CN115828161A

CN115828161A - Automobile fault type prediction method and device based on recurrent neural network

Info

Publication number: CN115828161A
Application number: CN202211691920.3A
Authority: CN
Inventors: 徐俊涛; 万龙; 张宇
Original assignee: South Sagittarius Integration Co Ltd
Current assignee: South Sagittarius Integration Co Ltd
Priority date: 2022-12-28
Filing date: 2022-12-28
Publication date: 2023-03-21

Abstract

The invention provides a method and a device for predicting automobile fault types based on a recurrent neural network, wherein the method comprises the following steps: acquiring vehicle-mounted data reported by a vehicle-mounted terminal in an automobile; preprocessing the vehicle-mounted data to obtain standardized vehicle-mounted data; coding the standardized vehicle-mounted data to obtain vehicle-mounted characteristics; inputting the vehicle-mounted characteristics into a pre-trained recurrent neural network to obtain the fault type output by the recurrent neural network, and completing the prediction of the vehicle fault type. By applying the embodiment of the invention, the early warning of the vehicle fault type is realized, and the fault type prediction accuracy is improved.

Description

Automobile fault type prediction method and device based on recurrent neural network

Technical Field

The invention relates to the technical field of intelligent vehicle-mounted, in particular to a method and a device for predicting a type of an automobile fault based on a recurrent neural network.

Background

The automobile is a complex mechanical system, which is composed of thousands of different parts, the structure is complex, the working conditions are various, the automobile is influenced by various environments and driving conditions in the long-term use process, the parts of the automobile can change or the performance parameters deteriorate according to different laws and different strengths, and various fault alarms appear along with the increase of the use time. The vehicle fault alarm may be accompanied by the fault of vehicle parts, which affects the normal driving and use of the vehicle, and further may cause accidents endangering the safety of drivers and passengers.

Therefore, the method can accurately predict the vehicle fault types in advance, and is an important means for reducing personal and property injuries of drivers and passengers. In order to realize rapid advance prediction of fault types and reduce potential accident occurrence and secondary loss thereof, an automobile fault type prediction method is urgently needed.

Disclosure of Invention

The invention aims to overcome the defects of the prior art and provides a method and a device for predicting the type of an automobile fault based on a recurrent neural network so as to realize rapid advance prediction of the type of the automobile fault.

The invention is realized by the following steps:

in a first aspect, a method for predicting a type of vehicle failure based on a recurrent neural network includes:

acquiring vehicle-mounted data reported by a vehicle-mounted terminal in an automobile; preprocessing the vehicle-mounted data to obtain standardized vehicle-mounted data;

coding the standardized vehicle-mounted data to obtain vehicle-mounted characteristics;

inputting the vehicle-mounted characteristics into a pre-trained recurrent neural network to obtain the fault type output by the recurrent neural network, and completing the prediction of the vehicle fault type.

Optionally, the preprocessing the vehicle-mounted data to obtain standardized vehicle-mounted data includes:

deleting invalid data and repeated data in the vehicle-mounted data to obtain valid vehicle-mounted data;

and carrying out standardization processing on the effective vehicle-mounted data to obtain standardized vehicle-mounted data.

Optionally, the normalizing the valid vehicle-mounted data to obtain normalized vehicle-mounted data includes:

and based on normalization processing, carrying out standardization processing on the effective vehicle-mounted data to obtain standardized vehicle-mounted data.

Optionally, the vehicle-mounted data includes: the system comprises a vehicle frame number, data reporting time, speed, total meter mileage, a chip system, total battery voltage, total battery current, lowest value of single battery voltage, highest temperature value of a probe, lowest temperature value of the probe and highest alarm level.

Optionally, the recurrent neural network is obtained by pre-training in the following manner:

setting the number of times of an observation window of the initial recurrent neural network as a first preset value; setting the predicted lag times as a second preset value;

inputting a two-dimensional array consisting of first preset value continuous sample records into the initial cyclic neural network to obtain a predicted sample label output by the initial cyclic neural network after reaching the second preset value;

constructing a sample set based on the first preset value of continuous sample records and the sample label;

and training the initial recurrent neural network by using the sample set to obtain the trained recurrent neural network.

In a second aspect, the present invention provides a vehicle fault type prediction apparatus based on a recurrent neural network, the apparatus comprising:

the acquisition unit is used for acquiring vehicle-mounted data reported by a vehicle-mounted terminal in an automobile; preprocessing the vehicle-mounted data to obtain standardized vehicle-mounted data;

the coding unit is used for coding the standardized vehicle-mounted data to obtain vehicle-mounted characteristics;

and the prediction unit is used for inputting the vehicle-mounted characteristics into a pre-trained recurrent neural network to obtain the fault type output by the recurrent neural network, and completing the prediction of the automobile fault type.

Optionally, the obtaining unit preprocesses the vehicle-mounted data to obtain standardized vehicle-mounted data, and specifically includes:

Optionally, the obtaining unit performs standardization processing on the effective vehicle-mounted data to obtain standardized vehicle-mounted data, which specifically includes:

In a third aspect, the present invention provides an electronic device comprising:

a memory storing executable program code;

a processor coupled with the memory;

the processor calls the executable program code stored in the memory for executing any one of the above-mentioned vehicle fault type prediction methods based on the recurrent neural network.

In a fourth aspect, the present invention provides a computer-readable storage medium storing a computer program, wherein the computer program causes a computer to execute any one of the above-mentioned recurrent neural network-based automobile fault type prediction methods.

The invention has the following beneficial effects: by applying the embodiment of the invention, based on the vehicle-mounted data uploaded by the vehicle-mounted terminal, the vehicle-mounted data is preprocessed to obtain standardized vehicle-mounted data; coding the standardized vehicle-mounted data to obtain vehicle-mounted characteristics; the vehicle fault alarm type is predicted by adopting a cyclic neural network structure, so that the vehicle fault alarm type is predicted in advance, and the accuracy of a prediction result is improved.

Drawings

In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings used in the embodiments or the prior art descriptions will be briefly described below, it is obvious that the drawings in the following description are only some embodiments of the present invention, and other drawings can be obtained by those skilled in the art without creative efforts.

Fig. 1 is a schematic flowchart of a method for predicting a type of vehicle fault based on a recurrent neural network according to an embodiment of the present invention;

fig. 2 is a diagram of a GRU model structure provided in an embodiment of the present invention;

FIG. 3 is a diagram of an LSTM model architecture provided by an embodiment of the present invention;

fig. 4 is a schematic structural diagram of a vehicle fault type prediction apparatus based on a recurrent neural network according to an embodiment of the present invention;

fig. 5 is a schematic structural diagram of an electronic device according to an embodiment of the present invention.

Detailed Description

The technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the drawings in the embodiments of the present invention, and it is obvious that the described embodiments are only a part of the embodiments of the present invention, and not all of the embodiments. All other embodiments, which can be derived by a person skilled in the art from the embodiments given herein without making any creative effort, shall fall within the protection scope of the present invention.

Aiming at the technical field of intelligent vehicle-mounted technology, in order to realize the advance prediction of vehicle faults and further reduce the personal and property losses caused by potential accidents, the inventor of the present application has conducted research, and finds that in the prior art, the vehicle fault prediction cannot take the incidence relation between historical vehicle-mounted data and lagging fault levels into consideration, so that the accuracy of fault type prediction is affected.

The automobile fault type prediction method based on the recurrent neural network can be applied to any scene needing fault prediction of vehicle-mounted data. For example, the method can be applied to new energy vehicle fault alarm type prediction or common vehicle fault alarm type prediction and the like.

The inventor finds that the incidence relation between the historical message and the lagging fault grade can be captured through the recurrent neural network to predict the vehicle-mounted fault type information at the future moment, so that the recurrent neural network is applied to vehicle-mounted fault prediction, the vehicle fault grade can be rapidly, early and accurately predicted, the occurrence of vehicle faults in advance can be effectively reduced, and the potential fault secondary loss is reduced.

Through continuous research, the inventor finally provides an automobile fault type prediction method based on a recurrent neural network, and the basic concept of the scheme is as follows: acquiring vehicle-mounted data reported by a vehicle-mounted terminal in an automobile; preprocessing the vehicle-mounted data to obtain standardized vehicle-mounted data; coding the standardized vehicle-mounted data to obtain vehicle-mounted characteristics; inputting the vehicle-mounted characteristics into a pre-trained recurrent neural network to obtain the fault type output by the recurrent neural network, and completing the prediction of the vehicle fault type.

By applying the embodiment of the invention, the vehicle fault alarm type is predicted based on the vehicle data uploaded by the vehicle terminal in real time and by adopting the cyclic neural network structure, so that the accuracy of the prediction result is improved, and the advance prediction of the vehicle fault alarm type is realized.

The automobile fault type prediction method based on the recurrent neural network can be applied to electronic equipment with data processing capacity, the electronic equipment can be a server on a network side and can also be a terminal used by a user side, such as a PC (personal computer), a notebook computer, a smart phone and the like, and the server on the network side or the terminal used by the user side can process vehicle-mounted data according to the fault type prediction method provided by the application. In addition, the functional software for implementing the fault type prediction method provided by the embodiment of the invention may be special software with information processing capability, or may be a plug-in the software with information processing capability. The following embodiments are provided to describe the failure type prediction method provided in the present application.

Referring to fig. 1, a schematic flow chart of a method for predicting a type of vehicle fault based on a recurrent neural network according to an embodiment of the present application is shown, where the method may include:

s101, acquiring vehicle-mounted data reported by a vehicle-mounted terminal in an automobile; preprocessing the vehicle-mounted data to obtain standardized vehicle-mounted data;

interface data reported by one or more new energy vehicles T-box (vehicle-mounted terminal) in a period of time can be acquired, for example, the period of time can be limited to 2021-11-01 to 2022-03-31, and vehicle-mounted data reported by 100 new energy vehicles numbered 1-100 in the period of time is acquired. The original vehicle-mounted data is analyzed and stored, and the following effective field information such as one or more of the vehicle frame number, the data reporting time, the driving speed, the total meter mileage, the SOC, the total battery voltage, the total battery current, the lowest battery cell voltage value, the highest probe temperature value, the lowest probe temperature value, the highest alarm level and the like can be reserved.

The acquired vehicle-mounted data can be subjected to certain data preprocessing, so that standardized vehicle-mounted data can be acquired to meet the input and prediction requirements of a subsequent cyclic neural network model.

S102, coding the standardized vehicle-mounted data to obtain vehicle-mounted characteristics;

the coding mode can be a common coding mode and is mainly used for realizing that the coded vehicle-mounted characteristics can be adapted to the recurrent neural network model, so that the recurrent neural network can predict the fault type based on the coded vehicle-mounted characteristics.

S103, inputting the vehicle-mounted characteristics into a pre-trained recurrent neural network to obtain the fault type output by the recurrent neural network, and completing the prediction of the vehicle fault type.

The failure grade prediction is to predict the lagging failure grade based on the reported historical vehicle-mounted data, is a time series prediction problem and focuses on capturing the incidence relation between the historical message and the lagging failure grade. The recurrent neural network can better capture time dimension information of the sequence to form prediction of information of future time. Therefore, the fault type is predicted through the recurrent neural network, and the prediction accuracy can be improved.

The recurrent neural network needs to be trained in advance, and can be used for fault grade prediction after training. And pre-training based on training sample data to obtain an available recurrent neural network as a fault level prediction model. During specific training, a Neural Network layer architecture based on an RNN (Recurrent Neural Network) is adopted, corresponding loss functions, evaluation functions, training iteration times, batch processing sample sizes, optimizer settings and the like are configured, and then multiple rounds of training are carried out until the loss functions converge and the evaluation functions reach preset prediction accuracy and the like.

And for the trained model, an expected prediction effect is achieved, and the model can be deployed to a production environment to perform actual fault level prediction.

The output result of the recurrent neural network can be the probability value of each type of fault, and the fault level labels can be divided into four types, which can be respectively represented by 0,1, 2 and 3. In order to facilitate the model to automatically calculate the loss value of each training round according to the preset loss function, the fault level label of the original sample data can be converted, and specifically, the fault level can be encoded by using one-hot encoding.

The one-hot coding is the representation of a classification variable as a binary vector, under the representation, N fault levels correspond to N state codes, only one state code of a certain fault level label is marked as 1, and the rest state codes are marked as 0. If there are four categories of features to be processed (e.g., failure classes), the one-hot codes are respectively: 0001,0010,0100,1000, respectively corresponding to 4 different types of values for the tag.

In addition, consider that in the actual reported data, the proportion of the fault message is lower than 2%, that is, there is a serious imbalance between the fault and the non-fault. In order to improve the sensitivity of the recurrent neural network to the fault data, in this embodiment, the operation of resampling may be performed on training sample data. The specific operation is that the resampling operation of the training sample is carried out according to the proportion of non-fault to fault of 5:1; while the test samples for the model test phase do not perform the resampling operation.

For the resampled training samples, the present embodiment adopts the GRU and the LSTM to respectively perform model training, and the relevant parameter setting forms are as follows:

description of model part parameter settings:

loss function: categoricalCrossentpy;

evaluation function: accuracy;

training iteration times: 12;

batch sample size: 256 of;

gradient descent optimizer: adam (0.001);

after the parameters are set according to the above values, the following network structure can be built:

GRU network architecture: fig. 2 shows a diagram of a neural network structure of a GRU (neuron number 80, activation function = 'tanh'), dropout (ratio 0.2), a GRU network (neuron number 80, activation function = 'tanh'), dropout (ratio 0.2), and Dense (activation function = 'softmax'). In FIG. 2, GRU-input represents the GRU network model input; the input layer represents the GRU network model input layer, the input represents the input, and the output represents the output; GRU, dropout, and density represent different network layers in the GRU network, respectively.

LSTM network architecture: one layer of LSTM (number of neurons 80, activation function = 'tanh'), dropout (ratio 0.2), one layer of LSTM network (number of neurons 80, activation function = 'tanh'), dropout (ratio 0.2), dense (activation function = 'softmax'), and the neural network structure diagram of the LSTM model is shown in fig. 3. In FIG. 3, LSTM-input represents the LSTM network model input; the input layer represents an LSTM network model input layer, the input represents input, and the output represents output; LSTM, dropout, and density represent different network layers in the LSTM network, respectively.

And training the two recurrent neural network models based on the training samples according to the set model parameters and model structure. The model effect of this embodiment shows that the accuracy of the GRU model is improved to a certain extent compared with the LSTM, and the consolidation prediction accuracy can reach the service expectation. Therefore, the GRU model can be selected as a trained recurrent neural network model.

Based on the trained recurrent neural network model, the model file is saved and deployed to a production environment, and the model file can be used for predicting the actual fault level.

Of course, in other implementations, the recurrent neural network model may be optimized in other manners, and specifically, the following manners may be included:

1. in the aspect of model architecture, increasing the number of network layers or replacing the network architecture can be considered, for example, an attention mechanism is used to replace a common RNN architecture, so as to better capture the association relationship of the time series;

2. in the aspect of configuration optimization, different weights are given to the accuracy of vehicle fault early warning and non-fault early warning, and in actual service, if the actual condition of a vehicle is mainly concerned as the prediction accuracy when a fault occurs, the weight of fault data can be increased or the punishment weight when fault prediction is wrong is increased, so that the accuracy of fault data prediction is improved;

3. in the aspect of setting of the target label, the fault level of the fault can be adjusted to be the highest fault level in a period of time, so that the stability of model prediction can be improved;

4. the length and the number of observation windows of the features can be adjusted properly to determine a proper observation length; in addition to other information such as voltage values of each battery cell, temperature values of probes, and the like, other non-interface reported data information, such as vehicle static information, other external environment information, and the like, introduced by the vehicle data, may be considered to be added to actually report the vehicle data for model prediction.

By applying the technical scheme provided by the embodiment of the invention, various vehicle-mounted data during the running of the automobile can be collected and processed according to time and vehicle dimensions to form a two-dimensional matrix, a multi-dimensional time sequence prediction model based on a recurrent neural network is established, and the fault level at the future moment is predicted in advance. The vehicle-mounted data reported by new energy in real time are utilized, the cyclic neural network algorithm is fused, the rapid, advanced and accurate prediction of the vehicle fault level is realized, the advanced occurrence of the vehicle fault can be effectively reduced, and the potential fault secondary loss is reduced.

Therefore, the vehicle fault alarm type is predicted by applying the embodiment of the invention based on the vehicle data uploaded by the vehicle terminal in real time and adopting the recurrent neural network structure. The accuracy of the prediction result is improved, and the advance prediction of the vehicle fault alarm type is realized.

In one implementation, the preprocessing the vehicle-mounted data to obtain normalized vehicle-mounted data includes:

The specific manner of deleting invalid data may be designed according to actual requirements, which is not limited in the present invention. In one implementation mode, invalid data information of each field can be eliminated by combining with interface specification of a T-box; or, the non-effective index which does not meet the standard requirement can be deleted by referring to the definition of the vehicle-mounted effective index in the national standard. For example, reference may be made to the "Standard of the national Standard of the people's republic of China GB/T32960-2016"; according to the standard, threshold value description and requirements are made on the vehicle speed, the accumulated mileage, the SOC, the total voltage, the total current, the lowest value of the single battery voltage, the highest temperature value of the probe, the lowest temperature value of the probe, the highest alarm level and the like, and in the implementation, the requirements are referred to, and invalid data which do not accord with the standard are eliminated.

The specific manner of deleting the duplicated data may be designed according to actual requirements, which is not limited in the present invention. In consideration of the situation that repeated reporting and repeated storage of data possibly exist in vehicle-mounted data, the invention performs duplicate removal processing on the repeated data so as to improve the efficiency of subsequent data processing.

The input of the cyclic neural network model is various vehicle-mounted data acquired in real time, and particularly relates to data of types such as speed, total meter mileage, SOC, total battery voltage, total battery current, minimum battery voltage, maximum battery cell voltage, maximum temperature value and minimum temperature value, the distribution of values of different types of data has great difference, and in order to accelerate the convergence rate of the model and prevent gradient explosion problems during training, the vehicle-mounted data can be subjected to standardization processing before being input into the model so as to eliminate dimension difference of the vehicle-mounted data.

Normalization of data (normalization) is to scale data to fall within a small specific interval. In some index processing for comparison and evaluation, unit limitation of data is removed and converted into a dimensionless pure numerical value, so that indexes of different units or orders can be compared and weighted conveniently. The most typical of them is the normalization processing of data, i.e. the data is mapped onto the interval of [0,1] uniformly.

In one implementation, the normalizing the valid vehicle-mounted data to obtain normalized vehicle-mounted data includes:

The specific normalization processing method is not limited in the present invention, and for example, any one or a combination of the following two normalization methods may be adopted:

one is dispersion Normalization, also known as Min-Max Normalization, which is a linear transformation of the raw data such that the resulting values map between [0,1 ]. The transfer function is as follows:

wherein max is the maximum value of the sample data, min is the minimum value of the sample data, and x _morn Is a value obtained by normalizing the sample data x.

Another is the Z-score normalization method, which normalizes data based on the mean and standard deviation of the raw data. The processed data are in accordance with the standard normal distribution, i.e. the mean value is 0, the standard deviation is 1, and the conversion function is:

where μ is the mean of all sample data, σ is the standard deviation of all sample data, x _norm Is a value obtained by normalizing the sample data x.

Illustratively, the training data and the test data of the recurrent neural network model are divided according to the vehicle number, wherein the message data of 80 trolleys is used for model training, and the message data of the remaining 20 trolleys is used for model testing. The training data was Z-score normalized and the mean and standard deviation were recorded and the test data was similarly manipulated according to the mean and standard deviation.

For example, for the total voltage field data in the vehicle data. Assuming that a total of 200 ten thousand records of the training data set are recorded, each record records the battery voltage value of different corresponding vehicles at specific time, and the mean value mu and the standard deviation delta of the voltage are calculated.

Assume that a certain data voltage is 3.5; the voltage mean of the training data was 3.2, the standard deviation of the voltage of the training data was 0.5, and the value after normalization with Z-score was (3.5-3.2)/0.5 =0.6.

The normalization process can be performed according to the above method for each type of data in the vehicle data.

Similarly, each sample data used for training the recurrent neural network is processed according to the method, and the corresponding sample label can be subjected to one-hot encoding operation, so that the sample data after normalization processing and the sample label after one-hot encoding are obtained and used for training the recurrent neural network model. The invention does not limit the mode of the one-hot coding, for example, the fault level can include 4 types, that is, the one-hot coding is performed based on 0,1, 2 and 3, and the label characteristics representing different fault levels are respectively obtained, for example, the label characteristics represent serious, general, lighter and no fault; the operation of one-hot encoding the fault level is performed based on this.

In one implementation, the vehicle data includes: the vehicle frame number, the data reporting time, the driving speed, the total meter mileage, a chip System (SOC), the total battery voltage, the total battery current, the lowest battery cell voltage value, the highest probe temperature value, the lowest probe temperature value and the highest alarm level.

In one implementation, the recurrent neural network is pre-trained by:

The initial recurrent neural network is an untrained recurrent neural network, and after training, the trained recurrent neural network can be obtained for predicting the type of the vehicle fault.

The first preset value and the second preset value can be set in advance according to requirements, and can be 100 and 3 respectively.

It can be understood that, in the original vehicle-mounted data, one data record is a single message record of a specific time of a vehicle, and the single message record includes data of the vehicle speed, the accumulated mileage, the SOC, the total voltage, the total current, the lowest value of the cell voltage, the highest temperature value of the probe, the lowest temperature value of the probe, the highest fault level and the like recorded this time. In this embodiment, the 9 types of field data and 1 tag value are selected as one sample data.

For example, in this embodiment, the number of times of the observation window is set to be 100, and the predicted number of times of hysteresis is set to be 30, that is, the input of the recurrent neural network model is a two-dimensional array formed by 100 consecutive packet records, and the corresponding output is the fault type tag value of the 30 th reported data point after the packet combination.

Specifically, for a new energy automobile A with the number of 1, 1000 interface message records are stored in a preset time period, and are recorded as 1,2,3,4 and … according to the time sequence; each record contains 9 field data of (x 1, x2, x3, x4 …, x 9), which respectively correspond to 9 field values of vehicle speed, accumulated mileage, SOC, total voltage, total current, cell voltage lowest value, cell voltage highest value, probe highest temperature value, probe lowest temperature value and the like. One of the sample data sets is:

sample input: the number is 1,2,3,4 …, and the field values corresponding to all reported vehicle-mounted data records of 100 are two-dimensional matrixes of 100 × 9;

and (3) outputting a sample: predicted number [ 130 ] corresponding to the fault level.

Thus, sample combinations of observation window features and hysteresis fault classes are constructed in turn, namely:

for each car, this operation can be performed to construct a sample, i.e., a full set of model samples can be obtained. In addition, in the specific implementation, if the number of messages of a single trolley is less than 100+30, the messages can be directly rejected because the sample construction condition is not met.

And further, the recurrent neural network can be trained through the constructed sample set, so that the trained recurrent neural network is obtained. The method adopts a trained network layer of the recurrent neural network to predict the failure grade of the automobile, inputs of the recurrent neural network model are multi-time series data with multi-feature dimensionality, and outputs the failure grade after reporting times of certain fixed data.

Where data is input for a plurality of successive time nodes involving a plurality of features, it may be assembled into a fixed shape for model training or final failure level prediction. Specifically, the characteristic form of the model may be a two-dimensional matrix (observation window length k, characteristic dimension m), and the observation window length k may be determined in combination with specific data.

The output of the model is a fault level after a certain fixed data reporting frequency, and the data acquisition action of the vehicle is considered only in a power-on state, specifically, the two states of vehicle starting or vehicle flameout and charging are mainly considered. Considering that the vehicle does not upload data when the vehicle is shut off and not charged and the time duration is unpredictable, the number of data upload intervals is taken as a predicted step length. That is, assuming that the data reporting numbers input by the model are respectively recorded as 1,2,3,4, …, k, predicting the fault state of the [ k + n ] th reporting point, wherein k is the number of times that the model inputs continuously observes the reported data, and n is the predicted fault level lag number.

The method is not limited to a specific model training mode, and training related parameters such as a loss function, an evaluation function, training iteration times, batch processing sample size, optimizer setting and the like can be set according to requirements. And (3) for the trained model, an expected prediction effect is achieved, and the model can be deployed to a production environment to perform actual fault level prediction.

By applying the embodiment of the invention, the capturing of the incidence relation between the historical message and the hysteresis fault level is realized, and the accuracy of predicting the fault type of the vehicle at the future moment is improved.

The following describes the car fault type prediction device based on the recurrent neural network provided by the embodiment of the present application, and the car fault type prediction device based on the recurrent neural network described below and the car fault type prediction method based on the recurrent neural network described above may be referred to correspondingly.

Referring to fig. 4, a schematic structural diagram of an automobile fault type prediction apparatus based on a recurrent neural network according to an embodiment of the present application is shown, where the apparatus may include: acquisition unit 201, encoding unit 202, and prediction unit 203.

By applying the embodiment of the invention, based on the vehicle-mounted data uploaded by the vehicle-mounted terminal, the vehicle-mounted data is preprocessed to obtain standardized vehicle-mounted data; coding the standardized vehicle-mounted data to obtain vehicle-mounted characteristics; by adopting a cyclic neural network structure, the vehicle fault alarm type is predicted based on the vehicle-mounted characteristics, the vehicle fault alarm type is predicted in advance, and the accuracy of the prediction result is improved.

In a possible implementation manner, the obtaining unit preprocesses the vehicle-mounted data to obtain standardized vehicle-mounted data, specifically:

In a possible implementation manner, the obtaining unit performs normalization processing on the valid vehicle-mounted data to obtain normalized vehicle-mounted data, specifically:

In one possible implementation, the vehicle data includes: the vehicle frame number, the data reporting time, the speed, the total meter mileage, the chip system, the total battery voltage, the total battery current, the lowest value of the single battery voltage, the highest temperature value, the lowest temperature value and the highest alarm level.

In a possible implementation manner, the apparatus further includes a training unit, and the training unit is configured to pre-train the recurrent neural network by:

inputting a two-dimensional array consisting of continuous sample records with a first preset value into the initial cyclic neural network to obtain a predicted sample label output by the initial cyclic neural network after reaching the second preset value;

An embodiment of the present application further provides an electronic device, please refer to fig. 5, which shows a schematic structural diagram of the electronic device, and the electronic device may include: at least one processor 301, at least one communication interface 302, at least one memory 303 and at least one communication bus 304;

in the embodiment of the present application, the number of the processor 301, the communication interface 302, the memory 303 and the communication bus 304 is at least one, and the processor 301, the communication interface 302 and the memory 303 complete communication with each other through the communication bus 304;

the processor 301 may be a central processing unit CPU, or an Application Specific Integrated Circuit ASIC (Application Specific Integrated Circuit), or one or more Integrated circuits configured to implement embodiments of the present invention, etc.;

the memory 303 may include a high-speed RAM memory, and may further include a non-volatile memory (non-volatile memory) or the like, such as at least one disk memory;

wherein the memory stores a program and the processor can call the program stored in the memory, the program for:

a method for predicting a type of vehicle fault based on a recurrent neural network, the method comprising:

Alternatively, the detailed function and the extended function of the program may be as described above.

Embodiments of the present application further provide a readable storage medium, where a program suitable for being executed by a processor may be stored, where the program is configured to:

By applying the embodiment of the invention, the vehicle-mounted data uploaded by the vehicle-mounted terminal is preprocessed to obtain the standardized vehicle-mounted data; coding the standardized vehicle-mounted data to obtain vehicle-mounted characteristics; by adopting the cyclic neural network structure, the vehicle fault alarm type is predicted in advance, and the accuracy of the prediction result is improved.

Alternatively, the detailed function and the extended function of the program may refer to the above description.

Finally, it should also be noted that, herein, relational terms such as first and second, and the like may be used solely to distinguish one entity or action from another entity or action without necessarily requiring or implying any actual such relationship or order between such entities or actions. Also, the terms "comprises," "comprising," or any other variation thereof, are intended to cover a non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements does not include only those elements but may include other elements not expressly listed or inherent to such process, method, article, or apparatus. Without further limitation, an element defined by the phrase "comprising a … …" does not exclude the presence of another identical element in a process, method, article, or apparatus that comprises the element.

The embodiments in the present description are described in a progressive manner, each embodiment focuses on differences from other embodiments, and the same and similar parts among the embodiments are referred to each other.

The previous description of the disclosed embodiments is provided to enable any person skilled in the art to make or use the present invention. Various modifications to these embodiments will be readily apparent to those skilled in the art, and the generic principles defined herein may be applied to other embodiments without departing from the spirit or scope of the invention. Thus, the present invention is not intended to be limited to the embodiments shown herein but is to be accorded the widest scope consistent with the principles and novel features disclosed herein.

Claims

1. A vehicle fault type prediction method based on a recurrent neural network is characterized by comprising the following steps:

2. The method of claim 1, wherein the pre-processing the vehicle-mounted data to obtain normalized vehicle-mounted data comprises:

3. The method of claim 2, wherein the normalizing the valid vehicle-mounted data to obtain normalized vehicle-mounted data comprises:

4. The method of claim 1, wherein the vehicle data comprises: the system comprises a vehicle frame number, data reporting time, speed, total meter mileage, a chip system, total battery voltage, total battery current, minimum single battery voltage, maximum probe temperature, minimum probe temperature and at least one of the highest alarm level.

5. The method of claim 1, wherein the recurrent neural network is pre-trained by:

constructing a sample set based on the first preset number of consecutive sample records and the sample label;

6. An automobile fault type prediction device based on a recurrent neural network, characterized in that the device comprises:

7. The device according to claim 6, wherein the obtaining unit preprocesses the vehicle-mounted data to obtain standardized vehicle-mounted data, specifically:

deleting invalid data and repeated data in the vehicle-mounted data to obtain effective vehicle-mounted data;

8. The device according to claim 7, wherein the obtaining unit performs a normalization process on the valid vehicle-mounted data to obtain normalized vehicle-mounted data, specifically:

9. An electronic device, comprising:

a memory storing executable program code;

a processor coupled with the memory;

the processor calls the executable program code stored in the memory for executing the recurrent neural network-based automobile fault type prediction method of any one of claims 1 to 5.

10. A computer-readable storage medium storing a computer program, wherein the computer program causes a computer to execute the recurrent neural network-based vehicle failure type prediction method of any one of claims 1 to 5.