Abnormal state detection method, device, apparatus and storage medium

Abstract

The application relates to an abnormal state detection method, device and equipment and a storage medium. When the abnormal state is detected, the application first determines an abnormal mode to be detected, obtains corresponding to-be-detected data of the abnormal mode from a to-be-detected vehicle, the to-be-detected data comprising first to-be-detected data and second to-be-detected data; the first to-be-detected data and the second to-be-detected data are input into a pre-trained detection network to obtain prediction data; a target loss value is calculated according to the prediction data and the first to-be-detected data; the target loss value and an abnormal threshold are used to detect an abnormal state; it can be seen that the application can detect the abnormal state of the vehicle according to the target loss value; if the target loss value is too large, it indicates that the prediction data and the input first detection data have a large difference, so the vehicle is determined to have an abnormal state; in this way, the application can effectively identify the potential abnormal state in the case that abnormal state samples are scarce.

Description

TECHNICAL FIELD

[0001] The present application relates to the field of abnormal state detection, and particularly relates to an abnormal state detection method and device, equipment and a storage medium. BACKGROUND

[0002] In the field of new energy vehicles, abnormal states are relatively rare compared with normal operating states. The abnormal states can be abnormal events occurring in parts such as electrical systems, power systems and battery management of new energy vehicles, such as battery failure, motor abnormality and the like. If these abnormal states are not discovered and handled in time, they can cause serious safety problems. However, the types of abnormal states are diverse and difficult to collect comprehensively. When training based on abnormal samples by using a traditional machine learning method, the data set has a serious imbalance problem due to the fact that the number of abnormal samples is much less than that of normal samples, so it is difficult to accurately and efficiently detect abnormal conditions of vehicles.

[0003] Therefore, how to accurately detect abnormal states of vehicles is a problem to be solved by those skilled in the art. SUMMARY

[0004] The present application provides an abnormal state detection method, device, equipment and storage medium to accurately detect abnormal states of vehicles.

[0005] In a first aspect, the present application provides an abnormal state detection method, comprising:

[0006] determining an abnormal mode to be detected;

[0007] obtaining, from a vehicle to be detected, to-be-detected data corresponding to the abnormal mode; the to-be-detected data comprising first to-be-detected data and second to-be-detected data;

[0008] inputting the first to-be-detected data and the second to-be-detected data into a pre-trained detection network to obtain prediction data;

[0009] calculating a target loss value according to the prediction data and the first to-be-detected data;

[0010] detecting an abnormal state by using the target loss value and an abnormal threshold.

[0011] Optionally, the step of inputting the first to-be-detected data and the second to-be-detected data into a pre-trained detection network to obtain prediction data comprises:

[0012] inputting the first to-be-detected data and the second to-be-detected data into an encoder of the detection network to obtain a high-dimensional space feature vector;

[0013] input the high-dimensional space feature vector and the second to-be-detected data into a decoder of the detection network to obtain predicted data.

[0014] Optionally, the target loss value is calculated according to the predicted data and the first to-be-detected data, including:

[0015] a reconstruction loss value is calculated by using a reconstruction loss determination rule, the predicted data and the first to-be-detected data;

[0016] The reconstruction loss determination rule is:

[0017] l recom (θ,ζ,x0,x1)=MSE(Decoder ζ (z,x1),x0);

[0018] wherein, l recom is a reconstruction loss value, θ is an encoder coefficient, ζ is a decoder coefficient, x0 is the first to-be-detected data, x1 is the second to-be-detected data, z is a high-dimensional space feature vector, z=Encoder θ (x0,x1), Encoder is an encoder, Decoder is a decoder, MSE is a mean square error loss function, Decoder ζ (z,x1) is the predicted data.

[0019] Optionally, if the abnormal mode is a power battery abnormal mode, after the reconstruction loss value is calculated by using the reconstruction loss determination rule, the predicted data and the first to-be-detected data, the method further includes:

[0020] a regularization loss value is calculated by using a regularization loss determination rule and the high-dimensional space feature vector;

[0021] The regularization loss determination rule is:

[0022]

[0023] wherein, l reg is a regularization loss value, μ represents a mean value, σ represents a mean square error, tr represents a trace number, and log represents a logarithm.

[0024] Optionally, if the abnormal mode is a power battery abnormal mode, after the reconstruction loss value is calculated by using the reconstruction loss determination rule, the predicted data and the first to-be-detected data, the method further includes:

[0025] The high-dimensional space feature vector is input into a decoder of the detection network, and an output result is input into a multi-layer perceptron of the detection network to obtain a predicted mileage value;

[0026] A mileage loss value is calculated using a mileage loss determination rule, the predicted mileage value, and the actual mileage value;

[0027] The mileage loss determination rule is:

[0028] l mile (θ, x0, x1) = MSE(MLP ξ (Decoder ζ (z, θ)), mile);

[0029] wherein, l mile is a mileage loss value, MLP is a multi-layer perceptron, ξ is a multi-layer perceptron parameter, and mile is an actual mileage value.

[0030] Optionally, the detecting an abnormal state using the target loss value and an abnormal threshold value comprises:

[0031] Determining a target loss value of each signal sample in the to-be-detected data;

[0032] Selecting a target loss value to be processed from the target loss values of all signal samples according to an abnormal sample proportion coefficient;

[0033] Calculating an average loss value of all target loss values to be processed;

[0034] Determining whether the average loss value is greater than an abnormal threshold value;

[0035] If yes, it is determined that an abnormal state corresponding to the abnormal pattern is detected.

[0036] Optionally, after the determining that the abnormal state corresponding to the abnormal pattern is detected, the method further comprises:

[0037] Displaying the abnormal state in a digital twin model and generating an early warning signal;

[0038] Sending the early warning signal to a predetermined terminal device.

[0039] In a second aspect, the present application provides an abnormal state detection device, comprising:

[0040] A determination module configured to determine an abnormal pattern to be detected;

[0041] An acquisition module configured to acquire to-be-detected data corresponding to the abnormal pattern from a to-be-detected vehicle; the to-be-detected data comprises first to-be-detected data and second to-be-detected data;

[0042] An input module configured to input the first to-be-detected data and the second to-be-detected data into a pre-trained detection network to obtain predicted data;

[0043] a calculation module, configured to calculate a target loss value according to the prediction data and the first to-be-detected data;

[0044] a detection module, configured to detect an abnormal state by using the target loss value and an abnormal threshold.

[0045] In a third aspect, the present application provides an electronic device, comprising:

[0046] a processor, a memory, and a computer program stored in the memory and executable on the processor, wherein the processor executes the steps of the abnormal state detection method according to the computer program.

[0047] In a fourth aspect, the present application provides a computer storage medium, which stores computer executable instructions for executing the steps of the abnormal state detection method.

[0048] Compared with the prior art, the above technical solution provided by the embodiments of the present application has the following advantages: when detecting an abnormal state, the embodiments of the present application first determine an abnormal mode to be detected, and obtain to-be-detected data corresponding to the abnormal mode from a vehicle to be detected, the to-be-detected data including first to-be-detected data and second to-be-detected data; the first to-be-detected data and the second to-be-detected data are input into a pre-trained detection network to obtain prediction data; a target loss value is calculated according to the prediction data and the first to-be-detected data; and an abnormal state is detected by using the target loss value and an abnormal threshold; it can be seen that the present application can detect the abnormal state of a vehicle according to the target loss value, if the target loss value is too large, it means that the prediction data and the input first to-be-detected data have a large difference, and thus it is determined that the vehicle has an abnormal state; in this way, the present application can effectively identify a potential abnormal state in the case of a lack of abnormal state samples. BRIEF DESCRIPTION OF DRAWINGS

[0049] The accompanying drawings, which are incorporated herein and form part of the specification, illustrate embodiments consistent with the present application and, together with the description, further serve to explain the principles behind the application.

[0050] In order to more clearly illustrate the technical solutions of the embodiments of the present application or the prior art, the drawings needed in the embodiments or the prior art description will be briefly introduced as follows, and obviously, other drawings can also be obtained by those skilled in the art without any creative labor.

[0051] One or more embodiments are exemplarily illustrated by pictures in the drawings corresponding thereto, and these exemplary illustrations do not constitute a limitation on the embodiments, and elements with the same reference numerals in the drawings represent similar elements, unless otherwise specified, and the drawings do not constitute a proportional limitation.

[0052] Figure 1 A flow chart of an abnormal state detection method provided for an embodiment of the present application is shown in the figure;

[0053] Figure 2 A flow chart of another abnormal state detection method provided for an embodiment of the present application is shown in the figure;

[0054] Figure 3 A flow chart of a detection method based on a few-sample knowledge flexible migration network provided for an embodiment of the present application is shown in the figure;

[0055] Figure 4 A flow chart of an abnormal state detection method provided for an embodiment of the present application is shown in the figure;

[0056] Figure 5 A structure diagram of an abnormal state detection device provided for an embodiment of the present application is shown in the figure;

[0057] Figure 6 A structure diagram of an electronic device provided for an embodiment of the present application is shown in the figure. DETAILED DESCRIPTION

[0058] To make the objectives, technical solutions and advantages of the embodiments of the present application clearer, the technical solutions in the embodiments of the present application will be described below in connection with the drawings in the embodiments of the present application. Obviously, the described embodiments are only a part of the embodiments of the present application, rather than all the embodiments of the present application. Based on the embodiments in the present application, all other embodiments obtained by those of ordinary skill in the art without creative work fall within the scope of protection of the present application.

[0059] The following disclosure provides many different embodiments, or examples, for implementing different structures of the present application. For the purpose of simplicity, the components and arrangements of the examples that are described herein are shown and described separately. It is to be understood, however, that the implementations of the present application are not limited to those examples; indeed, many modifications are possible. In addition, the present application is well suited to achieving this object and others through the techniques of the embodiments of the present application along with any other components or arrangements not specifically described herein. Accordingly, the disclosure as set forth hereof describes illustrative embodiments that are examples in nature and are not intended to be limiting.

[0060] It should be noted that, in the current new energy vehicle driving process, there are few abnormal samples, and the sample collection and sorting cost is high, and the traditional abnormal detection algorithm needs a large number of samples, so it is difficult to accurately and efficiently detect the abnormal condition of the new energy vehicle and take reasonable control measures. Therefore, the embodiments of the present application disclose an abnormal state detection method, device, equipment and storage medium, which can accurately detect the abnormal state of the vehicle.

[0061] Referring to Figure 1 , Figure 1A flowchart of an abnormal state detection method provided in an embodiment of the present application is shown in FIG. 1. The method comprises the following steps:

[0062] S101, determining an abnormal mode to be detected.

[0063] In this embodiment, the abnormal mode refers to a type of abnormality of the vehicle to be detected, such as a power battery abnormal mode, a motor abnormal mode, etc. Different types of abnormal modes require different data to be detected, so the present application first determines which type of abnormal mode to be detected before performing the subsequent detection operation based on the abnormal mode.

[0064] S102, obtaining data to be detected corresponding to the abnormal mode from the vehicle to be detected; the data to be detected comprises first data to be detected and second data to be detected.

[0065] In this embodiment, the vehicle to be detected is a vehicle whose abnormal state needs to be detected, so the present application needs to obtain data to be detected corresponding to the abnormal mode from the vehicle to be detected when performing abnormal detection. The data to be detected is various data collected in real time during the operation of the vehicle. The present application can obtain the data to be detected through various sensors deployed on the vehicle, such as temperature sensors, pressure sensors, current sensors, etc., which are not specifically limited herein. Moreover, the type of data to be detected needs to be determined according to the abnormal mode. For example, if the abnormal mode is a power battery abnormal mode, the data to be detected includes temperature signals, voltage signals, current signals, and SOC (State of Charge) values. If the abnormal mode is a motor bearing abnormal mode, the data to be detected includes current signals, vibration signals, and voltage signals.

[0066] Further, after obtaining the data to be detected, the present application needs to divide the data to be detected into two categories according to the function. The first category of data needs to be used not only as input data of the detection network but also compared with the predicted data to determine the target loss value. The present application refers to the first category of data as first data to be detected. The second category of data only needs to be used as input data of the detection network. The present application refers to the second category of data as second data to be detected.

[0067] It should be noted that the first to be detected data and the second to be detected data in the embodiment are two types of data classified according to types, and the first to be detected data and the second to be detected data can each include one or more types of data. For example, if the abnormal mode is a power battery abnormal mode, the acquired to-be-detected data includes temperature signals, voltage signals, current signals, and SOC values. Therefore, the temperature signals and the voltage signals are taken as the first to be detected data, and the current signals and the SOC values are taken as the second to be detected data. If the abnormal mode is a motor bearing abnormal mode, the acquired to-be-detected data includes current signals, vibration signals, and voltage signals. The current signals can be taken as the first to be detected data, and the vibration signals and the voltage signals can be taken as the second to be detected data.

[0068] S103, inputting the first to be detected data and the second to be detected data into a pre-trained detection network to obtain prediction data;

[0069] It should be noted that the detection network is a pre-trained detection network, which is a few-shot knowledge flexible transfer network. The network can effectively detect abnormal states according to the hidden relationship between the captured signals. In the test stage, first, the training data needs to be acquired, and the type of the training data is consistent with the type of the to-be-detected data. Then, the detection network is trained by using the training data to obtain the detection network of the application. The first to be detected data and the second to be detected data are input into the pre-trained detection network, and the detection network can output corresponding prediction data. The prediction data is the prediction data when the vehicle does not have an abnormal condition.

[0070] For example, if the abnormal mode is a power battery abnormal mode, the first to be detected data is temperature signals and voltage signals, and the second to be detected data is current signals and SOC values. Then, after the temperature signals, the voltage signals, the current signals, and the SOC values are input into the detection network, the prediction data output by the detection network is prediction temperature signals and prediction voltage signals. If the abnormal mode is a motor bearing abnormal mode, the first to be detected data is current signals, and the second to be detected data is vibration signals and voltage signals. Then, after the current signals, the vibration signals, and the voltage signals are input into the detection network, the prediction data output by the detection network is prediction current signals.

[0071] S104, calculating a target loss value according to the prediction data and the first to be detected data;

[0072] It should be noted that, since in the training process, the number of normal samples in the training data is large, and the number of abnormal samples is small, after the detection network training is completed, if the vehicle is currently in a normal working state, the prediction data output by the detection network should be close to the first to-be-detected data input, and the loss value is small, if the vehicle currently has an abnormal state, the prediction data output by the detection network should be far away from the first to-be-detected data input, and the loss value is large. Based on this, the target loss value is calculated by the prediction data and the first to-be-detected data, so as to determine whether the vehicle has an abnormal state according to the target loss value.

[0073] In the embodiment, the loss value can be calculated in different ways when calculating the target loss value, such as: calculating the target loss value between the prediction data and the first to-be-detected data by mean squared error (MSE), and also calculating the target loss value between the prediction data and the first to-be-detected data by mean absolute error (MAE), which is not specifically limited here.

[0074] S105, detecting an abnormal state by using the target loss value and an abnormal threshold.

[0075] In the embodiment, the abnormal threshold is a threshold set in combination with actual data and engineer experience, and by comparing the target loss value with the abnormal threshold, it can be determined whether the vehicle has an abnormal state. For example: if the target loss value is less than the abnormal threshold, it is determined that the vehicle does not have an abnormal state, and if the target loss value is not less than the abnormal threshold, it is determined that the vehicle has an abnormal state.

[0076] As can be seen from the above, when detecting the abnormal state of the vehicle, the abnormal threshold is set as an evaluation standard, and the abnormal state is detected by comparing the abnormal threshold with the target loss value. If the target loss value is too large, it means that the prediction data is far away from the first to-be-detected data input, and the vehicle has an abnormal state, otherwise it is in a normal state. In this way, the application can effectively identify potential abnormal behavior in the case of scarce abnormal state samples, so as to timely warn, reduce the risk of accidents, and protect the safety of the vehicle and passengers.

[0077] Referring to Figure 2 , Figure 2 Another abnormal state detection method flowchart provided by the embodiment of the application is provided, which specifically includes the following steps:

[0078] S201, determining an abnormal mode to be detected;

[0079] S202, obtaining to-be-detected data corresponding to an abnormal mode from a to-be-detected vehicle; the to-be-detected data includes first to-be-detected data and second to-be-detected data;

[0080] S203, inputting the first to-be-detected data and the second to-be-detected data into an encoder of a detection network to obtain a high-dimensional space feature vector;

[0081] S204, inputting the high-dimensional space feature vector and the second to-be-detected data into a decoder of the detection network to obtain predicted data;

[0082] S205, calculating a target loss value according to the predicted data and the first to-be-detected data;

[0083] S206, detecting an abnormal state by using the target loss value and an abnormal threshold.

[0084] It should be noted that the detection network in the embodiment is a few-sample knowledge flexible migration network, which mainly includes an encoder, a decoder and a multi-layer perception machine, and the encoder and the decoder each include three layers of GRU (gated recurrent unit). Here, the process of outputting predicted data by the detection network is described by taking the abnormal mode of a power battery as an example.

[0085] Referring to Figure 3 , Figure 3 The detection method flowchart based on the few-sample knowledge flexible migration network provided by the embodiment of the application. As can be seen from the flowchart, in power battery abnormality detection, the data collected by a sensor mainly includes temperature signals, voltage signals, current signals and SOC values, and there is a strong correlation between these signals. The few-sample flexible knowledge migration network is used to capture the hidden relationship between the signals, so as to effectively detect an abnormal state. The collected data is divided into first to-be-detected data x0: temperature signals and voltage signals, and second to-be-detected data x1: current signals and SOC values. The first to-be-detected data x0 and the second to-be-detected data x1 are input into an encoder of a detection network to obtain a high-dimensional space feature vector, and then the high-dimensional space feature vector and the second to-be-detected data x1 are input into a decoder of the detection network to obtain predicted data x ′ 0, which includes predicted temperature signals and predicted voltage signals. If the battery is in a normal state, the output x ′ 0 can basically fit x0, and the loss function is small. If the battery is in an abnormal state, the output x ′ 0 is greatly different from x0, and the loss function is large. Therefore, the loss value is used as an evaluation index, and a reasonable abnormal threshold is set, so that whether the power battery is in an abnormal state can be effectively determined.

[0086] And, in order to complete the training of the detection network, the CPU (Central Processing Unit) of the computer used in the experiment is Intel Core i7-9750H, the memory is 16G, the graphics card is four NVIDIA 3080 with 8GB of video memory, the operating system is ubantu, and the framework is pytorch platform of python3.6. The batch size is set to 512, the iteration number is set to 500, the learning rate is set to 0.01, the loss function is defined as cross entropy, and the Adam algorithm is used for training.

[0087] It should be noted that in the prior art, when detecting an anomaly, a special model needs to be trained for a specific anomaly state detection task. When processing other types of anomaly state detection tasks, the structure of the model needs to be modified, so the detection model in the prior art is difficult to be directly applied to other types of anomaly state detection tasks, especially in the field of new energy vehicles, different brands and models of vehicles may have different abnormal modes. In the present application, the few-sample knowledge flexible migration network can be applied to different anomaly state detection tasks of different brands and different models of vehicles, without the need to modify the network structure. After retraining the network using training data of different anomaly state detection tasks, the trained network can perform different types of anomaly state detection tasks.

[0088] Based on the above embodiment, in the present embodiment, calculating a target loss value according to the prediction data and the first to-be-detected data comprises: calculating a reconstruction loss value using a reconstruction loss determination rule, the prediction data and the first to-be-detected data.

[0089] The reconstruction loss determination rule is:

[0090] l recom (θ,ζ,x0,x1)=MSE(Decoder ζ (z,x1),x0);

[0091] Wherein, l recom is a reconstruction loss value, θ is an encoder coefficient, ζ is a decoder coefficient, x0 is the first to-be-detected data, x1 is the second to-be-detected data, z is a high-dimensional space feature vector, z=Encoder θ (x0,x1), Encoder is an encoder, Decoder is a decoder, MSE is a mean square error loss function, and Decoder ζ (z,x1) is the prediction data.

[0092] In this embodiment, the reconstruction loss value can be calculated using the reconstruction loss determination rule, and this reconstruction loss value is used as the target loss value and compared with the anomaly threshold to detect abnormal states. However, with a small number of abnormal samples, the trained detection model is prone to overfitting the training data, resulting in poor performance on the test set or in real-world applications. To prevent overfitting and considering the strong correlation between driving range and battery signal, this application introduces regularization loss and mileage loss into the overall loss, thereby improving the model's detection accuracy. Therefore, after calculating the reconstruction loss value using the reconstruction loss determination rule, predicted data, and the first data to be detected, this application further includes:

[0093] 1. Calculate the regularization loss value using the regularization loss determination rule and the aforementioned high-dimensional feature vector; the regularization loss determination rule is as follows:

[0094]

[0095] Among them, l reg is the regularization loss value, μ represents the mean, σ represents the standard deviation, tr represents the trace, and log represents the logarithm.

[0096] 2. Input the high-dimensional feature vector into the decoder of the detection network, and input the output into the multilayer perceptron of the detection network to obtain the predicted mileage value;

[0097] The mileage loss value is calculated using the mileage loss determination rule, the predicted mileage value, and the actual mileage value; the mileage loss determination rule is as follows:

[0098] l mile (θ,x0,x1)=MSE(MLP ξ (Decoder ζ (z,θ)),mile);

[0099] Among them, l mile ξ represents the mileage loss value, MLP is the multilayer perceptron, ξ is the multilayer perceptron parameter, and mile is the actual mileage value.

[0100] pass Figure 3 As can be seen, in this embodiment, the regularization loss value is calculated based on the high-dimensional space feature vector and the regularization loss determination rule. The high-dimensional space feature vector is input into the decoder of the detection network, and the output result is input into the multilayer perceptron of the detection network to obtain the predicted mileage value. The overall loss can then be calculated based on the reconstruction loss value, the regularization loss value, and the mileage loss value. This overall loss is the target loss value in this application. The overall loss determination rule is as follows:

[0101] l(θ,ζ,ξ)=l recom (θ,ζ,x0,x1)+l reg(θ, x0, x1) + l mile (θ, x0, x1) + l

[0102] wherein, l(θ, ζ, ξ) is the total loss value.

[0103] As can be seen from the above, in the calculation of the target loss value, the reconstruction loss value is calculated by the reconstruction loss determination rule, the regularization loss value is calculated by the regularization loss determination rule, and the mileage loss value is calculated by the mileage loss determination rule. The abnormal state is detected by the total loss of the reconstruction loss value, the regularization loss value and the mileage loss value, the overfitting phenomenon is effectively prevented, and the detection accuracy of the abnormal state is improved.

[0104] Referring to Figure 4 , Figure 4 A flowchart of an abnormal state detection method provided by an embodiment of the present application is shown in the figure. The method specifically includes the following steps:

[0105] S301, determine an abnormal mode to be detected;

[0106] S302, obtain, from a vehicle to be detected, to-be-detected data corresponding to the abnormal mode; the to-be-detected data includes first to-be-detected data and second to-be-detected data;

[0107] S303, input the first to-be-detected data and the second to-be-detected data into an encoder of a detection network to obtain a high-dimensional space feature vector;

[0108] S304, input the high-dimensional space feature vector and the second to-be-detected data into a decoder of the detection network to obtain predicted data;

[0109] S305, calculate a target loss value according to the predicted data and the first to-be-detected data;

[0110] S306, determine the target loss value of each signal sample in the to-be-detected data;

[0111] S307, select a target loss value to be processed from the target loss values of all signal samples according to an abnormal sample proportion coefficient;

[0112] S308, calculate an average loss value of all target loss values to be processed;

[0113] S309, determine whether the average loss value is greater than an abnormal threshold value;

[0114] If yes, execute S310; if no, continue to execute S302;

[0115] S310, determine that an abnormal state corresponding to the abnormal mode is detected;

[0116] S311, display the abnormal state in the digital twin model and generate a warning signal;

[0117] S312, send the warning signal to a predetermined terminal device.

[0118] In the embodiment, in order to effectively evaluate the abnormal state of the power battery, the application sets two parameters to determine the abnormal state in combination with actual data and engineer experience, the parameters are: an abnormal sample proportion coefficient p and an abnormal threshold τ. The abnormal sample proportion coefficient p is used to select the first n target loss values to be processed with larger target loss values from m sample signals, that is: n = p x m, m is the total number of signal samples. Therefore, after determining the target loss value of each signal sample in the to-be-detected data, the number n of target loss values to be processed is calculated according to the abnormal sample proportion coefficient p and the total number m of signal samples, the target loss values of all signal samples are arranged in descending order, the first n target loss values to be processed with larger target loss values are selected, the average loss value of all target loss values to be processed is calculated, and whether there is an abnormal state is determined according to the average loss value and the abnormal threshold τ.

[0119] The specific process is as follows:

[0120] L vehicle = (l1(θ, ζ, ξ),..., lm(θ, ζ, ξ)); m (θ, ζ, ξ));

[0121]

[0122] wherein, L vehicle represents the target loss values of all signal samples, l1(θ, ζ, ξ) is the target loss value of the first signal sample, l m (θ, ζ, ξ) is the target loss value of the mth signal sample, and VehicleError is the average loss value of the n larger target loss values selected. Whether VehicleError is greater than the abnormal threshold τ is determined, if greater than the abnormal threshold τ, it is determined that an abnormal state occurs, otherwise it is determined that no abnormal state occurs.

[0123] Further, the application can also construct a digital twin model, and combine the digital twin model with the detection network to detect the abnormal state. The construction method of the digital twin model comprises: determining an abnormal mode of the digital twin model, according to specific abnormal working conditions and abnormal performances, deeply understanding the operation mechanism of the new energy vehicle, and determining key parameters and abnormal detection standards of the model that need to be monitored. The key parameters are the to-be-detected data in the abnormal mode, and the abnormal detection standards include an abnormal sample proportion coefficient and an abnormal threshold. Various key data in the vehicle operation process are collected in real time through various sensors deployed on the vehicle, and the digital twin model is constructed in combination with the detection network.

[0124] Further, the digital twin model can reflect the running state of the vehicle in real time, and once the detection network detects an abnormal state, the abnormal state can be displayed in the digital twin model, and the digital twin model will immediately issue a warning signal, wherein the warning signal comprises the abnormal state, the abnormal reason and the solving measures. The warning signal can be sent to a predetermined terminal device, such as a mobile phone of the driver or the maintenance personnel, so as to remind the driver or the maintenance personnel to pay attention to the abnormal state and take reasonable measures to reduce or eliminate the influence of the abnormal state.

[0125] As can be seen from the above, in order to effectively evaluate the abnormal state of the power battery, the application combines actual data and engineer experience, and migrates expert knowledge into the detection network through the two parameters of the abnormal sample proportion coefficient p and the abnormal threshold τ, realizes the detection process combining expert knowledge and data analysis, and thus effectively solves the problem that the power battery is difficult to detect due to few abnormal samples in actual working conditions. Further, the digital twin model constructed by the application can reflect the running state of the vehicle in real time, and once an abnormal state is detected, the digital twin model will immediately issue a warning signal to remind the driver or the maintenance personnel to pay attention and take reasonable measures to reduce or eliminate the influence of the abnormal state. Such a warning mechanism helps manufacturers and maintenance agencies to make targeted maintenance plans in advance, optimizes maintenance strategies, and reduces downtime and maintenance costs caused by sudden failures. As a new emerging transportation tool, the user experience of the new energy vehicle is one of the key factors for its market promotion, and the application can timely find and solve small problems in the vehicle operation, avoid accumulation of small problems into big problems, and thus improve the satisfaction and trust of users.

[0126] Reference Figure 5 , Figure 5 An abnormal state detection device structure schematic diagram provided by an embodiment of the application, the device specifically comprises:

[0127] A determination module 11 is configured to determine a to-be-detected abnormal mode;

[0128] The acquisition module 12 is configured to acquire, from a vehicle to be detected, to-be-detected data corresponding to the abnormal pattern; the to-be-detected data includes first to-be-detected data and second to-be-detected data.

[0129] The input module 13 is configured to input the first to-be-detected data and the second to-be-detected data into a pre-trained detection network to obtain prediction data.

[0130] The calculation module 14 is configured to calculate a target loss value according to the prediction data and the first to-be-detected data.

[0131] The detection module 15 is configured to detect an abnormal state by using the target loss value and an abnormal threshold.

[0132] As an optional embodiment, the input module includes:

[0133] The first input unit is configured to input the first to-be-detected data and the second to-be-detected data into an encoder of the detection network to obtain a high-dimensional space feature vector.

[0134] The second input unit is configured to input the high-dimensional space feature vector and the second to-be-detected data into a decoder of the detection network to obtain the prediction data.

[0135] As an optional embodiment, the calculation module includes:

[0136] The first calculation unit is configured to calculate a reconstruction loss value by using a reconstruction loss determination rule, the prediction data and the first to-be-detected data; the reconstruction loss determination rule is:

[0137] l recom (θ,ζ,x0,x1)=MSE(Decoder ζ (z,x1),x0);

[0138] wherein, l recom is the reconstruction loss value, θ is an encoder coefficient, ζ is a decoder coefficient, x0 is the first to-be-detected data, x1 is the second to-be-detected data, z is the high-dimensional space feature vector, z=Encoder θ (x0,x1), Encoder is the encoder, Decoder is the decoder, MSE is a mean square error loss function, Decoder ζ (z,x1) is the prediction data.

[0139] As an optional embodiment, the calculation module further includes:

[0140] The second calculation unit is configured to calculate a regularization loss value by using a regularization loss determination rule and the high-dimensional space feature vector; the regularization loss determination rule is:

[0141]

[0142] wherein, l reg is a regularization loss value, μ represents a mean value, σ represents a mean square deviation, tr represents a trace number, and log represents a logarithm.

[0143] As an optional embodiment, the computing module further comprises:

[0144] a third input unit configured to input the high-dimensional space feature vector into a decoder of the detection network, and input an output result into a multi-layer perceptron of the detection network to obtain a predicted mileage value;

[0145] a third computing unit configured to calculate a mileage loss value by using a mileage loss determination rule, the predicted mileage value, and an actual mileage value; the mileage loss determination rule is:

[0146] l mile (θ, x0, x1) = MSE(MLP ξ (Decoder ζ (z, θ)), mile);

[0147] wherein, l mile is a mileage loss value, MLP is a multi-layer perceptron, ξ is a multi-layer perceptron parameter, and mile is an actual mileage value.

[0148] As an optional embodiment, the detection module comprises:

[0149] a determining unit configured to determine a target loss value of each signal sample in the to-be-detected data;

[0150] a selecting unit configured to select a to-be-processed target loss value from the target loss values of all signal samples according to an abnormal sample proportion coefficient;

[0151] a fourth computing unit configured to calculate an average loss value of all to-be-processed target loss values;

[0152] a judging unit configured to judge whether the average loss value is greater than an abnormal threshold value; if yes, a determining unit is triggered;

[0153] the determining unit is configured to determine that an abnormal state corresponding to the abnormal pattern is detected.

[0154] As an optional embodiment, the device further comprises:

[0155] a display module configured to display the abnormal state in a digital twin model, and generate an early warning signal;

[0156] a sending module configured to send the early warning signal to a predetermined terminal device.

[0157] As to the apparatus in the above embodiments, the specific manners in which the respective modules perform operations have been described in detail in the embodiments of the method, and thus will not be elaborated here.

[0158] Referring to Figure 6 , Figure 6 An electronic device structure schematic diagram is provided for the embodiments of the present application, and the electronic device specifically includes:

[0159] The processor 21, the memory 22, and the computer program stored in the memory 22 and executable on the processor 21, the processor 21 executes the steps of the abnormal state detection method described in any method embodiment above through the computer program.

[0160] The processor 21 can include one or more processing cores, such as a 4-core processor, an 8-core processor, etc. The processor 21 can be implemented in the form of at least one of a DSP (Digital Signal Processing), an FPGA (Field-Programmable Gate Array), and a PLA (Programmable Logic Array). The processor 21 can also include a main processor and a coprocessor. The main processor is a processor for processing data in a wake-up state, also known as a CPU (Central Processing Unit). The coprocessor is a low-power processor for processing data in a standby state. In some embodiments, the processor 21 can be integrated with a GPU (Graphics Processing Unit) that is responsible for rendering and drawing the content required to be displayed on the display screen. In some embodiments, the processor 21 can also include an AI (Artificial Intelligence) processor for processing machine learning-related computing operations.

[0161] The memory 22 can include one or more computer-readable storage media. The computer-readable storage media can be non-transitory. The memory 22 can also include high-speed random access memory and can include non-volatile memory, such as one or more magnetic disk storage devices, flash memory devices. In this embodiment, the memory 22 is used to store at least the following computer program 221, wherein the computer program is loaded and executed by the processor 21, and can implement the related steps in the abnormal state detection method disclosed in any of the preceding embodiments. In addition, the resources stored in the memory 22 can also include an operating system 222 and data 223, etc., and the storage mode can be temporary storage or permanent storage. The operating system 222 can include Windows, Unix, Linux, etc.

[0162] In some embodiments, the electronic device can further include a display screen 23, an input / output interface 24, a communication interface 25, a sensor 26, a power supply 27, and a communication bus 28.

[0163] Of course, Figure 6 The structure of the electronic device shown does not constitute a limitation on the electronic device in the embodiments of the present application. In actual applications, the electronic device can include more or fewer components than those shown, or some components can be combined. Figure 6 The structure of the electronic device shown does not constitute a limitation on the electronic device in the embodiments of the present application. In actual applications, the electronic device can include more or fewer components than those shown, or some components can be combined.

[0164] In another exemplary embodiment, a computer storage medium is also provided, and the program instructions are executed by the processor to implement the steps of the abnormal state detection method described in any of the preceding method embodiments. The storage medium can include a U disk, a mobile hard disk, a read-only memory (ROM), a random access memory (RAM), a magnetic disk or an optical disk, and various media that can store program codes.

[0165] Optionally, the specific examples in the embodiments can refer to the examples described in the above-described embodiments, and the embodiments will not be described here.

[0166] It is to be understood that the terminology used herein is for the purpose of describing particular example embodiments only and is not intended to be limiting. As used herein, the singular forms "a", "an" and "the" are intended to include the plural forms as well, unless the context clearly indicates otherwise. The terms "comprises", "comprising", "includes", "including" and "has" are inclusive and therefore specify the presence of stated features, steps, operations, elements, and / or components, but do not preclude the presence or addition of one or more other features, steps, operations, elements, components, and / or groups thereof. The method steps, processes, and operations described herein are not to be construed as necessarily requiring their performance in the particular order

[0167] The above description is merely that of the specific embodiments of the application and as such is not to be taken in a limiting sense. Various modifications and alterations of the embodiments described herein will become apparent to those skilled in the art from the foregoing description, which does not limit the generality presented. It is the intention that all such modifications and alterations be considered equaliy by the spirit and scope of this application. It is therefore intended to cover in the appended claims all such changes and alterations that come within the scope of this application.

Claims

1. An abnormal state detection method characterized by comprising: The method comprises the following steps: determining an abnormal pattern to be detected; obtaining, from a vehicle to be detected, to-be-detected data corresponding to the abnormal pattern; the to-be-detected data comprises two types of data divided according to functions: first to-be-detected data and second to-be-detected data, and each of the first to-be-detected data and the second to-be-detected data comprises one or more types of data; inputting the first to-be-detected data and the second to-be-detected data into a pre-trained detection network to obtain prediction data; wherein the prediction data is prediction data under a normal state of the vehicle; calculating a target loss value according to the prediction data and the first to-be-detected data; the calculation method of the target loss value is as follows: calculating a reconstruction loss value by a reconstruction loss determination rule, calculating a regularization loss value by a regularization loss determination rule, calculating a mileage loss value by a mileage loss determination rule, and calculating the sum of the reconstruction loss value, the regularization loss value and the mileage loss value as the target loss value; detecting an abnormal state by using the target loss value and an abnormal threshold.

2. The abnormal state detection method according to claim 1, characterized by, The step of inputting the first to-be-detected data and the second to-be-detected data into a pre-trained detection network to obtain prediction data comprises the following steps: inputting the first to-be-detected data and the second to-be-detected data into an encoder of the detection network to obtain a high-dimensional space feature vector; inputting the high-dimensional space feature vector and the second to-be-detected data into a decoder of the detection network to obtain the prediction data.

3. The abnormal state detection method according to claim 2, characterized by, The step of calculating a target loss value according to the prediction data and the first to-be-detected data comprises the following steps: calculating a reconstruction loss value by using a reconstruction loss determination rule, the prediction data and the first to-be-detected data; the reconstruction loss determination rule is as follows: ； wherein, is a reconstruction loss value, is an encoder coefficient, is a decoder coefficient, is first to-be-detected data, is second to-be-detected data, is a high-dimensional space feature vector, , is an encoder, is a decoder, is a mean square error loss function, is predicted data.

4. The abnormal state detection method according to claim 3, characterized by, if the abnormal pattern is a power battery abnormal pattern, after the step of calculating a reconstruction loss value by using a reconstruction loss determination rule, the prediction data and the first to-be-detected data, the method further comprises the following steps: calculating a regularization loss value by using a regularization loss determination rule and the high-dimensional space feature vector; the regularization loss determination rule is as follows: ； wherein, is a regularization loss value, denotes a mean, denotes a mean square error, denotes a trace, denotes a logarithm.

5. The abnormal state detection method according to claim 3, characterized by, if the abnormal pattern is a power battery abnormal pattern, after the step of calculating a reconstruction loss value by using a reconstruction loss determination rule, the prediction data and the first to-be-detected data, the method further comprises the following steps: inputting the high-dimensional space feature vector into a decoder of the detection network, and inputting an output result into a multi-layer perceptron of the detection network to obtain a predicted mileage value; calculating a mileage loss value by using a mileage loss determination rule, the predicted mileage value and an actual mileage value; the mileage loss determination rule is as follows: ； wherein, is a mileage loss value, is a multilayer perceptron, is a multilayer perceptron parameter, is an actual mileage value.

6. The abnormal state detection method according to any one of claims 1 to 5, characterized by, the step of detecting an abnormal state by using the target loss value and an abnormal threshold comprises the following steps: determining a target loss value of each signal sample in the to-be-detected data; selecting to-be-processed target loss values from the target loss values of all signal samples according to an abnormal sample proportion coefficient; calculating an average loss value of all to-be-processed target loss values; judging whether the average loss value is greater than an abnormal threshold; if yes, it is determined that an abnormal state corresponding to the abnormal pattern is detected.

7. The abnormal state detection method according to claim 6, characterized by, after the step of determining that an abnormal state corresponding to the abnormal pattern is detected, the method further comprises the following steps: displaying the abnormal state in a digital twin model, and generating a warning signal; The early warning signal is sent to a predetermined terminal device.

8. An abnormal state detection device characterized by comprising: The method comprises the steps of: determining an abnormal pattern to be detected; obtaining, from a vehicle to be detected, to-be-detected data corresponding to the abnormal pattern; The to-be-detected data comprises two types of data divided according to functions: first to-be-detected data and second to-be-detected data, and each of the first to-be-detected data and the second to-be-detected data comprises one or more types of data; inputting the first to-be-detected data and the second to-be-detected data into a pre-trained detection network to obtain prediction data; wherein the prediction data is prediction data under a state in which the vehicle does not have an abnormality; calculating a target loss value according to the prediction data and the first to-be-detected data; the calculation method of the target loss value is as follows: calculating a reconstruction loss value through a reconstruction loss determination rule, calculating a regularization loss value through a regularization loss determination rule, calculating a mileage loss value through a mileage loss determination rule, and calculating the sum of the reconstruction loss value, the regularization loss value, and the mileage loss value as the target loss value; detecting an abnormal state by using the target loss value and an abnormal threshold.

9. An electronic device, comprising: The method comprises the steps of: a processor, a memory, and a computer program stored on the memory and executable on the processor, wherein the processor executes the steps of the abnormal state detection method according to any one of claims 1 to 7 by using the computer program.

10. A computer storage medium, characterized in that, The computer storage medium stores computer executable instructions for executing the steps of the abnormal state detection method according to any one of claims 1 to 7.

Images