Electric vehicle transportation accident situation prediction method and system based on time sequence evolution model

Abstract

The invention discloses an electric vehicle transportation accident situation prediction method and system based on a time sequence evolution model, and the method comprises the steps: collecting battery data and transportation vehicle data in the historical transportation process of an electric vehicle, and obtaining a historical dynamic risk sequence; processing parameter changes at continuous time points in the historical dynamic risk sequence, and determining historical risk evolution vectors; constructing a long short-term memory network, and training the long short-term memory network by using the historical risk evolution vector and the historical accident condition to obtain an accident evolution model; acquiring a current dynamic risk sequence, inputting the current dynamic risk sequence into the accident evolution model to simulate the risk state evolution of a future time step, and judging a potential accident trigger point; determining a specific accident type based on the potential accident trigger point; and generating a prediction precision evaluation report according to the determined accident type and outputting a trend intervention signal to obtain a final accident situation prediction result.

Description

Technical Field

[0001] This invention relates to the field of transportation and accident prediction technology, specifically to a method and system for predicting the situation of electric vehicle transportation accidents based on a time-series evolution model. Background Technology

[0002] Electric vehicle transportation, as a crucial link in the new energy vehicle industry, directly impacts supply chain efficiency and public safety. Accident prediction is vital for minimizing economic losses and ensuring safety. With the increasing scale of electric vehicle use, accidents involving batteries, vehicles, and charging infrastructure during transportation are becoming more frequent, necessitating accurate prediction of accident types and trends for early intervention. However, existing prediction methods lack applicability in complex scenarios and struggle to cope with the changing environment and dynamic risks during transportation, thus limiting the efficiency and accuracy of accident prevention and control.

[0003] Current accident prediction methods largely rely on static data analysis or monitoring of single risk points, neglecting the dynamic evolution of risks during transportation. When faced with various accident types in electric vehicle transportation (such as battery thermal runaway, collisions, loss of control, or charging failures), these methods often fail to accurately distinguish the occurrence patterns of different accident types. In particular, existing technologies are deficient in capturing the temporal evolution characteristics before an accident occurs, making it difficult to extract key patterns of risk evolution from continuous dynamic data. This results in predictions lacking specificity and failing to provide accurate early warnings for specific accident types.

[0004] A key technical challenge lies in effectively tracking the entire timeline from potential risk to actual occurrence of an accident. During transportation, the risk state changes continuously over time. For example, battery performance degradation under high temperatures or vibrations can gradually evolve into thermal runaway, but current technologies struggle to record and analyze the parameters of these state changes in real time. This lack of dynamic tracking directly prevents the establishment of an accurate mapping between different accident types and their temporal evolution characteristics. For instance, during long-distance transportation, batteries may suffer minor damage due to continuous vibration, but existing systems struggle to capture how these minor changes gradually accumulate into significant accident risks.

[0005] Another related technical challenge is the refined identification of accident categories. Because the causes and evolution paths of different types of accidents (such as battery-related accidents and collision accidents) differ significantly, the lack of a targeted classification system makes it difficult for prediction systems to determine the specific type of accident that may occur in the future. For example, charging accidents may be caused by abnormal voltage, while collision accidents may stem from driving behavior. However, existing technologies often confuse these characteristics, making it difficult to accurately distinguish accident categories and predict their probability of occurrence in dynamic transportation scenarios.

[0006] Therefore, how to track the temporal evolution characteristics of risk status in real time in the complex and dynamic scenario of electric vehicle transportation, and establish accurate mapping relationships between different accident categories, has become a key issue in building an efficient accident situation prediction system. Summary of the Invention

[0007] To address the above technical problems, this invention provides a method for predicting the situation of electric vehicle transportation accidents based on a time-series evolution model, comprising the following steps: Collect battery temperature, battery vibration data, battery voltage, transport vehicle speed, and transport vehicle location during the historical transportation process of electric vehicles to obtain a historical dynamic risk sequence; The parameter changes at consecutive time points in the historical dynamic risk sequence are processed using a time-series feature extraction method to determine the historical risk evolution vector; A long short-term memory network is constructed, and the network is trained using the historical risk evolution vector and historical accident information to obtain an accident evolution model. Obtain the current dynamic risk sequence and input it into the accident evolution model to simulate the evolution of risk states in future time steps and determine potential accident trigger points; If the potential accident trigger point exceeds the third preset threshold, the multidimensional features in the risk state evolution are processed by type differentiation to obtain the accident category probability distribution, and the specific accident type is determined based on the accident category probability distribution. Based on the determined accident type, a prediction accuracy assessment report is generated and a trend intervention signal is output to obtain the final accident situation prediction result.

[0008] Preferably, the method for obtaining the historical dynamic risk sequence includes: Collect battery temperature, battery vibration data, battery voltage, transport vehicle speed, and transport vehicle location data during the historical transportation process of electric vehicles to generate a raw dataset containing time series data. If the battery temperature exceeds the first preset threshold, anomalies in the time series are extracted using a sliding window algorithm to obtain a battery temperature anomaly sequence. Based on the battery temperature anomaly sequence, combined with the battery vibration data and the battery voltage, the K-means clustering algorithm is used to classify the data and determine the battery status risk level. If the battery status risk level is high, the mileage is calculated based on the speed and location of the transport vehicle to obtain a mileage sequence related to the battery status. Based on the driving mileage sequence and ambient temperature, a linear regression algorithm is used to predict the future trend of battery status and generate a dynamic risk sequence. By using the dynamic risk sequence and the collection time, the fluctuation cycle of the risk level is determined, the data frequency is adjusted, and the optimized historical dynamic risk sequence is generated.

[0009] Preferably, the method for determining the historical risk evolution vector includes: The statistical features of the historical dynamic risk sequence are extracted using the sliding window method to obtain a time-series feature set; Based on the time series feature set, the risk change trend is determined. If the fluctuation range of the risk change trend exceeds the second preset threshold, the principal component analysis method is used to reduce the dimensionality of the time series feature set to obtain a dimensionality-reduced feature set. Based on the dimensionality reduction feature set, a clustering method is used to classify the risk change trend, determine the risk category, and then a vector mapping method is used to combine the risk category with the dimensionality reduction feature set to generate an initial risk evolution vector. The initial risk evolution vector is standardized to obtain a standardized vector, and the standardized vector and the time series are fused using a weighted average method to obtain the historical risk evolution vector.

[0010] Preferably, the method for obtaining the accident type includes: If the trigger point exceeds the third preset threshold, the risk evolution vector is obtained, multi-dimensional feature data is extracted, and the multi-dimensional feature data is dimensionality reduced by the principal component analysis algorithm to obtain a dimensionality-reduced feature set. Based on the dimensionality reduction feature set, an initial accident category probability distribution is generated using the support vector machine algorithm. If the maximum probability value in the initial accident category probability distribution is lower than a preset classification threshold, the dimensionality reduction feature set is then classified a second time using the random forest algorithm to obtain the accident category probability distribution. If the probability distribution of the accident category meets the preset threshold, the accident type is determined by the trained classification model.

[0011] The present invention also provides an electric vehicle transportation accident situation prediction system based on a time-series evolution model. The system applies the above-mentioned method and includes: a data acquisition module, a feature extraction module, a model construction module, an accident simulation module, a type determination module, and a result prediction module. The data acquisition module is used to collect battery temperature, battery vibration data, battery voltage, transport vehicle speed and transport vehicle location during the historical transportation process of electric vehicles, and obtain historical dynamic risk sequences. The feature extraction module uses a temporal feature extraction method to process parameter changes at consecutive time points in the historical dynamic risk sequence to determine the historical risk evolution vector; The model building module is used to construct a long short-term memory network, and the long short-term memory network is trained using the historical risk evolution vector and historical accident information to obtain an accident evolution model; The accident simulation module is used to obtain the current dynamic risk sequence, input it into the accident evolution model to simulate the evolution of risk state in future time steps, and determine potential accident trigger points; In the type determination module, if the potential accident trigger point exceeds the third preset threshold, the multidimensional features in the risk state evolution are processed by type differentiation to obtain the accident category probability distribution, and the specific accident type is determined based on the accident category probability distribution. The result prediction module generates a prediction accuracy assessment report and outputs a trend intervention signal based on the determined accident type to obtain the final accident situation prediction result.

[0012] Preferably, the workflow of the data acquisition module includes: Collect battery temperature, battery vibration data, battery voltage, transport vehicle speed, and transport vehicle location data during the historical transportation process of electric vehicles to generate a raw dataset containing time series data. If the battery temperature exceeds the first preset threshold, anomalies in the time series are extracted using a sliding window algorithm to obtain a battery temperature anomaly sequence. Based on the battery temperature anomaly sequence, combined with the battery vibration data and the battery voltage, the K-means clustering algorithm is used to classify the data and determine the battery status risk level. If the battery status risk level is high, the mileage is calculated based on the speed and location of the transport vehicle to obtain a mileage sequence related to the battery status. Based on the driving mileage sequence and ambient temperature, a linear regression algorithm is used to predict the future trend of battery status and generate a dynamic risk sequence. By using the dynamic risk sequence and the collection time, the fluctuation cycle of the risk level is determined, the data frequency is adjusted, and the optimized historical dynamic risk sequence is generated.

[0013] Preferably, the workflow of the feature extraction module includes: The statistical features of the historical dynamic risk sequence are extracted using the sliding window method to obtain a time-series feature set; Based on the time series feature set, the risk change trend is determined. If the fluctuation range of the risk change trend exceeds the second preset threshold, the principal component analysis method is used to reduce the dimensionality of the time series feature set to obtain a dimensionality-reduced feature set. Based on the dimensionality reduction feature set, a clustering method is used to classify the risk change trend, determine the risk category, and then a vector mapping method is used to combine the risk category with the dimensionality reduction feature set to generate an initial risk evolution vector. The initial risk evolution vector is standardized to obtain a standardized vector, and the standardized vector and the time series are fused using a weighted average method to obtain the historical risk evolution vector.

[0014] Preferably, the workflow of the type determination module includes: If the trigger point exceeds the third preset threshold, the risk evolution vector is obtained, multi-dimensional feature data is extracted, and the multi-dimensional feature data is dimensionality reduced by the principal component analysis algorithm to obtain a dimensionality-reduced feature set. Based on the dimensionality reduction feature set, an initial accident category probability distribution is generated using the support vector machine algorithm. If the maximum probability value in the initial accident category probability distribution is lower than a preset classification threshold, the dimensionality reduction feature set is then classified a second time using the random forest algorithm to obtain the accident category probability distribution. If the probability distribution of the accident category meets the preset threshold, the accident type is determined by the trained classification model.

[0015] Compared with the prior art, the beneficial effects of the present invention are as follows: This invention addresses the potential accident risks arising from the dynamic changes of multiple parameters such as battery temperature, vibration, voltage, vehicle speed, and position during electric vehicle transportation. It generates a dynamic risk sequence by real-time acquisition of multi-dimensional parameter data, processes parameter changes at continuous time points using a time-series feature extraction method, generates a risk evolution vector, and utilizes a Long Short-Term Memory (LSTM) network to capture long-term dependencies. This constructs an accident tracking model to simulate the evolution of future risk states and identify potential accident trigger points. When a trigger point exceeds a threshold, a type differentiation module is activated. By combining historical accident data with a random forest classifier, it accurately identifies accident types such as battery thermal runaway or collisions, ultimately generating a prediction accuracy assessment report and outputting trend intervention signals. This invention, through multi-dimensional data fusion and deep learning technology, achieves accurate prediction and classification of potential accidents during transportation, significantly improving safety management efficiency and accident prevention capabilities, and providing intelligent safety assurance for electric vehicle transportation. Attached Figure Description

[0016] To more clearly illustrate the technical solution of the present invention, the drawings used in the embodiments are briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0017] Figure 1This is a schematic diagram of the method flow according to an embodiment of the present invention. Detailed Implementation

[0018] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0019] It should be noted that, unless otherwise defined, the technical or scientific terms used in the embodiments of this disclosure should have the ordinary meaning understood by one of ordinary skill in the art to which this disclosure pertains. The terms "first," "second," and similar terms used in the embodiments of this disclosure do not indicate any order, quantity, or importance, but are merely used to distinguish different components. Terms such as "comprising" or "including" mean that the element or object preceding the word encompasses the elements or objects listed following the word and their equivalents, without excluding other elements or objects. Terms such as "connected" or "linked" are not limited to physical or mechanical connections, but can include electrical connections, whether direct or indirect. Terms such as "upper," "lower," "left," and "right" are used only to indicate relative positional relationships; when the absolute position of the described object changes, the relative positional relationship may also change accordingly.

[0020] Example 1 In this embodiment, as Figure 1 As shown, a method for predicting the situation of electric vehicle transportation accidents based on a time-series evolution model includes the following steps: S1. Collect battery temperature, battery vibration data, battery voltage, transport vehicle speed, and transport vehicle location during the historical transportation process of electric vehicles to obtain a historical dynamic risk sequence.

[0021] The method for obtaining the historical dynamic risk sequence includes: collecting battery temperature, battery vibration data, battery voltage, transport vehicle speed, and transport vehicle location during the historical transportation process of electric vehicles to generate a raw dataset containing time series data; if the battery temperature exceeds a first preset threshold, outliers in the time series are extracted using a sliding window algorithm to obtain a battery temperature anomaly sequence; based on the battery temperature anomaly sequence, combined with battery vibration data and battery voltage, the data is classified using a K-means clustering algorithm to determine the battery state risk level; if the battery state risk level is high, the mileage is calculated using the transport vehicle speed and transport vehicle location to obtain a mileage sequence related to the battery state; based on the mileage sequence and ambient temperature, a linear regression algorithm is used to predict the future trend of battery state changes to generate a dynamic risk sequence; and by using the dynamic risk sequence and the collection time, the fluctuation period of the risk level is determined, the data frequency is adjusted, and an optimized historical dynamic risk sequence is generated.

[0022] S2. Use time-series feature extraction methods to process parameter changes at consecutive time points in the historical dynamic risk sequence and determine the historical risk evolution vector.

[0023] The method for determining the historical risk evolution vector includes: extracting statistical features of historical dynamic risk sequences using a sliding window method to obtain a time-series feature set; determining the risk change trend based on the time-series feature set; if the fluctuation range of the risk change trend exceeds a second preset threshold, using principal component analysis to reduce the dimensionality of the time-series feature set to obtain a dimensionality-reduced feature set; classifying the risk change trend using a clustering method based on the dimensionality-reduced feature set to determine the risk category, and combining the risk category with the dimensionality-reduced feature set using a vector mapping method to generate an initial risk evolution vector; standardizing the initial risk evolution vector to obtain a standardized vector, and fusing the standardized vector and the time series using a weighted average method to obtain the historical risk evolution vector.

[0024] In this embodiment, a sliding window method is first used to process the historical dynamic risk sequence to extract parameter change features at continuous time points. The size of the sliding window is set to a fixed value. For each window, multiple statistical features are calculated, including mean, maximum, minimum, standard deviation, and slope, thus forming a time-series feature set. Subsequently, the fluctuation amplitude of the risk change trend is analyzed, and volatility is assessed by calculating the standard deviation or range of the features. If the fluctuation amplitude exceeds a second preset threshold (set to 0.5 in this embodiment), it indicates that the data has significant noise or redundancy, and dimensionality reduction processing is required to optimize the feature expression. Next, principal component analysis (PCA) is used to reduce the dimensionality of the time-series feature set. PCA calculates the covariance matrix of the features, performs eigenvalue decomposition, and selects principal components to retain most of the variance, thereby transforming high-dimensional features into a low-dimensional feature set, reducing data redundancy and highlighting key risk information. Based on the dimensionality-reduced feature set, the K-means clustering algorithm is applied to classify the risk change trend, setting the number of clusters to correspond to low, medium, and high risk levels, and measuring the similarity of data points using Euclidean distance. Subsequently, a vector mapping method is used to combine the risk category labels obtained from clustering with dimensionality-reduced feature values ​​to generate an initial risk evolution vector. This vector numerically encodes the category and characteristic attributes of the risk state. Finally, the initial risk evolution vector is standardized using the Z-score method to transform each feature value into a distribution with a mean of 0 and a standard deviation of 1, ensuring the comparability of features at different scales. Next, a weighted average method is used to fuse the standardized vector and time series data, with weights allocated according to a time decay factor to enhance time dependence and smooth series fluctuations, generating a historical risk evolution vector.

[0025] S3. Construct a long short-term memory network, and train the long short-term memory network using historical risk evolution vectors and historical accident scenarios to obtain an accident evolution model.

[0026] In this embodiment, a Long Short-Term Memory (LSTM) network is constructed as the core architecture of the accident evolution model to handle the temporal dependencies of the historical risk evolution vector. The LSTM network includes: an input layer with the same dimension as the number of features in the historical risk evolution vector; a hidden layer containing 128 LSTM units, using the tanh activation function to handle nonlinear relationships; and an output layer employing the softmax activation function, with the number of nodes corresponding to the possible accident categories to output a probability distribution. Furthermore, this embodiment adds a dropout layer (with a dropout rate of 0.2) after the hidden layers to reduce the risk of overfitting and ensure the model's generalization ability.

[0027] In the training data preparation phase, historical risk evolution vectors are used as input feature sequences and aligned with historical accident information. Historical accident information, including accident type and occurrence time, is extracted from records and transformed into one-hot encoded vectors as the target for supervised learning. The input sequence is generated using a sliding window with a window size of 10 time steps to form continuous sequence samples. Data preprocessing includes Z-score normalization to ensure consistent input scale, and the dataset is split into training, validation, and test sets in a 7:2:1 ratio to evaluate model performance and prevent data leakage.

[0028] The training process employs a supervised learning framework, using classification cross-entropy as the loss function to quantify the difference between predicted probabilities and true labels. The optimizer is Adam, with an initial learning rate of 0.001, and a learning rate decay strategy (halving every 20 epochs) is introduced to accelerate convergence. The batch size is set to 32, the number of training epochs is fixed at 100, and an early stopping mechanism is implemented: training is terminated early if the validation set loss does not decrease for 10 consecutive epochs. During training, LSTM weights are updated using backpropagation through-time (BPTT) algorithm, gradients are calculated, and parameters are adjusted to minimize the loss function. Furthermore, we monitor the accuracy and F1 score on both the training and validation sets, performing hyperparameter tuning (such as adjusting the number of hidden layer units) to ensure the model stably learns the risk evolution pattern.

[0029] Ultimately, the trained accident evolution model can output an accident probability distribution based on the input sequence, which can be used for prediction in subsequent steps. The model was evaluated on a test set, achieving an average accuracy of over 90% and an F1 score exceeding 0.85, indicating its effective ability to capture the dynamic characteristics of risk evolution.

[0030] S4. Obtain the current dynamic risk sequence, input it into the accident evolution model to simulate the evolution of risk state in future time steps, and determine potential accident trigger points.

[0031] In this embodiment, based on real-time collected electric vehicle transportation data, including battery temperature, battery vibration data, battery voltage, and vehicle speed and location, a current dynamic risk sequence is obtained using a processing method similar to that used for historical data: Battery temperature anomalies are extracted using a sliding window algorithm; if the temperature exceeds a first preset threshold, an anomaly sequence is generated. Combining battery vibration and voltage data, a K-means clustering algorithm is applied to assess the battery state risk level. If the risk level is high, the driving mileage is calculated, and the battery state change trend is predicted using linear regression based on ambient temperature, ultimately generating an optimized current dynamic risk sequence. This sequence has adjusted data frequency to match the risk fluctuation cycle, ensuring consistency with the input requirements of the accident evolution model.

[0032] Next, the current dynamic risk sequence is input into the pre-trained accident evolution model. The input layer dimension of the LSTM model is consistent with the number of features in the historical risk evolution vector. The hidden layer contains 128 LSTM units, and the tanh activation function is used to handle nonlinear relationships. The output layer generates the accident probability distribution through the softmax function. The input sequence is generated using a sliding window with a window size of 10 time steps to form continuous sequence samples. The model calculates the output probability for each time step through forward propagation. Then, the evolution of the risk state in future time steps is simulated. Multi-step prediction is performed using the recursive properties of the LSTM model: based on the current input sequence, the model outputs the accident probability distribution for the next time step; subsequently, this output is integrated into the input sequence (through state propagation or sequence expansion), and subsequent time steps are recursively predicted. The above process is repeated to generate the probability distribution of future risk states.

[0033] Finally, based on the probability distribution of future risk states obtained from the simulation, it is checked whether the probability of an accident at each time step exceeds a third preset threshold (e.g., set to 0.75). If the probability of any accident category at a certain time step exceeds this threshold, that time point is marked as a potential accident trigger point; at the same time, the probability trend is analyzed, such as a continuous rise or the appearance of a peak, to enhance the accuracy of the judgment.

[0034] S5. If the potential accident trigger point exceeds the third preset threshold, the multidimensional features in the risk state evolution are processed by type differentiation to obtain the accident category probability distribution, and the specific accident type is determined based on the accident category probability distribution.

[0035] The method for obtaining the accident type includes: if the trigger point exceeds a third preset threshold, obtaining the risk evolution vector, extracting multi-dimensional feature data, and using principal component analysis to reduce the dimensionality of the multi-dimensional feature data to obtain a dimensionality-reduced feature set; based on the dimensionality-reduced feature set, using the support vector machine algorithm to generate an initial accident category probability distribution; if the maximum probability value in the initial accident category probability distribution is lower than a preset classification threshold, then using the random forest algorithm to perform secondary classification on the dimensionality-reduced feature set to obtain the accident category probability distribution; if the accident category probability distribution meets the preset threshold, then using the trained classification model to determine the accident type.

[0036] In this embodiment, when a potential accident trigger point exceeds a third preset threshold (e.g., set to 0.75), the system initiates a refined accident type identification process. First, a historical risk evolution vector is obtained from the risk state evolution. This vector contains multi-dimensional feature data, such as abnormal battery temperature sequences, battery vibration data, battery voltage fluctuations, and transport vehicle speed and location information. This feature data is extracted using a sliding window to form a high-dimensional time-series dataset. To reduce data redundancy and improve computational efficiency, Principal Component Analysis (PCA) is used to reduce the dimensionality of the multi-dimensional feature data: the covariance matrix of the features is calculated, eigenvalue decomposition is performed, and the top k principal components are selected (e.g., retaining 95% of the variance), transforming the original features into a low-dimensional reduced feature set.

[0037] Next, based on the dimensionality-reduced feature set, the Support Vector Machine (SVM) algorithm is applied to generate an initial accident category probability distribution. SVM uses a radial basis function (RBF) as the kernel function to handle nonlinear classification problems by maximizing the classification margin. During training, a one-to-one strategy is used to construct multiple binary classifiers, outputting the probability value for each accident category (e.g., battery overheating, short circuit, or mechanical failure). If the maximum probability value in the initial accident category probability distribution is lower than a preset classification threshold (set to 0.7 in this embodiment), it indicates high classification uncertainty, requiring secondary classification to improve accuracy. At this point, the system switches to the Random Forest (RF) algorithm, which integrates multiple decision trees to vote on the dimensionality-reduced feature set: each tree uses bootstrapping to generate a training subset and selects the optimal split point using Gini impurity. Finally, the outputs of all trees are aggregated to obtain a more robust accident category probability distribution.

[0038] Finally, the system checks whether the probability distribution of accident categories after secondary classification meets a preset threshold, such as the probability of a certain category exceeding 0.8. If it does, the pre-trained classification model is used for the final judgment to determine the specific accident type.

[0039] S6. Generate a prediction accuracy assessment report based on the determined accident type and output trend intervention signals to obtain the final accident situation prediction results.

[0040] In this embodiment, a prediction accuracy evaluation report is first generated based on the determined accident type, including accuracy, precision, recall, and F1 score. Specifically, a confusion matrix analysis is used to assess the classification performance of each accident category (e.g., battery overheating, short circuit), and the classification report function from the scikit-learn library is applied to calculate the macro-average F1 score (e.g., exceeding 0.85 in this embodiment) to evaluate overall performance. Simultaneously, the report integrates the loss curve and ROC curve from the validation set, monitors the model's generalization ability through the AUC value (typically greater than 0.9), and uses time-series cross-validation to check prediction bias and ensure reliability. The report is output in a visual format, including indicator tables and trend graphs, facilitating user understanding of the model's performance on dynamic risk sequences. Subsequently, the system outputs a trend intervention signal and generates the final accident situation prediction result.

[0041] It should be noted that the method of this embodiment can be executed by a single device, such as a computer or server. The method of this embodiment can also be applied to a distributed scenario, where multiple devices cooperate to complete the task. In such a distributed scenario, one of these devices may execute only one or more steps of the method of this embodiment, and the multiple devices will interact with each other to complete the method.

[0042] It should be noted that the above description describes some embodiments of this disclosure. Other embodiments are within the scope of the appended claims. In some cases, it should be understood that the sequence number of each step in the above embodiments does not imply the order of execution; the execution order of each process should be determined by its function and internal logic, and should not constitute any limitation on the implementation process of the embodiments of the present invention. The actions or steps recorded in the claims can be performed in a different order than that in the above embodiments and can still achieve the desired result. In addition, the processes depicted in the drawings do not necessarily require a specific or sequential order to achieve the desired result. In some embodiments, multitasking and parallel processing are also possible or may be advantageous.

[0043] Example 2 In this embodiment, an electric vehicle transportation accident situation prediction system based on a time-series evolution model includes: a data acquisition module, a feature extraction module, a model construction module, an accident simulation module, a type determination module, and a result prediction module.

[0044] The data acquisition module is used to collect data on battery temperature, battery vibration, battery voltage, transport vehicle speed, and transport vehicle location during the historical transportation process of electric vehicles, thereby obtaining a historical dynamic risk sequence.

[0045] The workflow of the data acquisition module includes: collecting battery temperature, battery vibration data, battery voltage, transport vehicle speed, and transport vehicle location during the historical transportation of electric vehicles, generating a raw dataset containing time series data; if the battery temperature exceeds a first preset threshold, outliers in the time series are extracted using a sliding window algorithm to obtain a battery temperature anomaly sequence; based on the battery temperature anomaly sequence, combined with battery vibration data and battery voltage, the data is classified using a K-means clustering algorithm to determine the battery status risk level; if the battery status risk level is high, the mileage is calculated using the transport vehicle speed and location to obtain a mileage sequence related to the battery status; based on the mileage sequence and ambient temperature, a linear regression algorithm is used to predict the future trend of battery status changes, generating a dynamic risk sequence; through the dynamic risk sequence and the collection time, the fluctuation period of the risk level is determined, the data frequency is adjusted, and an optimized historical dynamic risk sequence is generated.

[0046] The feature extraction module uses a time-series feature extraction method to process parameter changes at consecutive time points in the historical dynamic risk sequence and determine the historical risk evolution vector.

[0047] The workflow of the feature extraction module includes: extracting statistical features of historical dynamic risk sequences using a sliding window method to obtain a time-series feature set; determining risk change trends based on the time-series feature set; if the fluctuation range of the risk change trend exceeds a second preset threshold, using principal component analysis to reduce the dimensionality of the time-series feature set to obtain a dimensionality-reduced feature set; classifying risk change trends using a clustering method based on the dimensionality-reduced feature set to determine risk categories, and combining the risk categories with the dimensionality-reduced feature set using a vector mapping method to generate an initial risk evolution vector; standardizing the initial risk evolution vector to obtain a standardized vector, and fusing the standardized vector and the time series using a weighted average method to obtain a historical risk evolution vector.

[0048] The model building module is used to construct a long short-term memory network. The long short-term memory network is trained using historical risk evolution vectors and historical accident scenarios to obtain an accident evolution model.

[0049] The accident simulation module is used to obtain the current dynamic risk sequence, input it into the accident evolution model to simulate the evolution of risk states in future time steps, and determine potential accident trigger points.

[0050] In the type determination module, if the potential accident trigger point exceeds the third preset threshold, the multidimensional features in the risk state evolution are processed by type differentiation to obtain the accident category probability distribution, and the specific accident type is determined based on the accident category probability distribution.

[0051] The workflow of the type determination module includes: if the trigger point exceeds the third preset threshold, the risk evolution vector is obtained, multi-dimensional feature data is extracted, and principal component analysis is used to reduce the dimensionality of the multi-dimensional feature data to obtain a dimensionality-reduced feature set; based on the dimensionality-reduced feature set, a support vector machine algorithm is used to generate an initial accident category probability distribution; if the maximum probability value in the initial accident category probability distribution is lower than the preset classification threshold, the dimensionality-reduced feature set is reclassified using a random forest algorithm to obtain the accident category probability distribution; if the accident category probability distribution meets the preset threshold, the accident type is determined using the trained classification model.

[0052] The results prediction module generates a prediction accuracy assessment report based on the determined accident type and outputs a trend intervention signal to obtain the final accident situation prediction result.

[0053] The system described in the above embodiments is used to implement the electric vehicle transportation accident situation prediction method based on the time-series evolution model in any of the foregoing embodiments, and has the beneficial effects of the corresponding method embodiments, which will not be repeated here.

[0054] It should be noted that the electric vehicle transportation accident situation prediction system based on the time-series evolution model described above is embodied in the form of functional units. The term "module" here can be implemented in software and / or hardware, without specific limitations.

[0055] For example, a "module" can be a software program, hardware circuit, or a combination of both that implements the above functions. Hardware circuits may include application-specific integrated circuits (ASICs), electronic circuits, processors (e.g., shared processors, proprietary processors, or group processors) and memory for executing one or more software or firmware programs, combined logic circuits, and / or other suitable components that support the described functions.

[0056] This disclosure is intended to cover all such substitutions, modifications, and variations that fall within the broad scope of the appended claims. Therefore, any omissions, modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of this disclosure should be included within the scope of protection of this disclosure.

Claims

1. A method for predicting the situation of electric vehicle transportation accidents based on a time-series evolution model, characterized in that, Includes the following steps: Collect battery temperature, battery vibration data, battery voltage, transport vehicle speed, and transport vehicle location during the historical transportation process of electric vehicles to obtain a historical dynamic risk sequence; The parameter changes at consecutive time points in the historical dynamic risk sequence are processed using a time-series feature extraction method to determine the historical risk evolution vector; A long short-term memory network is constructed, and the network is trained using the historical risk evolution vector and historical accident data to obtain an accident evolution model. Obtain the current dynamic risk sequence and input it into the accident evolution model to simulate the evolution of risk states in future time steps and determine potential accident trigger points; If the potential accident trigger point exceeds the third preset threshold, the multidimensional features in the risk state evolution are processed by type differentiation to obtain the accident category probability distribution, and the specific accident type is determined based on the accident category probability distribution. Based on the determined accident type, a prediction accuracy assessment report is generated and a trend intervention signal is output to obtain the final accident situation prediction result.

2. The method for predicting the situation of electric vehicle transportation accidents based on a time-series evolution model according to claim 1, characterized in that, The methods for obtaining the historical dynamic risk sequence include: Collect battery temperature, battery vibration data, battery voltage, transport vehicle speed, and transport vehicle location data during the historical transportation process of electric vehicles to generate a raw dataset containing time series data. If the battery temperature exceeds the first preset threshold, anomalies in the time series are extracted using a sliding window algorithm to obtain a battery temperature anomaly sequence. Based on the battery temperature anomaly sequence, combined with the battery vibration data and the battery voltage, the K-means clustering algorithm is used to classify the data and determine the battery status risk level. If the battery status risk level is high, the mileage is calculated based on the speed and location of the transport vehicle to obtain a mileage sequence related to the battery status. Based on the driving mileage sequence and ambient temperature, a linear regression algorithm is used to predict the future trend of battery status and generate a dynamic risk sequence. By using the dynamic risk sequence and the collection time, the fluctuation cycle of the risk level is determined, the data frequency is adjusted, and the optimized historical dynamic risk sequence is generated.

3. The method for predicting the situation of electric vehicle transportation accidents based on a time-series evolution model according to claim 1, characterized in that, The methods for determining the historical risk evolution vector include: The statistical features of the historical dynamic risk sequence are extracted using the sliding window method to obtain a time-series feature set; Based on the time series feature set, the risk change trend is determined. If the fluctuation range of the risk change trend exceeds the second preset threshold, the principal component analysis method is used to reduce the dimensionality of the time series feature set to obtain a dimensionality-reduced feature set. Based on the dimensionality reduction feature set, a clustering method is used to classify the risk change trend, determine the risk category, and then a vector mapping method is used to combine the risk category with the dimensionality reduction feature set to generate an initial risk evolution vector. The initial risk evolution vector is standardized to obtain a standardized vector, and the standardized vector and the time series are fused using a weighted average method to obtain the historical risk evolution vector.

4. The method for predicting the situation of electric vehicle transportation accidents based on a time-series evolution model according to claim 1, characterized in that, The methods for obtaining the accident type include: If the trigger point exceeds the third preset threshold, the risk evolution vector is obtained, multi-dimensional feature data is extracted, and the multi-dimensional feature data is dimensionality reduced by the principal component analysis algorithm to obtain a dimensionality-reduced feature set. Based on the dimensionality reduction feature set, an initial accident category probability distribution is generated using the support vector machine algorithm. If the maximum probability value in the initial accident category probability distribution is lower than a preset classification threshold, the dimensionality reduction feature set is then classified a second time using the random forest algorithm to obtain the accident category probability distribution. If the probability distribution of the accident category meets the preset threshold, the accident type is determined by the trained classification model.

5. A predictive system for electric vehicle transportation accidents based on a time-series evolution model, wherein the system applies the method described in any one of claims 1-4, characterized in that, include: The system includes a data acquisition module, a feature extraction module, a model building module, an accident simulation module, a type determination module, and a result prediction module. The data acquisition module is used to collect battery temperature, battery vibration data, battery voltage, transport vehicle speed and transport vehicle location during the historical transportation process of electric vehicles, and obtain historical dynamic risk sequences. The feature extraction module uses a temporal feature extraction method to process parameter changes at consecutive time points in the historical dynamic risk sequence to determine the historical risk evolution vector; The model building module is used to construct a long short-term memory network, and the long short-term memory network is trained using the historical risk evolution vector and historical accident information to obtain an accident evolution model; The accident simulation module is used to obtain the current dynamic risk sequence, input it into the accident evolution model to simulate the evolution of risk state in future time steps, and determine potential accident trigger points; In the type determination module, if the potential accident trigger point exceeds the third preset threshold, the multidimensional features in the risk state evolution are processed by type differentiation to obtain the accident category probability distribution, and the specific accident type is determined based on the accident category probability distribution. The result prediction module generates a prediction accuracy assessment report and outputs a trend intervention signal based on the determined accident type to obtain the final accident situation prediction result.

6. The electric vehicle transportation accident situation prediction system based on a time-series evolution model according to claim 5, characterized in that, The workflow of the data acquisition module includes: Collect battery temperature, battery vibration data, battery voltage, transport vehicle speed, and transport vehicle location data during the historical transportation process of electric vehicles to generate a raw dataset containing time series data. If the battery temperature exceeds the first preset threshold, anomalies in the time series are extracted using a sliding window algorithm to obtain a battery temperature anomaly sequence. Based on the battery temperature anomaly sequence, combined with the battery vibration data and the battery voltage, the K-means clustering algorithm is used to classify the data and determine the battery status risk level. If the battery status risk level is high, the mileage is calculated based on the speed and location of the transport vehicle to obtain a mileage sequence related to the battery status. Based on the driving mileage sequence and ambient temperature, a linear regression algorithm is used to predict the future trend of battery status and generate a dynamic risk sequence. By using the dynamic risk sequence and the collection time, the fluctuation cycle of the risk level is determined, the data frequency is adjusted, and the optimized historical dynamic risk sequence is generated.

7. The electric vehicle transportation accident situation prediction system based on a time-series evolution model according to claim 5, characterized in that, The workflow of the feature extraction module includes: The statistical features of the historical dynamic risk sequence are extracted using the sliding window method to obtain a time-series feature set; Based on the time series feature set, the risk change trend is determined. If the fluctuation range of the risk change trend exceeds the second preset threshold, the principal component analysis method is used to reduce the dimensionality of the time series feature set to obtain a dimensionality-reduced feature set. Based on the dimensionality reduction feature set, a clustering method is used to classify the risk change trend, determine the risk category, and then a vector mapping method is used to combine the risk category with the dimensionality reduction feature set to generate an initial risk evolution vector. The initial risk evolution vector is standardized to obtain a standardized vector, and the standardized vector and the time series are fused using a weighted average method to obtain the historical risk evolution vector.

8. The electric vehicle transportation accident situation prediction system based on a time-series evolution model according to claim 5, characterized in that, The workflow of the type determination module includes: If the trigger point exceeds the third preset threshold, the risk evolution vector is obtained, multi-dimensional feature data is extracted, and the multi-dimensional feature data is dimensionality reduced by the principal component analysis algorithm to obtain a dimensionality-reduced feature set. Based on the dimensionality reduction feature set, an initial accident category probability distribution is generated using the support vector machine algorithm. If the maximum probability value in the initial accident category probability distribution is lower than a preset classification threshold, the dimensionality reduction feature set is then classified a second time using the random forest algorithm to obtain the accident category probability distribution. If the probability distribution of the accident category meets the preset threshold, the accident type is determined by the trained classification model.

Images