CN-121989751-A - Electric vehicle battery dynamic safety monitoring method and system based on Internet of things

CN121989751ACN 121989751 ACN121989751 ACN 121989751ACN-121989751-A

Abstract

The invention discloses an electric vehicle battery dynamic safety monitoring method and system based on the Internet of things, and relates to the technical field of battery safety monitoring; the method comprises the steps of firstly obtaining low-frequency initial battery data of a target vehicle, generating initial samples through rapid preprocessing and time sequence division, inputting a hybrid prediction model to obtain initial two-dimensional prediction characteristics of the highest battery cell voltage and probe temperature, detecting the initial two-dimensional prediction characteristics through a target algorithm to obtain initial fault probability, adaptively adjusting sampling frequency according to the initial two-dimensional prediction characteristics, collecting high-frequency update data, uploading the high-frequency update data and the initial fault probability to a cloud, updating parameters of the hybrid prediction model through an optimization algorithm by the cloud, substituting the update data into the optimized model to obtain target two-dimensional characteristics, detecting abnormality and outputting accurate fault probability, and accordingly making a hierarchical monitoring scheme. The problem of current electric motor car battery safety monitoring because parameter and model dual solidification lead to electric motor car battery safety monitoring inefficiency is solved.

Inventors

YAO XIAOQING
XU YONG
LI YUAN
WANG DEYING

Assignees

安徽超电新能源发展有限公司

Dates

Publication Date: 20260508
Application Date: 20260318

Claims (10)

1. The method for dynamically monitoring the safety of the electric vehicle battery based on the Internet of things is characterized by comprising the following steps of: After battery running state data acquired by a target vehicle at an initial sampling frequency are acquired and are subjected to rapid preprocessing, dividing the preprocessed battery running state data by a preset time step to obtain an initial input sample set; Substituting the initial input sample set into a mixed prediction model to obtain initial two-dimensional characteristics, wherein the initial two-dimensional characteristics are predicted values corresponding to the highest single-cell voltage and the highest probe temperature in different time steps; performing fault detection on the initial two-dimensional characteristics based on a target algorithm to obtain initial fault probability; Aiming at the preliminary fault probability, determining a target sampling frequency based on a preset rule, and acquiring the battery running state of a target vehicle at the target sampling frequency to obtain updated battery sampling data; uploading the updated battery sampling data and the preliminary fault probability to a cloud server, so that the cloud server performs parameter updating on the hybrid prediction model by applying a preset optimization algorithm to obtain optimization parameters; updating the hybrid prediction model according to the optimization parameters, and substituting the updated battery sampling data into the updated hybrid prediction model to obtain target two-dimensional characteristics; and carrying out anomaly detection on the target two-dimensional characteristics based on a target algorithm to obtain target fault probability, and determining a monitoring scheme according to the target fault probability.
2. The method for monitoring the dynamic safety of the electric vehicle battery based on the Internet of things according to claim 1, wherein the hybrid prediction model comprises a TCN feature extraction sub-module, an LSTM time sequence modeling sub-module and a parallel prediction output layer which are connected in sequence; the TCN feature extraction submodule sequentially comprises a linear embedded layer, three causal expansion convolution layers, a residual error connecting layer, a Dropout layer and a linear compression layer from input to output, wherein the convolution kernels of the three causal expansion convolution layers are 3, and expansion factors are sequentially set to be 1,2 and 4; The LSTM time sequence building module sequentially comprises a first LSTM layer, a second LSTM layer and a Dropout layer from input to output, wherein the hidden layer dimension of the first LSTM layer is 128, a return sequence mode is started, the hidden layer dimension of the second LSTM layer is 64, the return sequence mode is closed, and the discarding rate of the Dropout layer is set to 0.5; the parallel prediction output layer comprises two stages of full-connection layers.
3. The method for monitoring the dynamic safety of the electric vehicle battery based on the internet of things according to claim 1, wherein the step of performing anomaly detection on the initial two-dimensional feature based on a target algorithm to obtain the initial fault probability comprises the following steps: Taking each element in the initial two-dimensional feature matrix as a predicted data point; calculating the distance between each predicted data point and other data points based on the Euclidean distance, and determining the k neighborhood of each data point; calculating an reachable distance of each data point of the predicted data point in the k neighborhood aiming at each predicted data point, and determining the local reachable density of the predicted data point according to the reachable distance; calculating a local anomaly factor for each predicted data point based on the local reachable density of the predicted data point and the local reachable density of the corresponding k neighborhood; By the formula And converting the local abnormal factor into a preliminary fault probability, wherein A is the local abnormal factor, and when the local abnormal factor is smaller than or equal to 1, the preliminary fault probability is zero.
4. The method for dynamically monitoring safety of an electric vehicle battery based on the internet of things according to claim 3, wherein determining the target sampling frequency based on the preset rule for the preliminary failure probability comprises: When the local reachable density corresponding to the predicted data point is greater than 1, the data point is marked as an abnormal detection point; Calculating the duty ratio of the abnormal detection points in all the predicted data points to obtain an abnormal duty ratio; If the abnormal occupation ratio is larger than a preset threshold value, calculating a preliminary fault probability average value of all abnormal detection points; And determining a target sampling frequency according to the preliminary fault probability average value and a preset fault threshold value.
5. The method for dynamically monitoring the safety of the electric vehicle battery based on the internet of things according to claim 1, wherein uploading the updated battery sampling data and the preliminary fault probability to a cloud server, so that the cloud server performs parameter updating on the hybrid prediction model by applying a preset optimization algorithm to obtain the optimized parameters comprises: Taking updated battery sampling data as a model input sample, and taking corresponding preliminary fault probability as a supervision tag to construct a training sample set for model optimization; Constructing a model training loss function according to the deviation between the predicted value output by the hybrid prediction model and the true value in the updated battery sampling data; Iteratively adjusting network coefficients of the hybrid prediction model based on the gradient of the loss function through a preset optimization algorithm; And stopping parameter iteration when the loss function converges to a preset threshold value or the iteration times reach the preset times, and determining model parameters meeting convergence conditions after iteration is completed as optimization parameters.
6. Electric vehicle battery dynamic safety monitoring system based on thing networking, its characterized in that, the system includes: The data dividing module is used for acquiring battery running state data acquired by a target vehicle at an initial sampling frequency, performing rapid preprocessing, and dividing the preprocessed battery running state data by a preset time step to obtain an initial input sample set; the initial two-dimensional characteristic determining module is used for substituting the initial input sample set into a mixed prediction model to obtain an initial two-dimensional characteristic, wherein the initial two-dimensional characteristic is a predicted value corresponding to the highest single-cell voltage and the highest probe temperature in different time steps; the initial fault probability determining module is used for carrying out fault detection on the initial two-dimensional characteristics based on a target algorithm to obtain initial fault probability; the target sampling frequency determining module is used for determining a target sampling frequency based on a preset rule aiming at the preliminary fault probability, and acquiring the battery running state of the target vehicle at the target sampling frequency to obtain updated battery sampling data; The model optimization parameter determining module is used for uploading the updated battery sampling data and the preliminary fault probability to a cloud server, so that the cloud server performs parameter updating on the hybrid prediction model by applying a preset optimization algorithm to obtain optimization parameters; The target two-dimensional feature determining module is used for updating the hybrid prediction model according to the optimization parameters, and substituting the updated battery sampling data into the updated hybrid prediction model to obtain target two-dimensional features; the monitoring scheme generation module is used for carrying out anomaly detection on the target two-dimensional characteristics based on a target algorithm to obtain target fault probability, and determining a monitoring scheme according to the target fault probability.
7. The system for monitoring the dynamic safety of the electric vehicle battery based on the Internet of things according to claim 6, wherein the hybrid prediction model comprises a TCN feature extraction sub-module, an LSTM time sequence modeling sub-module and a parallel prediction output layer which are connected in sequence; the TCN feature extraction submodule sequentially comprises a linear embedded layer, three causal expansion convolution layers, a residual error connecting layer, a Dropout layer and a linear compression layer from input to output, wherein the convolution kernels of the three causal expansion convolution layers are 3, and expansion factors are sequentially set to be 1,2 and 4; The LSTM time sequence building module sequentially comprises a first LSTM layer, a second LSTM layer and a Dropout layer from input to output, wherein the hidden layer dimension of the first LSTM layer is 128, a return sequence mode is started, the hidden layer dimension of the second LSTM layer is 64, the return sequence mode is closed, and the discarding rate of the Dropout layer is set to 0.5; the parallel prediction output layer comprises two stages of full-connection layers.
8. The electric vehicle battery dynamic safety monitoring system based on the internet of things of claim 6, wherein the preliminary fault probability determination module comprises: The predicted data point generation module is used for taking each element in the initial two-dimensional feature matrix as a predicted data point; The neighborhood dividing module is used for calculating the distance between each predicted data point and other data points based on the Euclidean distance and determining the k neighborhood of each data point; The local reachable density determining module is used for calculating the reachable distance of each data point in the k neighborhood of each predicted data point according to each predicted data point, and determining the local reachable density of the predicted data point according to the reachable distance; The local anomaly factor determining module is used for calculating the local anomaly factor of each predicted data point based on the local reachable density of the predicted data point and the local reachable density of the corresponding k neighborhood; A local abnormal factor conversion module for passing through the formula And converting the local abnormal factor into a preliminary fault probability, wherein A is the local abnormal factor, and when the local abnormal factor is smaller than or equal to 1, the preliminary fault probability is zero.
9. The system for dynamic safety monitoring of an electric vehicle battery based on the internet of things of claim 8, wherein the target sampling frequency determining module comprises: the abnormal detection point marking module is used for marking the data point as an abnormal detection point when the local reachable density corresponding to the predicted data point is greater than 1; the abnormal occupation ratio determining module is used for calculating the occupation ratio of the abnormal detection points in all the predicted data points to obtain an abnormal occupation ratio; the preliminary fault probability average value determining module is used for calculating the preliminary fault probability average value of all the abnormal detection points if the abnormal occupation ratio is larger than a preset threshold value; and the target sampling frequency generation module is used for determining the target sampling frequency according to the preliminary fault probability average value and the preset fault threshold value.
10. The system for monitoring the dynamic safety of the electric vehicle battery based on the internet of things according to claim 6, wherein the model optimization parameter determining module comprises: The training sample set generation module is used for taking updated battery sampling data as a model input sample and corresponding preliminary fault probability as a supervision tag to construct a training sample set for model optimization; the loss function construction module is used for constructing a loss function of model training according to the deviation between the predicted value output by the hybrid prediction model and the true value in the updated battery sampling data; the network coefficient optimization module is used for iteratively adjusting the network coefficient of the hybrid prediction model based on the gradient of the loss function through a preset optimization algorithm; and the optimization parameter generation module is used for stopping parameter iteration when the loss function converges to a preset threshold value or the iteration number reaches the preset number, and determining model parameters meeting convergence conditions after the iteration is completed as optimization parameters.

Description

Electric vehicle battery dynamic safety monitoring method and system based on Internet of things Technical Field The invention belongs to the technical field of battery safety monitoring, and particularly relates to an electric vehicle battery dynamic safety monitoring method and system based on the Internet of things. Background With the rapid development of the electric automobile industry, the power battery is used as a core power source of the electric automobile, and the running safety of the power battery is directly related to the running safety of the automobile, the personal safety and the property safety of drivers and passengers, and is one of key constraint factors of the large-scale development of the electric automobile industry. Currently, an electric vehicle battery safety monitoring system based on the internet of things technology has become a mainstream technical direction of industry, and the core requirements are that early warning and dynamic monitoring of battery faults are realized by collecting battery running state data and analyzing battery running trend, so that safety accidents such as battery thermal runaway and ignition are avoided. The existing monitoring scheme generally adopts a monitoring mode with fixed sampling frequency and fixed model parameters, the core contradiction exists in that the requirements of prediction precision, terminal calculation power consumption and internet of things transmission bandwidth cannot be simultaneously considered, the two difficulties of sacrificing safety and wasting resources are overcome, specifically, if the low initial sampling frequency is adopted, the vehicle end power consumption can be reduced, the internet of things bandwidth is saved, the edge calculation pressure is lightened, but the battery running state data acquisition is not dense, the tiny voltage and temperature abnormal signals are easy to miss, the early failure feature cannot be captured by the mixed prediction model, the failure detection rate is increased, the core requirement of dynamic safety monitoring cannot be met, if the high initial sampling frequency is adopted, the tiny running abnormality of a battery can be captured, the model prediction precision is improved, the early failure early warning is realized, the continuous high-frequency sampling of the vehicle end can cause the power consumption to be increased sharply, the continuous driving of an electric vehicle is seriously influenced, a large amount of sampling data can occupy too much internet of things bandwidth and increase the transmission cost, meanwhile, the edge end needs to process massive data, the calculation pressure is too large, the model training efficiency can be reduced after receiving the data, and iteration calculation resource waste is caused. Further, the existing scheme also has secondary hidden dangers caused by model solidification, which is also an important reason for causing long-term attenuation of monitoring precision, after the vehicle model is deployed, parameters of the vehicle model are fixed and cannot adapt to battery operation state changes caused by battery aging, environment changes and use scene differences, in the battery operation process, the change rule of voltage and temperature can change along with the use duration and environment changes, a model aging problem of increased prediction deviation can be gradually caused by a mixed prediction model with fixed parameters, the accuracy of fault detection is further reduced, early faults which can be detected originally can gradually become missed detection, and finally the core significance of monitoring is lost. In summary, the existing electric vehicle battery safety monitoring system based on the internet of things has the core defect of dual curing of parameters and models, the sampling frequency and the model parameters are fixedly set, the sampling frequency curing directly causes the triangle contradiction of data redundancy, power consumption waste and effective information shortage, the model parameter curing causes the problem that the model cannot adapt to battery aging, working condition change and continuous precision attenuation, and the two curing problems are mutually restricted and commonly influenced, so that the electric vehicle battery safety monitoring efficiency is low. Disclosure of Invention The invention aims to solve the problem that the existing electric vehicle battery safety monitoring commonly has double curing of parameters and models, so that the electric vehicle battery safety monitoring efficiency is low, and provides an electric vehicle battery dynamic safety monitoring method and system based on the Internet of things. In a first aspect of the present invention, firstly, an electric vehicle battery dynamic security monitoring method based on internet of things is provided, the method includes: After battery running state data acquired by a target vehicle at an initial sampling