) +ω fi C (t-1) +b f ) In the formula, ω xf ,ω hf With ω cf are the historical charging input data of the electric vehicle at the current moment, the LSTM unit output at the previous moment, and the connection weight of the cell unit output at the previous moment to the forget gate; b f is the bias of the forget gate; S340: Calculate the current LSTM memory cell state value: In the formula It means the operation of finding the product of staying in a hotel; S350: Calculate the value of the output gate of the LSTM network: The (t) =sigmoid(ω xo x (t) +oh ho h (t-1) +oh co C (t-1) +b o ) Where: xo ,ω ho With ω co are the connection weights of the current input, the previous LSTM unit output and the previous cell unit output to the output gate, respectively, b o is the bias of the output gate; S360: Integrate steps S310 - S350: to obtain the output of the LSTM memory cell at time t as:

6. A method for warning of electric vehicle charging based on the A-LSTM algorithm according to claim 1, wherein, the S400 performs adaptive optimization on the output of the data fitting model by adopting an error-related linear analysis method, constructs an A-LSTM network model, improves the data fitting model, and obtains the predicted value of the electric vehicle charging data, specifically including: S410: Establish a relational expression for the relationship between the historical prediction error and the input: e pre =f(x 1 ,...,x n ) In the formula, e pre represents the LSTM historical prediction error; f(x 1 ,…,x n ) is a linear function of the input, where x 1 ,…,x n Represents input; S420: The prediction model after error correction is as follows: g'=g(x 1+1 ,...,x n+1 )+f(x 1+1 ,...,x n+1 ) In the formula, g(x 1+1 ,…,x n+1 ) is the established LSTM prediction model; f(x 1+1 ,…,x n+1 ) is the error linear correction function after least squares fitting; g' is the current prediction result of the A-LSTM algorithm.

7. A method for warning of electric vehicle charging based on the A-LSTM algorithm according to claim 1, wherein, the specific calculation steps for the S500 to establish a dynamic threshold model are: S510: Take data with a time length of l to perform anomaly detection and obtain error data e (t) : e (t) =g′ (t) -x (t) S520: Smoothing the error matrix based on SG filtering to obtain a new error matrix e n : And n =[and n(t-l) ,...,And n(t-1) ,And n(t) ] Where, e is the original error data; e j is the filtered data; u i is the coefficient when filtering the i-th time series data value; N refers to the number of convolutions; coefficient j refers to the coefficient of the original time series data set; l is the length of the filter window, which controls the smoothing effect together with the degree of the smoothing polynomial; analyze the smoothed results and set the initial warning threshold k through the following formula; k=µ(e n )+zσ(e n ) Dm(e n )=μ(e n )-μ({e n ∈e n ,e n (k) Ds(e n )=σ(e n )-σ({e n ∈e n ,e n (k) And a ={and n ∈e n ,And n >k} Where k is the threshold vector, μ(e n ) is e n The expectation of σ(e n ) is e n The standard deviation of a is an outlier, E seq for e a Continuous sequence in ; S530: Select the maximum values ​​of abnormal data and normal data and arrange them in descending order to form e max , use the following formula to correct the anomaly: r i =(and max(i-1) -And max(i) ) / And max(i-1) S540: Determine r i The value of is used to update the threshold.

8. A method for warning of electric vehicle charging based on the A-LSTM algorithm according to claim 1, wherein, S540: Determine r i The value of is used to update the threshold, including: S541: If ri > p, then e(i - 1) is still an outlier; S542: If ri < p, then e(i) and subsequent values are reclassified as normal values; wherein, the setting range of the parameter p is: 0.2 > p > 0.

05.

9. An electric vehicle charging warning system based on the A-LSTM algorithm using the method for warning of electric vehicle charging based on the A-LSTM algorithm according to any one of claims 1 - 8, wherein, it includes: an acquisition module, configured to acquire data during the normal charging process of an electric vehicle and calibrate it as historical charging data; a preprocessing module, connected to the acquisition module, configured to preprocess the historical charging data; a construction module, connected to the preprocessing module, configured to construct a data fitting deep learning network model based on the LSTM algorithm with the charging voltage of the electric vehicle, the charging current of the electric vehicle, the highest voltage of the single battery in the battery pack, and the highest temperature of the single battery as the outputs of the model respectively, and use the preprocessed historical charging data as the input of the model to train the LSTM deep learning network to obtain a data fitting model for the normal charging state of the electric vehicle; a processing module, connected to the construction module, configured to perform adaptive optimization on the output of the data fitting model by adopting an error-related linear analysis method, construct an A-LSTM network model, improve the data fitting model, and obtain the predicted value of the electric vehicle charging data; An updating module, connected to the processing module, for establishing a dynamic threshold model to determine and optimize the warning threshold, and updating the warning threshold in real time according to the difference in the status of the electric vehicle; The early warning module is connected to the processing module and the updating module, and is used to construct a comprehensive early warning model for electric vehicle charging safety, input the real-time charging data of the electric vehicle into the model, screen the on-board battery type, battery pack initial SOC, battery pack initial temperature, motor initial temperature, vehicle charging type and vehicle charging time of the electric vehicle to determine the initial charging state of the electric vehicle, and monitor the charging state of the electric vehicle in real time based on the comprehensive early warning model for electric vehicle charging safety, the predicted value of the electric vehicle charging data and the early warning threshold. When the charging data of the electric vehicle deviates from the set early warning threshold, a charging safety early warning is issued according to a preset early warning rule.

Images