A driving style recognition method and system based on an entropy weight method

Abstract

The application belongs to the technical field of artificial intelligence, and provides a driving style recognition method and system based on an entropy weight method, wherein the driving style recognition method based on the entropy weight method comprises: processing original driving data of a target vehicle and an artificial neural network model to obtain a predicted driving period label of each time segment and a predicted driving style label of each time segment; storing the predicted driving period label of each time segment and the predicted driving style label of each time segment; calculating based on a time number and the predicted driving style label to obtain an average style score; calculating based on the predicted driving period label and the predicted driving style label to obtain an entropy weight value; and performing weighted calculation based on the average style score and the entropy weight value to obtain a target driving style score, and determining a target driving style based on the target driving style score and a preset score, thereby significantly improving the objectivity and reproducibility of the score.

Description

Technical Field

[0001] This invention relates to the field of artificial intelligence technology, and in particular to a driving style recognition method and system based on entropy weighting. Background Technology

[0002] In recent years, as cars have become increasingly intelligent, users have placed higher demands on the adaptive intelligent control of vehicles, leading to the emergence of driving style recognition technology. Driving style recognition technology mainly involves determining data features based on collected data and training a model based on machine learning or deep learning algorithms and these data features to identify the driver's driving style as accurately as possible.

[0003] Currently, driving style recognition technology often uses the Analytic Hierarchy Process (AHP) to assign importance to driving cycles. However, AHP requires domain experts to set up pairwise comparison matrices based on their personal experience or cognition. The judgment criteria provided by different experts vary, leading to highly subjective weights that are difficult to maintain consistently across different environments. Once the AHP judgment matrix is ​​set, its weights remain fixed and cannot be dynamically adjusted according to different driving environments or changes in driver behavior, making it difficult to adapt to the diversity and complexity of actual driving processes. Furthermore, when there are many types of cycles or complex judgment criteria, manually constructing the judgment matrix by experts can easily lead to logical contradictions, requiring consistency checks and increasing design complexity and the probability of errors.

[0004] Therefore, how to improve the objectivity and consistency of scoring results in driving style recognition technology is a problem that urgently needs to be solved by those skilled in the art. Summary of the Invention

[0005] In view of this, embodiments of the present invention provide a driving style recognition method and system based on the entropy weight method to solve the problem of how to improve the objectivity and consistency of the scoring results in the prior art; that is, embodiments of the present invention can significantly improve the objectivity and reproducibility of the scoring.

[0006] According to one aspect of the present invention, a driving style recognition method based on entropy weighting is provided. The method includes: processing raw driving data of a target vehicle and an artificial neural network model to obtain a predicted driving cycle label and a predicted driving style label for each time segment; storing the predicted driving cycle label and the predicted driving style label for each time segment, and recording the time number corresponding to each time segment; counting the number of stored predicted driving cycle labels or predicted driving style labels for each time segment, and determining whether the stored number is equal to or greater than a preset storage number; when the stored number is equal to or greater than the preset storage number, calculating an average style score based on the time number and the predicted driving style label; when the stored number is equal to or greater than the preset storage number, calculating an entropy weight based on the predicted driving cycle label and the predicted driving style label; performing a weighted calculation based on the average style score and the entropy weight to obtain a target driving style score, and determining the target driving style based on the target driving style score and the preset score.

[0007] In one embodiment, the processing of raw driving data of the target vehicle and an artificial neural network model to obtain a predicted driving cycle label and a predicted driving style label for each time segment includes: real-time acquisition of raw driving data of the target vehicle, and cleaning the raw driving data to obtain target driving data, wherein the raw driving data includes training raw driving data and non-training raw driving data; dividing the target driving data into time segments based on a preset driving duration to obtain target driving data for at least one time segment; and performing feature extraction based on the target driving data for each time segment to obtain driving cycle features and driving style for the corresponding time segment. Features; dimensionality reduction processing is performed on the driving cycle features and driving style features to obtain driving principal component features, wherein the driving principal component features include trained driving principal component features and untrained driving principal component features; an artificial neural network model is trained based on the trained driving principal component features to obtain a target driving cycle artificial neural network model and a target driving style artificial neural network model; the target driving cycle artificial neural network model is called, and recognition processing is performed based on the untrained driving principal component features to obtain the predicted driving cycle label, and the target driving style artificial neural network model is called, and recognition processing is performed based on the untrained driving principal component features to obtain the predicted driving style label.

[0008] In one embodiment, training the artificial neural network model based on the training driving principal component features to obtain a target driving cycle artificial neural network model and a target driving style artificial neural network model includes: performing clustering processing based on the training driving principal component features to obtain corresponding real driving cycle labels and corresponding real driving style labels; and training the artificial neural network model based on the training driving principal component features, the real driving cycle labels, and the real driving style labels to obtain a target driving cycle artificial neural network model and a target driving style artificial neural network model.

[0009] In one embodiment, training the artificial neural network model based on the training driving principal component features, the real driving cycle label, and the real driving style label to obtain a target driving cycle artificial neural network model and a target driving style artificial neural network model includes: obtaining a prediction probability based on the training driving principal component features; calculating a target loss value based on the prediction probability, the real driving cycle label, and the real driving style label; and optimizing the model parameters in the artificial neural network model in the direction of reducing the target loss value to obtain the target driving cycle artificial neural network model and the target driving style artificial neural network model.

[0010] In one embodiment, when the storage quantity is equal to or greater than the preset storage quantity, the average style score is calculated based on the time number and the predicted driving style label, including: calculating a time decay weight based on the time number; and performing a weighted calculation based on the predicted driving style label and the time decay weight to obtain the average style score.

[0011] In one embodiment, the step of weighting the predicted driving style label and the time decay weight to obtain the average style score includes: calculating the frequency of occurrence of each type of driving style label based on the predicted driving cycle label and the predicted driving style label; calculating the entropy value based on the frequency of occurrence of each type of driving style label; and calculating the entropy weight based on the entropy value.

[0012] In one embodiment, the target driving style score is obtained by weighting the average style score and the entropy weight, and the target driving style score is:

[0013]

[0014] Among them, w i Here, avg_score represents the entropy weight, i represents the type of the predicted driving cycle label, and avg_score represents the driving cycle label. i(t) represents the average style score.

[0015] According to another aspect of the present invention, a driving style recognition system based on the entropy weight method is provided. The driving style recognition system based on the entropy weight method includes: a data processing module, a data storage module, a comparison module, a first calculation module, a second calculation module, and a style determination module. The data processing module processes raw driving data of a target vehicle using an artificial neural network model to obtain a predicted driving cycle label and a predicted driving style label for each time segment. The data storage module stores the predicted driving cycle label and the predicted driving style label for each time segment and records the time number corresponding to each time segment. The comparison module compares the stored data for each time segment. The number of predicted driving cycle tags for each time segment or the number of predicted driving style tags for each time segment is counted, and it is determined whether the number of stored tags is equal to or greater than a preset number of stored tags. When the number of stored tags is equal to or greater than the preset number of stored tags, the first calculation module is used to calculate an average style score based on the time number and the predicted driving style tags. When the number of stored tags is equal to or greater than the preset number of stored tags, the second calculation module is used to calculate an entropy weight based on the predicted driving cycle tags and the predicted driving style tags. The style determination module is used to perform a weighted calculation based on the average style score and the entropy weight to obtain a target driving style score, and to determine the target driving style based on the target driving style score and the preset score.

[0016] In one embodiment, the data processing module includes a data acquisition module, a time segmentation module, a feature extraction module, a feature processing module, a model training module, and a label generation module. The data acquisition module acquires raw driving data of the target vehicle in real time and performs cleaning processing on the raw driving data to obtain target driving data. The raw driving data includes training raw driving data and non-training raw driving data. The time segmentation module divides the target driving data into time segments based on a preset driving duration to obtain target driving data for at least one time segment. The feature extraction module performs feature extraction on the target driving data for each time segment to obtain driving cycle features and driving style features for the corresponding time segment. The feature processing module is used to perform dimensionality reduction processing based on the driving cycle features and the driving style features to obtain driving principal component features, wherein the driving principal component features include training driving principal component features and non-training driving principal component features; the model training module is used to train the artificial neural network model based on the training driving principal component features to obtain the target driving cycle artificial neural network model and the target driving style artificial neural network model respectively; the label generation module is used to call the target driving cycle artificial neural network model, perform recognition processing based on the non-training driving principal component features to obtain the predicted driving cycle label, and call the target driving style artificial neural network model, perform recognition processing based on the non-training driving principal component features to obtain the predicted driving style label.

[0017] In one embodiment, the first calculation module includes a first calculation module and a second calculation module, wherein the first calculation module is used to calculate based on the time number to obtain a time decay weight; and the second calculation module is used to perform a weighted calculation based on the predicted driving style label and the time decay weight to obtain the average style score.

[0018] In summary, in this embodiment of the invention, based on the original driving data of the target vehicle and an artificial neural network model, a predicted driving cycle label and a predicted driving style label for each time segment are obtained. These labels are then stored, and the corresponding time number for each time segment is recorded. The number of stored predicted driving cycle labels or predicted driving style labels for each time segment is counted, and it is determined whether the stored number is equal to or greater than a preset storage number. When the stored number is equal to or greater than a preset storage number... When the number of stored items exceeds the preset storage quantity, an average style score is calculated based on the time number and the predicted driving style label. When the number of stored items is equal to or greater than the preset storage quantity, an entropy weight is calculated based on the predicted driving cycle label and the predicted driving style label. The target driving style score is obtained by weighted calculation based on the average style score and the entropy weight. The target driving style is determined based on the target driving style score and the preset score. This not only significantly improves the objectivity and reproducibility of the score, but also flexibly adapts to the actual characteristics of different drivers, different road conditions or time periods, and has strong environmental adaptability and generalization ability. Attached Figure Description

[0019] Further details, features, and advantages of the invention are disclosed in the following description of exemplary embodiments in conjunction with the accompanying drawings, in which:

[0020] Figure 1 A flowchart illustrating a driving style recognition method based on entropy weight method disclosed in an embodiment of this application is shown.

[0021] Figure 2 It shows Figure 1 The flowchart of step S110 shown is a schematic diagram.

[0022] Figure 3 It shows Figure 2 The flowchart of step S115 is shown below;

[0023] Figure 4 It shows Figure 3 The flowchart of step S1152 shown is a schematic diagram.

[0024] Figure 5 It shows Figure 1 The flowchart of step S140 shown is a schematic diagram.

[0025] Figure 6 It shows Figure 1 The flowchart of step S150 shown is a schematic diagram.

[0026] Figure 7A schematic diagram of the structure of a driving style recognition system based on the entropy weight method disclosed in an embodiment of this application is shown. Detailed Implementation

[0027] Embodiments of the present invention will now be described in more detail with reference to the accompanying drawings. While some embodiments of the invention are shown in the drawings, it should be understood that the invention can be implemented in various forms and should not be construed as limited to the embodiments set forth herein. Rather, these embodiments are provided to provide a more thorough and complete understanding of the invention. It should be understood that the accompanying drawings and embodiments are for illustrative purposes only and are not intended to limit the scope of protection of the invention.

[0028] It should be understood that the various steps described in the method embodiments of the present invention may be performed in different orders and / or in parallel. Furthermore, the method embodiments may include additional steps and / or omit the steps shown. The scope of the present invention is not limited in this respect.

[0029] The term "comprising" and its variations as used herein are open-ended, meaning "including but not limited to". The term "based on" means "at least partially based on". The term "one embodiment" means "at least one embodiment"; the term "another embodiment" means "at least one additional embodiment"; the term "some embodiments" means "at least some embodiments". Definitions of other terms will be given in the following description. It should be noted that the concepts of "first", "second", etc., mentioned in this invention are used only to distinguish different devices, modules, or units, and are not intended to limit the order of functions performed by these devices, modules, or units or their interdependencies.

[0030] It should be noted that the terms "a" and "a plurality of" used in this invention are illustrative rather than restrictive. Those skilled in the art should understand that, unless otherwise expressly indicated in the context, they should be understood as "one or more".

[0031] The names of the messages or information exchanged between the multiple devices in the embodiments of the present invention are for illustrative purposes only and are not intended to limit the scope of these messages or information.

[0032] It should be noted that the execution subject of the driving style recognition method based on entropy weight method provided in this embodiment of the invention can be one or more electronic devices, and this invention does not limit this; wherein, the electronic device can be a terminal (i.e., a client) or a server. Therefore, when the execution subject includes multiple electronic devices, and among the multiple electronic devices includes at least one terminal and at least one server, the driving style recognition method based on entropy weight method provided in this embodiment of the invention can be jointly executed by the terminal and the server. Accordingly, the terminal mentioned herein can include, but is not limited to: smartphones, tablets, laptops, desktop computers, smartwatches, smart voice interaction devices, smart home appliances, vehicle terminals, aircraft, etc. The server mentioned herein can be an independent physical server, a server cluster or distributed system composed of multiple physical servers, or a cloud server providing basic cloud computing services such as cloud services, cloud databases, cloud computing, cloud functions, cloud storage, network services, cloud communication, middleware services, domain name services, security services, CDN (Content Delivery Network), and big data and artificial intelligence platforms, etc.

[0033] Based on the above description, this embodiment of the invention proposes a driving style recognition method based on entropy weighting. This method can be executed by the aforementioned electronic device (terminal or server); or, it can be executed jointly by the terminal and the server. For ease of explanation, the following description will use the execution of this driving style recognition method based on entropy weighting by an electronic device as an example.

[0034] Please see Figure 1 This is a flowchart illustrating a driving style recognition method based on entropy weighting, as disclosed in an embodiment of this application. This method addresses the issue of improving the objectivity and consistency of scoring results in driving style recognition technology. It should be noted that the driving style recognition method based on entropy weighting in this application is not limited to... Figure 1 The flowchart shown illustrates the steps and their order. Depending on different needs, the steps in the flowchart can be added, removed, or their order changed. In this embodiment, as... Figure 1 As shown, the process of a driving style recognition method based on entropy weight method includes at least the following steps.

[0035] S110. Based on the original driving data of the target vehicle and the artificial neural network model, the predicted driving cycle label and the predicted driving style label of each time segment are obtained.

[0036] like Figure 2As shown, in an embodiment of the present invention, Figure 2 Step S110 includes at least the following steps:

[0037] S111. Acquire the raw driving data of the target vehicle in real time, and perform cleaning processing on the raw driving data to obtain the target driving data, wherein the raw driving data includes training raw driving data and non-training raw driving data.

[0038] In this embodiment of the invention, the raw driving data of the vehicle may include at least vehicle speed data, pedal signal data, time series data, abnormal data, and fluctuation data. Specifically, vehicle speed data may include at least vehicle speed, acceleration, and jerk; pedal signal data may include at least accelerator pedal opening, accelerator pedal rate of change, brake pedal opening, and brake pedal rate of change; time series data may include at least segments with discontinuous or missing sampling times; and abnormal data may include at least acceleration with an absolute value exceeding 10 m / s². 2 Outliers and fluctuations in the data can include at least the noise from vehicle speed fluctuations and the noise from pedal data fluctuations.

[0039] The raw driving data undergoes cleaning processing to obtain the target driving data. Specifically, firstly, due to sensor errors or environmental interference, the raw driving data may contain noise, which can affect the accuracy of data analysis; secondly, during the data sampling process, there may be discontinuities in time or data loss, requiring interpolation to supplement the data; thirdly, extreme values ​​(i.e., abnormal accelerations) may appear during the data sampling process, distorting the statistical results and requiring removal or correction; fourthly, data fluctuations (i.e., sudden changes in vehicle speed) require filtering for smoothing to reflect the actual driving situation; and fifthly, the target driving data obtained through cleaning processing can more accurately represent the driving cycle and driving style, thereby ensuring the stability and accuracy of the Artificial Neural Network (ANN) model.

[0040] S112. Divide the target driving data into time segments based on a preset driving time to obtain target driving data for at least one time segment.

[0041] In this embodiment of the invention, the target driving data is divided into time segments based on a preset driving duration to obtain target driving data for at least one time segment. The preset driving duration can be 40 seconds. It is understood that the preset driving duration can be set based on actual needs, and this embodiment does not limit it in this regard.

[0042] S113. Based on the target driving data of each time segment, feature extraction is performed to obtain the driving cycle features and driving style features of the corresponding time segment.

[0043] In this embodiment of the invention, feature extraction is performed on the target driving data for each time segment to obtain the driving cycle features and driving style features for the corresponding time segment. The driving cycle features can be extracted based on vehicle speed, acceleration, and time distribution in the target driving data, while the driving style features can be extracted based on pedal data and accelerometer readings in the target driving data.

[0044] S114. Dimensionality reduction is performed based on the driving cycle features and the driving style features to obtain driving principal component features, wherein the driving principal component features include trained driving principal component features and untrained driving principal component features.

[0045] In this embodiment of the invention, Principal Component Analysis (PCA) is performed to reduce the dimensionality of driving cycle features and driving style features, resulting in driving principal component features. Specifically, orthogonal transformations are used to project the driving cycle features and driving style features into a low-dimensional space, thereby preserving the direction of maximum variance. Dimensionality reduction using PCA reduces computational complexity, generates uncorrelated principal components through orthogonal transformations to avoid redundant information interfering with the ANN model, makes the processed data more compact, and significantly improves the efficiency of the ANN model while maintaining model accuracy.

[0046] S115. Based on the training driving principal component features, the artificial neural network model is trained to obtain the target driving cycle artificial neural network model and the target driving style artificial neural network model, respectively.

[0047] like Figure 3 As shown, in an embodiment of the present invention, Figure 3 Step S115 includes at least the following steps:

[0048] S1151. Based on the training driving principal component features, perform clustering processing to obtain the corresponding real driving cycle label and the corresponding real driving style label.

[0049] In this embodiment of the invention, K-means clustering is performed based on the principal component features of the training driving data to obtain corresponding real driving cycle labels and corresponding real driving style labels. Specifically, driving cycle labels can be categorized into four types, and driving style labels can be categorized into three types. The four driving cycle labels can include at least 0-3, where 0 represents congestion, 1 represents low speed, 2 represents medium speed, and 3 represents high speed; the three driving style labels can include at least 0-2, where 0 represents conservative, 1 represents normal, and 2 represents aggressive.

[0050] S1152. The artificial neural network model is trained based on the training driving principal component features, the real driving cycle label, and the real driving style label to obtain the target driving cycle artificial neural network model and the target driving style artificial neural network model, respectively.

[0051] like Figure 4 As shown, in an embodiment of the present invention, Figure 4 Step S1152 includes at least the following steps:

[0052] S1152a. The predicted probability is obtained based on the training driving principal component features.

[0053] In this embodiment of the invention, calculation is performed through forward propagation, and the calculation process is as follows:

[0054] data h =ω1·data PAC +b1 formula (1)

[0055] data out =ω2·data ReLU +b2 Formula (2)

[0056] Among them, data h The output data for the first model, ω1 is the first weight coefficient, and data PAC To train the principal component features of driving, b1 is the first bias, and data out The output data for the second model, ω2 is the second weight coefficient, data ReLU b1 represents the input data processed by the activation function, and b2 represents the second bias.

[0057] The ReLU activation function is shown in the following formula:

[0058]

[0059] Where x is the input data.

[0060] Then, based on the Softmax normalization, the output model predicts the probability, which is given by the following formula:

[0061]

[0062] Among them, data out,i Output data for the second model corresponding to the i-th data point. The output data of the second model is the j-th data point, where k is 4 (i.e., the number of types of driving cycle labels) or 3 (i.e., the number of types of driving style labels).

[0063] S1152b: The target loss value is calculated based on the predicted probability, the real driving cycle label, and the real driving style label.

[0064] In this embodiment of the invention, a target loss value is obtained by calculating the predicted probability, the actual driving cycle label, and the actual driving style label using Cross-Entropy Loss (CEE). The target loss value is calculated using the following formula:

[0065]

[0066] Where, q i For real-time driving cycle labels or real-time driving style labels, prob i To predict probabilities.

[0067] S1152c. Optimize the model parameters in the artificial neural network model in the direction of reducing the target loss value to obtain the target driving cycle artificial neural network model and the target driving style artificial neural network model.

[0068] S116. Call the target driving cycle artificial neural network model, perform recognition processing based on the untrained driving principal component features to obtain the predicted driving cycle label, and call the target driving style artificial neural network model, perform recognition processing based on the untrained driving principal component features to obtain the predicted driving style label.

[0069] S120. Store the predicted driving cycle label and the predicted driving style label for each time segment, and record the time number corresponding to each time segment.

[0070] In this embodiment of the invention, the predicted driving cycle label and predicted driving style label for each time segment are stored in a sliding window, and the time number of each time segment is recorded. Specifically, taking the first time segment as an example, its time number is 1, and its corresponding predicted driving cycle label and predicted driving style label are stored.

[0071] S130. Count the number of predicted driving cycle tags or predicted driving style tags for each time segment stored, and determine whether the number of stored tags is equal to or greater than the preset number of stored tags.

[0072] In this embodiment of the invention, when the storage quantity is less than the preset storage quantity, the accumulation continues. It is understood that the preset storage quantity can be set based on actual needs, and this embodiment does not limit this setting.

[0073] S140. When the storage quantity is equal to or greater than the preset storage quantity, the average style score is calculated based on the time number and the predicted driving style label.

[0074] like Figure 5 As shown, in an embodiment of the present invention, Figure 5 Step S140 includes at least the following steps:

[0075] S141. Calculate the time decay weight based on the time number.

[0076] In this embodiment of the invention, time segments with the same predicted driving cycle label are placed into a time segment set. The calculation process for the time decay weight of each time segment is as follows:

[0077]

[0078] Where λ is the time decay coefficient, λ>0, f ik t is the time number of the k-th time segment in the time segment set, and t is the time number of the current time segment.

[0079] S142. The average style score is obtained by weighting the predicted driving style label and the time decay weight.

[0080] In this embodiment of the invention, a time decay weight is used based on each time segment. And the style score v corresponding to the predicted driving style label for each time segment ik A weighted average style score is obtained by performing a weighted calculation. The average style score is avg_score. i The calculation process for (t) is as follows:

[0081]

[0082] Among them, v ik For each time segment, predict the style score corresponding to the driving style label. The time decay weight for each time segment is defined by k, where k is the time number of the time segment.

[0083] S150. When the storage quantity is equal to or greater than the preset storage quantity, the entropy weight is calculated based on the predicted driving cycle label and the predicted driving style label.

[0084] like Figure 6 As shown, in an embodiment of the present invention, Figure 6 Step S150 includes at least the following steps:

[0085] S151. Based on the predicted driving cycle label and the predicted driving style label, calculate the frequency of occurrence of each type of driving style label.

[0086] In this embodiment of the invention, the frequency of occurrence of each type of predicted driving style label is calculated in the time segment of each type of predicted driving cycle label to obtain the frequency of occurrence of each type of driving style label.

[0087] S152. Calculate the entropy value based on the frequency of occurrence of each type of driving style label.

[0088] In this embodiment of the invention, an entropy value is calculated based on the frequency of occurrence of each type of driving style label. The entropy value reflects the dispersion or uncertainty of the predicted driving style labels within a time segment of each predicted driving cycle label. A smaller entropy value indicates a more concentrated style for that cycle, resulting in stronger discrimination and higher weight; a larger entropy value indicates a more mixed style for that cycle, resulting in weaker discrimination and lower weight. The entropy value is calculated using the following formula:

[0089]

[0090] Where i represents the type of the predicted driving cycle label, k is the normalization coefficient, and s ij Let be the frequency of the predicted driving style label of the j-th class within the time segment of the predicted driving cycle label of the i-th class.

[0091] S153. Calculate the entropy weight based on the entropy value.

[0092] In this embodiment of the invention, the entropy weight is calculated based on the entropy value. The calculation process for the entropy weight is as follows:

[0093]

[0094] Among them, 1-E i Let be the amount of effective information in the time segment of the i-th class predicting the driving cycle label.

[0095] S160. A target driving style score is obtained by weighted calculation based on the average style score and the entropy weight, and the target driving style is determined based on the target driving style score and the preset score.

[0096] In this embodiment of the invention, the target driving style score is obtained by weighted summation based on the average style score and entropy weight. The calculation process of the target driving style score is as follows:

[0097]

[0098] When the target driving style scoretotal If the score is >85, the target driving style is determined to be conservative; if the score is <75, the target driving style score is determined to be conservative. total If the score is ≤85, the target driving style is determined to be conventional; when the target driving style score is ≤85, the target driving style is determined to be conventional. total If the value is ≤75, then the target driving style is determined to be aggressive.

[0099] In summary, the driving style recognition method based on entropy weighting in this application processes the original driving data of the target vehicle and an artificial neural network model to obtain a predicted driving cycle label and a predicted driving style label for each time segment. These labels are then stored, and the corresponding time number for each time segment is recorded. The number of stored predicted driving cycle labels or predicted driving style labels for each time segment is counted, and it is determined whether the stored number is equal to or greater than a preset storage number. When the number of stored items is equal to or greater than the preset number of stored items, an average style score is calculated based on the time number and the predicted driving style label. When the number of stored items is equal to or greater than the preset number of stored items, an entropy weight is calculated based on the predicted driving cycle label and the predicted driving style label. The target driving style score is obtained by weighted calculation based on the average style score and the entropy weight. The target driving style is determined based on the target driving style score and the preset score. This not only significantly improves the objectivity and reproducibility of the score, but also flexibly adapts to the actual characteristics of different drivers, different road conditions or time periods, and has strong environmental adaptability and generalization ability.

[0100] Please see Figure 7 This is a schematic diagram of the structure of a driving style recognition system based on the entropy weight method disclosed in an embodiment of this application. In one embodiment, such as Figure 7 As shown, this application provides a driving style recognition system 100 based on the entropy weight method. The driving style recognition system 100 based on the entropy weight method may include at least: a data processing module 110, a data storage module 130, a comparison module 150, a first calculation module 170, a second calculation module 180, and a style determination module 190. Information interaction exists between the data processing module 110 and the data storage module 130; between the data storage module 130 and the comparison module 150; between the comparison module 150 and the first calculation module 170; between the first calculation module 170 and the second calculation module 180; and between the second calculation module 180 and the style determination module 190.

[0101] The data processing module 110 is used to process the raw driving data of the target vehicle and the artificial neural network model to obtain the predicted driving cycle label and the predicted driving style label for each time segment. The data processing module 110 may include at least: a data acquisition module 111, a time segment segmentation module 113, a feature extraction module 115, a feature processing module 116, a model training module 117, and a label generation module 118.

[0102] The data acquisition module 111 is used to acquire the raw driving data of the target vehicle in real time, and to clean and process the raw driving data to obtain the target driving data. The raw driving data includes training raw driving data and non-training raw driving data. Specifically, the raw driving data of the vehicle may include at least vehicle speed data, pedal signal data, time series data, abnormal data, and fluctuation data. The vehicle speed data may include at least vehicle speed, acceleration, and jerk; the pedal signal data may include at least accelerator pedal opening, accelerator pedal rate of change, brake pedal opening, and brake pedal rate of change; the time series data may include at least segments with discontinuous or missing sampling times; and the abnormal data may include at least acceleration with an absolute value exceeding 10 m / s². 2 Outliers and fluctuations in the data can include at least noise from vehicle speed fluctuations and pedal data fluctuations. The raw driving data undergoes cleaning to obtain the target driving data. Specifically, firstly, due to sensor errors or environmental interference, the raw driving data may contain noise, which can affect the accuracy of data analysis; secondly, during data sampling, there may be discontinuities in time or data loss, requiring interpolation to supplement the data; thirdly, extreme values ​​(i.e., abnormal accelerations) may occur during data sampling, distorting the statistical results and requiring removal or correction; fourthly, data fluctuations (i.e., sudden changes in vehicle speed) require filtering for smoothing to reflect the actual driving situation; fifthly, the target driving data obtained through cleaning can more accurately represent the driving cycle and driving style, thereby ensuring the stability and accuracy of the Artificial Neural Network (ANN) model.

[0103] The time segmentation module 113 is used to segment the target driving data into time segments based on a preset driving duration, obtaining at least one time segment of target driving data. Specifically, the time segmentation module 113 segments the target driving data into time segments based on a preset driving duration, obtaining at least one time segment of target driving data. The preset driving duration can be 40 seconds. It is understood that the preset driving duration can be set based on actual needs, and this embodiment does not limit it.

[0104] The feature extraction module 115 is used to extract features from the target driving data of each time segment to obtain the driving cycle features and driving style features of the corresponding time segment. Specifically, the feature extraction module 115 extracts features from the target driving data of each time segment to obtain the driving cycle features and driving style features of the corresponding time segment. The driving cycle features can be extracted based on vehicle speed, acceleration, and time distribution in the target driving data, while the driving style features can be extracted based on pedal data and accelerometer readings in the target driving data.

[0105] The feature processing module 116 is used to perform dimensionality reduction based on the driving cycle features and the driving style features to obtain driving principal component features, wherein the driving principal component features include trained driving principal component features and non-trained driving principal component features. Specifically, the feature processing module 116 performs Principal Component Analysis (PCA) dimensionality reduction based on the driving cycle features and the driving style features to obtain driving principal component features. Specifically, the driving cycle features and driving style features are projected into a low-dimensional space through orthogonal transformation, thereby preserving the direction of maximum variance. Dimensionality reduction through PCA can reduce computational complexity, generate uncorrelated principal components through orthogonal transformation to avoid redundant information interfering with the ANN model, make the processed data more compact, and significantly improve the efficiency of the ANN model while maintaining model accuracy.

[0106] The model training module 117 is used to train the artificial neural network model based on the principal component features of the training driving, to obtain the target driving cycle artificial neural network model and the target driving style artificial neural network model, respectively. The model training module 117 may include at least a clustering processing unit 1171 and a training unit 1173.

[0107] The clustering processing unit 1171 is used to perform clustering processing based on the training driving principal component features to obtain corresponding real driving cycle labels and corresponding real driving style labels. Specifically, the clustering processing unit 1171 performs K-means clustering processing based on the training driving principal component features to obtain corresponding real driving cycle labels and corresponding real driving style labels. Driving cycle labels can be divided into 4 types, and driving style labels can be divided into 3 types. Specifically, the 4 driving cycle labels can include at least 0-3, where 0 represents congestion, 1 represents low speed, 2 represents medium speed, and 3 represents high speed; the 3 driving style labels can include at least 0-2, where 0 represents conservative, 1 represents normal, and 2 represents aggressive.

[0108] Training unit 1173 is used to train the artificial neural network model based on the training driving principal component features, the real driving cycle label, and the real driving style label, to obtain the target driving cycle artificial neural network model and the target driving style artificial neural network model, respectively. Specifically, training unit 1173 obtains the prediction probability based on the training driving principal component features. In this embodiment of the invention, the calculation is performed through forward propagation, and the calculation process is as follows:

[0109] data h =ω1·data PAC +b1 formula (1)

[0110] data out =ω2·data ReLU +b2 Formula (2)

[0111] Among them, data h The output data for the first model, ω1 is the first weight coefficient, and data PAC To train the principal component features of driving, b1 is the first bias, and data out The output data for the second model, ω2 is the second weight coefficient, data ReLU b1 represents the input data processed by the activation function, and b2 represents the second bias.

[0112] The ReLU activation function is shown in the following formula:

[0113]

[0114] Where x is the input data.

[0115] Then, based on the Softmax normalization, the output model predicts the probability, which is given by the following formula:

[0116]

[0117] Among them, data out,i Output data for the second model corresponding to the i-th data point. The output data of the second model is the j-th data point, where k is 4 (i.e., the number of types of driving cycle labels) or 3 (i.e., the number of types of driving style labels).

[0118] Training unit 1173 calculates the target loss value based on the predicted probability, the real driving cycle label, and the real driving style label. In this embodiment of the invention, the target loss value is calculated by cross-entropy loss (CEE) on the predicted probability, the real driving cycle label, and the real driving style label. The target loss value is calculated using the following formula:

[0119]

[0120] Where, q i For real-time driving cycle labels or real-time driving style labels, prob i To predict probabilities.

[0121] Training unit 1173 optimizes the model parameters in the artificial neural network model in the direction of reducing the target loss value to obtain the target driving cycle artificial neural network model and the target driving style artificial neural network model.

[0122] The label generation module 118 is used to call the target driving cycle artificial neural network model, perform recognition processing based on the non-trained driving principal component features to obtain the predicted driving cycle label, and call the target driving style artificial neural network model, perform recognition processing based on the non-trained driving principal component features to obtain the predicted driving style label.

[0123] The data storage module 130 stores the predicted driving cycle label and the predicted driving style label for each time segment, and records the corresponding time number for each time segment. Specifically, the data storage module 130 stores the predicted driving cycle label and the predicted driving style label for each time segment in a sliding window, and records the time number for each time segment. Taking the first time segment as an example, its time number is 1, and its corresponding predicted driving cycle label and predicted driving style label are stored.

[0124] The comparison module 150 is used to count the number of predicted driving cycle tags or predicted driving style tags for each time segment stored in the database, and to determine whether the stored number is equal to or greater than a preset storage number. Specifically, when the stored number is less than the preset storage number, the accumulation continues. It is understood that the preset storage number can be set based on actual needs, and this embodiment does not limit it.

[0125] When the storage quantity is equal to or greater than the preset storage quantity, the first calculation module 170 is used to calculate an average style score based on the time number and the predicted driving style label. The first calculation module 170 may include at least: a first calculation module 171 and a second calculation module 173.

[0126] The first calculation module 171 is used to calculate the time decay weight based on the time number. Specifically, the first calculation module 171 puts time segments with the same predicted driving cycle label into a time segment set. The calculation process for the time decay weight of each time segment is as follows:

[0127]

[0128] Where λ is the time decay coefficient, λ>0, f ik t is the time number of the k-th time segment in the time segment set, and t is the time number of the current time segment.

[0129] The second calculation module 173 is used to perform a weighted calculation based on the predicted driving style label and the time decay weight to obtain the average style score. Specifically, the second calculation module 173 calculates the average style score based on the time decay weight of each time segment. And the style score v corresponding to the predicted driving style label for each time segment ik A weighted average style score is obtained by performing a weighted calculation. The average style score is avg_score. i The calculation process for (t) is as follows:

[0130]

[0131] Among them, v ik For each time segment, predict the style score corresponding to the driving style label. The time decay weight for each time segment is defined by k, where k is the time number of the time segment.

[0132] When the storage quantity is equal to or greater than the preset storage quantity, the second calculation module 180 is used to calculate the entropy weight based on the predicted driving cycle label and the predicted driving style label. The second calculation module 180 may include at least: a third calculation module 181, a fourth calculation module 183, and a fifth calculation module 185.

[0133] The third calculation module 181 is used to calculate the frequency of occurrence of each type of driving style label based on the predicted driving cycle label and the predicted driving style label. Specifically, it calculates the frequency of occurrence of each type of predicted driving style label within the time segment of each type of predicted driving cycle label to obtain the frequency of occurrence of each type of driving style label.

[0134] The fourth calculation module 183 is used to calculate the entropy value based on the frequency of occurrence of each type of driving style label. Specifically, the fourth calculation module 183 calculates the entropy value based on the frequency of occurrence of each type of driving style label. The entropy value reflects the dispersion or uncertainty of the predicted driving style label in the time segment of each type of predicted driving cycle label. The smaller the entropy value, the more concentrated the style of this cycle, the stronger the discrimination, and the higher the weight; the larger the entropy value, the more mixed the style of this cycle, the weaker the discrimination, and the lower the weight. The calculation process of the entropy value is as follows:

[0135]

[0136] Where i represents the type of the predicted driving cycle label, k is the normalization coefficient, and s ij Let be the frequency of the predicted driving style label of the j-th class within the time segment of the predicted driving cycle label of the i-th class.

[0137] The fifth calculation module 185 is used to calculate the entropy weight based on the entropy value. Specifically, the fifth calculation module 185 calculates the entropy weight based on the entropy value. The calculation process of the entropy weight is as follows:

[0138]

[0139] Among them, 1-E i Let be the amount of effective information in the time segment of the i-th class predicting the driving cycle label.

[0140] The style determination module 190 is used to perform a weighted calculation based on the average style score and the entropy weight to obtain a target driving style score, and to determine the target driving style based on the target driving style score and a preset score. Specifically, the style determination module 190 performs a weighted summation based on the average style score and the entropy weight to obtain the target driving style score. The calculation process of the target driving style score is as follows:

[0141]

[0142] When the target driving style score total If the score is >85, the target driving style is determined to be conservative; if the score is <75, the target driving style score is determined to be conservative. total If the score is ≤85, the target driving style is determined to be conventional; when the target driving style score is ≤85, the target driving style is determined to be conventional. total If the value is ≤75, then the target driving style is determined to be aggressive.

[0143] In summary, in the driving style recognition system based on the entropy weight method of this application, the data processing module 110 processes the original driving data of the target vehicle and the artificial neural network model to obtain the predicted driving cycle label and the predicted driving style label for each time segment. The data storage module 130 stores the predicted driving cycle label and the predicted driving style label for each time segment and records the time number corresponding to each time segment. The comparison module 150 counts the number of the stored predicted driving cycle label or the predicted driving style label for each time segment and determines whether the stored number is equal to or greater than the preset storage number. When the number of stored items is equal to or greater than the preset number of stored items, the first calculation module 170 calculates an average style score based on the time number and the predicted driving style label. When the number of stored items is equal to or greater than the preset number of stored items, the second calculation module 180 calculates an entropy weight based on the predicted driving cycle label and the predicted driving style label. The style determination module 190 performs a weighted calculation based on the average style score and the entropy weight to obtain a target driving style score, and determines the target driving style based on the target driving style score and the preset score. This not only significantly improves the objectivity and reproducibility of the score, but also flexibly adapts to the actual characteristics of different drivers, different road conditions or time periods, and has strong environmental adaptability and generalization ability.

[0144] In the description of this specification, the references to terms such as "an embodiment," "some embodiments," "example," "specific example," "a implementation," "a preferred implementation," or "some examples," etc., indicate that a specific feature, structure, material, or characteristic described in connection with that embodiment or example is included in at least one embodiment or example of the present invention. In this specification, the illustrative expressions of the above terms do not necessarily refer to the same embodiment or example. Furthermore, the specific features, structures, materials, or characteristics described may be combined in any suitable manner in one or more embodiments or examples.

[0145] Although embodiments of the invention have been shown and described, those skilled in the art will understand that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the claims and their equivalents.

Claims

1. A driving style recognition method based on entropy weight method, characterized in that, The driving style recognition method based on entropy weight method includes: Based on the original driving data of the target vehicle and the artificial neural network model, the predicted driving cycle label and the predicted driving style label for each time segment are obtained. Based on the target driving data of each time segment, the driving cycle features and driving style features of the corresponding time segment are extracted. The driving cycle features are extracted based on the vehicle speed, acceleration and time distribution in the target driving data, and the driving style features are extracted based on the pedal data and acceleration in the target driving data. The predicted driving cycle label and the predicted driving style label for each time segment are stored, and the time number corresponding to each time segment is recorded. The number of predicted driving cycle tags or predicted driving style tags for each time segment stored is counted, and it is determined whether the number stored is equal to or greater than the preset number of storage tags. When the storage quantity is equal to or greater than the preset storage quantity, an average style score is calculated based on the time number and the predicted driving style label. When the storage quantity is equal to or greater than the preset storage quantity, the entropy weight is calculated based on the predicted driving cycle label and the predicted driving style label. The target driving style score is obtained by weighting the average style score and the entropy weight, and the target driving style is determined based on the target driving style score and the preset score. The calculation of the average style score based on the time number and the predicted driving style label includes at least the following steps: The time decay weight is calculated based on the time number. Time segments with the same predicted driving cycle label are placed into a time segment set. The calculation process for the time decay weight of each time segment is as follows: ； in, The time decay coefficient, , The time number of the k-th time segment in the set of time segments. The time number of the current time segment; The average style score is obtained by weighting the predicted driving style label and the time decay weight. Time decay weight based on each time segment And the style score corresponding to the predicted driving style label for each time segment. A weighted calculation is performed to obtain the average style score; the average style score The calculation process is as follows: ； in, For each time segment, predict the style score corresponding to the driving style label. For each time segment, assign a time decay weight. The time number of the time segment; The calculation of entropy weights based on the predicted driving cycle label and the predicted driving style label includes at least the following steps: The frequency of each type of driving style label is calculated based on the predicted driving cycle label and the predicted driving style label. The entropy value is calculated based on the frequency of occurrence of each type of driving style label. The entropy value is calculated using the following formula: ； ； in, To predict the type of driving cycle label, The normalization coefficient is... In the first In the time segment of the predicted driving cycle label, the first Predict the frequency of occurrence of driving style tags; The entropy weight is obtained by calculating based on the entropy value; The calculation process for entropy weights is as follows: ； in, For the first The amount of effective information in time segments of driving cycle labels is predicted; The target driving style score is calculated using the following formula: ； in, For entropy weights, To predict the type of driving cycle label, This represents the average style score.

2. The driving style recognition method based on entropy weight method according to claim 1, characterized in that, The process, based on the target vehicle's original driving data and an artificial neural network model, yields a predicted driving cycle label and a predicted driving style label for each time segment, including: The original driving data of the target vehicle is acquired in real time, and the target driving data is obtained by cleaning the original driving data. The original driving data includes training original driving data and non-training original driving data. The target driving data is divided into time segments based on a preset driving time to obtain target driving data for at least one time segment. Based on the target driving data of each time segment, feature extraction is performed to obtain the driving cycle features and driving style features of the corresponding time segment; The driving principal component features are obtained by dimensionality reduction based on the driving cycle features and the driving style features, wherein the driving principal component features include trained driving principal component features and untrained driving principal component features. The artificial neural network model is trained based on the principal component features of the training driving to obtain the target driving cycle artificial neural network model and the target driving style artificial neural network model, respectively. The target driving cycle artificial neural network model is invoked, and the predicted driving cycle label is obtained by recognition processing based on the non-trained driving principal component features. The target driving style artificial neural network model is also invoked, and the predicted driving style label is obtained by recognition processing based on the non-trained driving principal component features.

3. The driving style recognition method based on entropy weight method according to claim 2, characterized in that, The process of training the artificial neural network model based on the principal component features of the training driving system to obtain the target driving cycle artificial neural network model and the target driving style artificial neural network model includes: Clustering is performed based on the training driving principal component features to obtain the corresponding real driving cycle label and the corresponding real driving style label; The artificial neural network model is trained based on the training driving principal component features, the real driving cycle label, and the real driving style label to obtain the target driving cycle artificial neural network model and the target driving style artificial neural network model, respectively.

4. The driving style recognition method based on entropy weight method according to claim 3, characterized in that, The process of training the artificial neural network model based on the training driving principal component features, the real driving cycle labels, and the real driving style labels to obtain the target driving cycle artificial neural network model and the target driving style artificial neural network model, respectively, includes: The predicted probability is obtained based on the trained driving principal component features; The target loss value is calculated based on the predicted probability, the actual driving cycle label, and the actual driving style label. The model parameters in the artificial neural network model are optimized in the direction of reducing the target loss value to obtain the target driving cycle artificial neural network model and the target driving style artificial neural network model.

5. A driving style recognition system based on the entropy weight method, characterized in that, The driving style recognition system based on the entropy weight method includes: a data processing module, a data storage module, a comparison module, a first calculation module, a second calculation module, and a style determination module, wherein, The data processing module processes the target vehicle's raw driving data and artificial neural network model to obtain the predicted driving cycle label and the predicted driving style label for each time segment. It also extracts features from the target driving data for each time segment to obtain the corresponding driving cycle features and driving style features. The driving cycle features are extracted based on vehicle speed, acceleration, and time distribution in the target driving data, while the driving style features are extracted based on pedal data and acceleration in the target driving data. The data storage module is used to store the predicted driving cycle label and the predicted driving style label of each time segment, and to record the time number corresponding to each time segment; The comparison module is used to count the number of predicted driving cycle tags or predicted driving style tags for each time segment stored, and to determine whether the stored number is equal to or greater than the preset storage number. When the storage quantity is equal to or greater than the preset storage quantity, the first calculation module is used to calculate based on the time number and the predicted driving style label to obtain an average style score; When the storage quantity is equal to or greater than the preset storage quantity, the second calculation module is used to calculate based on the predicted driving cycle label and the predicted driving style label to obtain the entropy weight. The style determination module is used to perform a weighted calculation based on the average style score and the entropy weight to obtain a target driving style score, and to determine the target driving style based on the target driving style score and the preset score; The calculation of the average style score based on the time number and the predicted driving style label includes at least the following steps: The time decay weight is calculated based on the time number. Time segments with the same predicted driving cycle label are placed into a time segment set. The calculation process for the time decay weight of each time segment is as follows: ； in, The time decay coefficient, , The time number of the k-th time segment in the set of time segments. The time number of the current time segment; The average style score is obtained by weighting the predicted driving style label and the time decay weight. Time decay weight based on each time segment And the style score corresponding to the predicted driving style label for each time segment. A weighted calculation is performed to obtain the average style score; the average style score The calculation process is as follows: ； in, For each time segment, predict the style score corresponding to the driving style label. For each time segment, assign a time decay weight. The time number of the time segment; The calculation of entropy weights based on the predicted driving cycle label and the predicted driving style label includes at least the following steps: The frequency of each type of driving style label is calculated based on the predicted driving cycle label and the predicted driving style label. The entropy value is calculated based on the frequency of occurrence of each type of driving style label. The entropy value is calculated using the following formula: ； ； in, To predict the type of driving cycle label, The normalization coefficient is... In the first In the time segment of the predicted driving cycle label, the first Predict the frequency of occurrence of driving style tags; The entropy weight is obtained by calculating based on the entropy value; The calculation process for entropy weights is as follows: ； in, For the first The amount of effective information in time segments of driving cycle labels is predicted; The target driving style score is calculated using the following formula: ； in, For entropy weights, To predict the type of driving cycle label, This represents the average style score.

6. A driving style recognition system based on entropy weight method according to claim 5, characterized in that, The data processing module includes a data acquisition module, a time segmentation module, a feature extraction module, a feature processing module, a model training module, and a label generation module. The data acquisition module is used to acquire the raw driving data of the target vehicle in real time, and to clean the raw driving data to obtain the target driving data. The raw driving data includes training raw driving data and non-training raw driving data. The time segmentation module is used to divide the target driving data into time segments based on a preset driving time, so as to obtain target driving data in at least one time segment. The feature extraction module is used to extract features from the target driving data of each time segment to obtain the driving cycle features and driving style features of the corresponding time segment; The feature processing module is used to perform dimensionality reduction processing based on the driving cycle features and the driving style features to obtain driving principal component features, wherein the driving principal component features include trained driving principal component features and untrained driving principal component features; The model training module is used to train the artificial neural network model based on the training driving principal component features, to obtain the target driving cycle artificial neural network model and the target driving style artificial neural network model respectively; The label generation module is used to call the target driving cycle artificial neural network model, perform recognition processing based on the non-trained driving principal component features to obtain the predicted driving cycle label, and call the target driving style artificial neural network model, perform recognition processing based on the non-trained driving principal component features to obtain the predicted driving style label.

Images