Driving intention prediction method and device, equipment, medium and vehicle

Abstract

The invention relates to a driving intention prediction method and device, equipment, a medium and a vehicle, and the method comprises the steps: obtaining a current event instance of a to-be-predicted vehicle, the current event instance comprising a user behavior guidance event, an equipment state change event and a recording event of a target application at a current moment and within a preset time period before the current moment; according to a preset rule, performing graph construction on the user behavior guidance event, the equipment state change event and the record event of the target application to form a current event graph of the to-be-predicted vehicle; and analyzing the current event atlas by using an atlas content analysis method included in the target prediction model to obtain driving intention data of the to-be-predicted vehicle. Therefore, the driving intention can be predicted and recommended based on event fusion of multiple dimensions, the function of the intelligent recommendation technology is expanded, and meanwhile, various intelligent recommendation requirements of the user are met.

Description

Technical Field

[0001] The present disclosure relates to the field of intelligent recommendation technology, and in particular to a driving intention prediction method, device, equipment, medium and vehicle. Background Art

[0002] With the continuous development of vehicle technology, users are driving vehicles more and more convenient. In vehicle computers, mobile phones, tablets, computers and other terminal devices, usually intelligent recommendation applications can be set up to recommend content of interest to users.

[0003] Existing intelligent recommendation applications are difficult to integrate user behavior events, device status change events, and application record events to predict driving intentions, and therefore cannot make driving intention recommendations based on the predicted results of driving intentions. It can be seen that the existing intelligent recommendation technology is not intelligent enough and it is difficult to meet users' various intelligent recommendation needs. Summary of the invention

[0004] In order to solve the above technical problems or at least partially solve the above technical problems, the present disclosure provides a driving intention prediction method, device, equipment, medium and vehicle.

[0005] In a first aspect, the present disclosure provides a driving intention prediction method, the method comprising:

[0006] Obtaining a current event instance of the vehicle to be predicted, wherein the current event instance includes user behavior guidance events, device state change events, and target application record events at the current moment and within a preset time period before the current moment;

[0007] According to preset rules, a graph is constructed for the user behavior guidance event, the device state change event, and the recorded event of the target application included in the current event instance to form a current event graph of the vehicle to be predicted;

[0008] The current event graph is analyzed using the graph content analysis method included in the target prediction model to obtain the driving intention data of the vehicle to be predicted.

[0009] In some embodiments of the present disclosure, the graph content parsing method includes a data encoding method and a data decoding method; the graph content parsing method included in the target prediction model is used to parse the current event graph to obtain the driving intention data of the vehicle to be predicted, including:

[0010] Based on the data encoding method corresponding to the transformer encoder in the target prediction model, the current event graph is encoded to obtain a plurality of encoding features corresponding to the current event graph;

[0011] Based on the data decoding method corresponding to the autoregressive decoder in the target prediction model, a plurality of encoding features corresponding to the current event map are decoded to obtain the driving intention data of the vehicle to be predicted.

[0012] In some embodiments of the present disclosure, the data encoding method includes a first multi-head self-attention processing mechanism and a first feature processing method; the data encoding method corresponding to the transformer encoder in the target prediction model encodes the current event graph to obtain a plurality of encoding features corresponding to the current event graph, including:

[0013] Based on a first multi-head self-attention processing mechanism corresponding to a first multi-head self-attention network in the transformer encoder, performing multi-head self-attention processing on the current event graph to obtain a first multi-head self-attention feature;

[0014] Based on the first feature processing method corresponding to the feedforward neural network in the transformer encoder, feature transformation and feature extraction are performed on the first multi-head self-attention feature to obtain a plurality of encoding features corresponding to the current event map.

[0015] In some embodiments of the present disclosure, the data decoding method includes a second multi-head self-attention processing mechanism, a multi-head attention mechanism, and a second feature processing method; the data decoding method corresponding to the autoregressive decoder in the target prediction model decodes the multiple encoding features corresponding to the current event map to obtain the driving intention data of the vehicle to be predicted, including:

[0016] Based on a second multi-head self-attention processing mechanism corresponding to the second multi-head self-attention network in the autoregressive decoder, multi-head self-attention processing is performed on the multiple encoding features to obtain second multi-head self-attention features;

[0017] Performing multi-head attention processing on the second multi-head self-attention feature based on the multi-head attention mechanism corresponding to the multi-head attention network in the autoregressive decoder to obtain a multi-head attention feature;

[0018] Based on the second feature processing method corresponding to the feedforward neural network in the autoregressive decoder, the multi-head attention feature is transformed and extracted to obtain a predicted driving text, and the predicted driving text is used as the driving intention data of the vehicle to be predicted.

[0019] In some embodiments of the present disclosure, the step of obtaining the current event instance of the vehicle to be predicted includes:

[0020] Obtaining user behavior guidance data, device state change data, and target application record data of the vehicle to be predicted;

[0021] Based on multiple data items in a preset instance format, respectively extracting data values ​​corresponding to the multiple data items from the user behavior guidance data, the device state change data, and the record data of the target application;

[0022] Filling the data values ​​respectively corresponding to the multiple data items into the preset instance format to obtain multiple event instances;

[0023] The multiple event instances are respectively configured with corresponding event identifiers to obtain a current event instance including the user behavior guidance event, the device state change event, and the target application's recorded event.

[0024] In some embodiments of the present disclosure, the user behavior guidance event, the device state change event, and the target application record event included in the current event instance are graphed according to preset rules to form the current event graph of the vehicle to be predicted, including:

[0025] Sorting the user behavior guidance event, the device state change event, and the target application record event included in the current event instance in chronological order to obtain a sorted current event instance;

[0026] According to the identification information respectively corresponding to each event instance in the sorted current event instance, each event instance in the sorted current event instance is spliced ​​one by one to form the current event map of the vehicle to be predicted.

[0027] In some embodiments of the present disclosure, the user behavior guidance event includes one or more combinations of the following: voice guidance data, gesture guidance data;

[0028] The device state change event includes one or more of the following combinations: door state change data, seat belt state change data, light state change data, throttle state change data;

[0029] The recorded events of the target application include one or more of the following combinations: navigation recorded data, environment recorded data, and user portrait recorded data.

[0030] In some embodiments of the present disclosure, before obtaining the current event instance of the vehicle to be predicted, the method further includes:

[0031] Constructing a graph of historical event instances of a reference vehicle within a preset time period before a historical moment to form a historical event graph of the reference vehicle, wherein the historical event instances include a user behavior guidance reference, a device state change reference, and a target application record reference;

[0032] Based on the historical event graph and the reference driving intention of the reference vehicle, supervised training is performed on the pre-trained base model to obtain a supervised fine-tuning model;

[0033] Based on the evaluation statement corresponding to the reference driving intention, the historical event map and the reference driving intention, the supervised fine-tuning model is trained to obtain a reward model;

[0034] The supervised fine-tuning model is optimized using the reward model and the proximal strategy optimization method to obtain the target prediction model.

[0035] In a second aspect, the present disclosure provides a driving intention prediction device, the device comprising:

[0036] An acquisition module, used to acquire a current event instance of the vehicle to be predicted, wherein the current event instance includes a user behavior guidance event, a device state change event, and a target application record event at the current moment and within a preset time period before the current moment;

[0037] A formation module, configured to construct a graph for the user behavior guidance event, the device state change event, and the target application record event contained in the current event instance according to preset rules, so as to form a current event graph of the vehicle to be predicted;

[0038] The prediction module is used to analyze the current event graph using the graph content analysis method included in the target prediction model to obtain the driving intention data of the vehicle to be predicted.

[0039] In a third aspect, the present disclosure provides a driving intention prediction device, comprising:

[0040] processor;

[0041] A memory for storing executable instructions;

[0042] Among them, the processor is used to read executable instructions from the memory and execute the executable instructions to implement the driving intention prediction method of the first aspect.

[0043] In a fourth aspect, the present disclosure provides a computer-readable storage medium storing a computer program. When the computer program is executed by a processor, the processor implements the driving intention prediction method of the first aspect.

[0044] In a fifth aspect, the present disclosure provides a vehicle, comprising:

[0045] The driving intention prediction device as described in the second aspect; or

[0046] The driving intention prediction device as described in the third aspect; or

[0047] A computer-readable storage medium as described in the fourth aspect.

[0048] Compared with the prior art, the technical solution provided by the embodiments of the present disclosure has the following advantages:

[0049] The disclosed embodiment provides a method, device, equipment, medium and vehicle for predicting driving intention, the method comprising: obtaining a current event instance of a vehicle to be predicted, wherein the current event instance includes user behavior guidance events, device state change events and target application record events at the current moment and within a preset time period before the current moment; constructing a graph of the user behavior guidance events, device state change events and target application record events contained in the current event instance according to preset rules to form a current event graph of the vehicle to be predicted; parsing the current event graph using the graph content parsing method contained in the target prediction model to obtain driving intention data of the vehicle to be predicted. Thus, it is possible to achieve the fusion of user behavior events, device state change events and application record events to predict driving intention and make driving intention recommendations, which expands the functions of intelligent recommendation technology and also meets the user's various intelligent recommendation needs. BRIEF DESCRIPTION OF THE DRAWINGS

[0050] The above and other features, advantages and aspects of the embodiments of the present disclosure will become more apparent with reference to the following detailed description in conjunction with the accompanying drawings. Throughout the accompanying drawings, the same or similar reference numerals represent the same or similar elements. It should be understood that the drawings are schematic and the originals and elements are not necessarily drawn according to probability.

[0051] Figure 1 A flow chart of a driving intention prediction method provided by an embodiment of the present disclosure;

[0052] Figure 2 A flowchart of another driving intention prediction method provided by an embodiment of the present disclosure;

[0053] Figure 3 A flowchart of a method for fine-tuning a pre-trained base model provided in an embodiment of the present disclosure;

[0054] Figure 4 A schematic diagram of the structure of a driving intention prediction device provided by an embodiment of the present disclosure;

[0055] Figure 5 A schematic diagram of the structure of a driving intention prediction device provided in an embodiment of the present disclosure. DETAILED DESCRIPTION

[0056] Embodiments of the present disclosure will be described in more detail below with reference to the accompanying drawings. Although certain embodiments of the present disclosure are shown in the accompanying drawings, it should be understood that the present disclosure can be implemented in various forms and should not be construed as being limited to the embodiments described herein, which are instead provided for a more thorough and complete understanding of the present disclosure. It should be understood that the drawings and embodiments of the present disclosure are only for exemplary purposes and are not intended to limit the scope of protection of the present disclosure.

[0057] It should be understood that the various steps described in the method embodiments of the present disclosure may be performed in different orders and / or in parallel. In addition, the method embodiments may include additional steps and / or omit the steps shown. The scope of the present disclosure is not limited in this respect.

[0058] The term "including" and its variations used herein are open inclusions, i.e., "including but not limited to". The term "based on" means "based at least in part 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". The relevant definitions of other terms will be given in the following description.

[0059] It should be noted that the concepts such as "first" and "second" mentioned in the present disclosure are only used to distinguish different devices, modules or units, and are not used to limit the order or interdependence of the functions performed by these devices, modules or units.

[0060] It should be noted that the modifications of "one" and "plurality" mentioned in the present disclosure are illustrative rather than restrictive, and those skilled in the art should understand that unless otherwise clearly indicated in the context, it should be understood as "one or more".

[0061] This embodiment provides a driving intention prediction method, which can be executed by a controller in a target vehicle. Figure 1 A flow chart of a driving intention prediction method provided by an embodiment of the present disclosure is shown.

[0062] like Figure 1 As shown, the driving intention prediction method may include the following steps.

[0063] S110. Obtain a current event instance of the vehicle to be predicted, wherein the current event instance includes user behavior guidance events, device state change events, and recorded events of the target application at the current moment and within a preset time period before the current moment.

[0064] In this embodiment, after starting the smart recommendation application on the vehicle, if you want to use the smart recommendation application to predict driving intention, you need to obtain events in multiple dimensions such as user behavior guidance events, device status change events, and target application record events of the vehicle to be predicted at the current moment and within a preset time period before the current moment to determine the current event instance of the vehicle to be predicted.

[0065] The vehicle to be predicted is any vehicle that needs to plan a navigation path.

[0066] The preset time period before the current time is any time period for collecting data, including but not limited to 5 minutes, 10 minutes or 30 minutes.

[0067] The user behavior guidance event can be understood as an event that guides the intelligent recommendation application to predict driving intention. Optionally, the user behavior guidance event includes but is not limited to a voice guidance event and a gesture guidance event.

[0068] The device state change event can be understood as the event of different devices of the vehicle to be predicted. Optionally, the device state change event includes but is not limited to: a door state change event, a seat belt state change event, a light state change event, and a throttle state change event.

[0069] Among them, the recorded events of the target application can be understood as historical events recorded by the driving intention recording application. Specifically, the target application can be buried in advance so that when the driving intention is predicted, the buried records of the target application can be directly obtained as the recorded events of the target application. Optionally, the target application includes but is not limited to navigation applications, automatic driving control applications, environmental information display applications, user portrait display applications, etc. The recorded events of the target application include but are not limited to: navigation recorded events, automatic driving recorded events, environmental recorded events, and user portrait recorded events.

[0070] In order to facilitate the analysis of user behavior guidance events, device state change events, and target application record events, features of events in these three dimensions can be constructed in the time dimension, space dimension, and event dimension, so as to facilitate the determination of the current event instance based on the data of the three dimensions after feature construction.

[0071] When constructing features in the time dimension, user behavior guidance events, device status change events, and target application record events are divided into 24 time periods, each time period is 1 hour, and the events in each time period include but are not limited to holidays, non-holidays, weekdays, weekends and other date types.

[0072] When constructing features in the spatial dimension, the geographic location of the vehicle is determined based on user behavior guidance events, device state change events, and target application record events, and events belonging to the same geographic location are determined as current event instances. Optionally, the geographic location includes but is not limited to longitude and latitude discretized according to different areas, point of interest (POI) type, POI name, POI identifier, etc.

[0073] When constructing features in the event dimension, user behavior guidance events, device state change events, and target application record events include events in different time series, which are the sequences before the behavior intention prediction, and the events in different time series are taken as current event instances.

[0074] S120. According to preset rules, a graph is constructed for user behavior guidance events, device state change events, and recorded events of the target application to form a current event graph of the vehicle to be predicted.

[0075] It can be understood that since the event graph is compatible with events of multiple dimensions of different events and has good scalability, it can integrate multiple dimensional events such as user behavior guidance events, device state change events, and target application record events contained in the current event instance into an event graph according to preset rules, so that the driving intention of the predicted vehicle can be predicted using the above-mentioned multiple dimensional events.

[0076] The preset rules refer to the integration method of events in multiple dimensions. Specifically, the preset rules may include sorting rules and splicing rules, so as to form a current event map based on the sorting rules and splicing rules.

[0077] In order to improve the efficiency of driving intention prediction, the specific implementation method of S120 includes but is not limited to the following methods: sorting the user behavior guidance events, device status change events and target application record events contained in the current event instance in chronological order to obtain the sorted current event instance; according to the identification information corresponding to each event instance in the sorted current event instance, splicing each event instance in the sorted current event instance one by one to form a current event map of the vehicle to be predicted.

[0078] In some embodiments, the identification information includes an event identification. Accordingly, the specific implementation method of successive splicing includes: according to the event identifications corresponding to the respective event instances in the sorted current event instances, each event instance in the sorted current event instances is spliced ​​together one by one to form a current event map of the vehicle to be predicted.

[0079] The event identifier is used to distinguish different event instances, that is, different event instances correspond to different event identifiers.

[0080] In other embodiments, the identification information includes the event name. Accordingly, the specific implementation method of successive splicing includes: if there are event instances with the same event name in the sorted current event instances, the event instances with the same event name are spliced ​​to obtain a preliminary splicing result; according to the event name, the remaining event instances in the sorted current event instances and the preliminary splicing results are continued to be spliced ​​to form a current event map of the vehicle to be predicted.

[0081] In some other embodiments, the identification information includes the event name, and the specific implementation method of successive splicing includes: if there is no event instance with the same event name in the sorted current event instance, then directly splice each event instance in the sorted current event instance one by one according to the event name corresponding to each event instance in the sorted current event instance, so as to form a current event map of the vehicle to be predicted.

[0082] The event name refers to the action transformation name corresponding to different event instances. Different event instances may correspond to the same event name or different event names.

[0083] For example, when constructing the current event graph, if events of two dimensions are obtained. The events of the first dimension are: event identifier (such as 10001), event name (such as the door from opening to closing), entity (such as door), predicate (such as opening to closing), occurrence date (such as 2023.1.1), occurrence time (such as 0:00), timestamp, event type (such as behavior type, voice type), vehicle and other attributes; the events of the second dimension are: event identifier (such as 10002), event name (such as navigation application record data), entity (such as navigation application), predicate (such as driving on the highway), occurrence date (such as 2023.1.2), occurrence time (such as 0:00), timestamp, event type (such as behavior type, voice type), vehicle and other attributes. After sorting the events of the above two dimensions in chronological order, according to the event names corresponding to the events of the above two dimensions (including the door from opening to closing and the navigation application record data) and / or the event identifiers corresponding to the events of the two dimensions (including 10001 and 10002), the events of the above two dimensions are spliced ​​one by one to form the current event map. In this way, the current event instances at the current moment and within a preset time period (eg, 5 minutes) before the current moment are constructed into a current event graph according to preset rules.

[0084] Therefore, when predicting the driving intention of the predicted vehicle, the current event graph can be constructed by first splicing the event instances in the current event instance in chronological order, and then constructing the current event graph according to the identification information corresponding to each event instance in the sorted current event instance. In this way, the reliability and orderliness of the construction of the current event graph are guaranteed.

[0085] S130. Analyze the current event graph using the graph content analysis method included in the target prediction model to obtain the driving intention data of the vehicle to be predicted.

[0086] In this embodiment, the intelligent recommendation application on the vehicle to be predicted downloads the target prediction model from the cloud in advance or calls it from the database, and then uses the current event instance as the input data of the target prediction model to predict the driving intention and outputs the driving intention data of the vehicle to be predicted.

[0087] The target prediction model can be a large-scale pre-trained model (Large Pretrained Language Model, LPLM) obtained by pre-training and fine-tuning. The large-scale pre-trained model can also be understood as a large language model (Large Language Models, LLM), a foundation model or a foundation model (Foundation Models, FM).

[0088] Specifically, the target prediction model is embedded with a graph content analysis method, so that the target prediction model can analyze the current event graph based on the graph content analysis method to realize driving intention prediction processing of the current event graph and output the driving intention data of the vehicle to be predicted, thereby realizing the effect of intelligently predicting driving intention data based on data from multiple dimensions.

[0089] Optionally, the target prediction model includes but is not limited to a transformer-based language model (ChatGPT), and may also be other types of base models.

[0090] The graph content analysis method can be understood as a method for converting the current event graph into driving intention data. Optionally, the graph content analysis method includes a data encoding method and a data decoding method.

[0091] Optionally, the driving intention data includes but is not limited to the navigation destination, the predicted navigation route, the operating status data, and the application usage preference data during driving.

[0092] The disclosed embodiment provides a method for predicting driving intention, the method comprising: obtaining a current event instance of a vehicle to be predicted, wherein the current event instance includes user behavior guidance events, device state change events, and target application record events at the current moment and within a preset time period before the current moment; constructing a graph of user behavior guidance events, device state change events, and target application record events according to preset rules to form a current event graph of the vehicle to be predicted; parsing the current event graph using a graph content parsing method included in a target prediction model to obtain driving intention data of the vehicle to be predicted. Thus, it is possible to achieve the fusion of user behavior events, device state change events, and application record events to predict driving intention and make driving intention recommendations, which expands the functions of intelligent recommendation technology and also meets the user's various intelligent recommendation needs.

[0093] In another embodiment of the present disclosure, in order to facilitate understanding of the method for determining a navigation destination, a process for acquiring a current event instance and a processing principle of a driving intention prediction model are specifically described.

[0094] Figure 2 A flow chart of another driving intention prediction method provided by an embodiment of the present disclosure is shown.

[0095] like Figure 2 As shown, the driving intention prediction method may include the following steps.

[0096] S210: Obtain user behavior guidance data, device state change data, and target application record data of the vehicle to be predicted.

[0097] In this embodiment, sensors with different functions are installed at multiple locations of the vehicle to be predicted. The sensors on the vehicle to be predicted collect user behavior guidance signals, current device status change signals, and target application recording signals in real time. Then, these three signals are converted into corresponding data respectively to obtain user behavior guidance data, device status change data, and target application recording data.

[0098] Among them, the user behavior guidance signal can be the user's historical conversation log, which records the user's behavior guidance data. The current device state change signal can be a historical log when the state of various hardware devices in the vehicle to be predicted changes, which records the state change data of the hardware devices. The target application recording signal refers to the embedded point record of the target application in the vehicle, including but not limited to navigation information, autonomous driving information, environmental recording information, and user portrait recording information.

[0099] Among them, the method for obtaining user behavior guidance data includes but is not limited to the following methods: performing behavior recognition processing on the user behavior guidance signal to obtain the user behavior guidance data.

[0100] Optionally, if the user behavior guidance signal includes a voice guidance signal, the behavior recognition processing method includes the following steps: obtaining the voice guidance signal of the vehicle to be predicted; extracting keywords and numerical values ​​corresponding to the keywords from the voice guidance signal, and using the keywords and numerical values ​​corresponding to the keywords as user behavior guidance data.

[0101] The voice guidance signal is voice information collected by a voice sensor. Optionally, the voice guidance signal includes but is not limited to a chat voice signal of a user in the car, a voice signal emitted by an in-car entertainment device, and a current voice guidance signal.

[0102] Optionally, if the user behavior guidance signal includes a gesture guidance signal, the behavior recognition processing method includes the following steps: identifying a guidance object from the gesture guidance signal, and using the guidance object as user behavior guidance data.

[0103] Optionally, the gesture guidance signal is collected by an induction sensor.

[0104] The method for acquiring the device state change data includes, but is not limited to, the following manner: identifying the first signal identifier and the change information corresponding to the first signal identifier from the device state change signal to obtain the device state change data.

[0105] Optionally, the first signal identifier includes but is not limited to a door signal identifier, a seat belt signal identifier, a light signal identifier, and a throttle signal identifier. Accordingly, the change information corresponding to the first signal identifier includes but is not limited to door state change data, seat belt state change data, light state change data, and throttle state change data.

[0106] The method for acquiring the recorded data of the target application includes, but is not limited to, the following manner: extracting the second identification signal identifier and the key field corresponding to the second signal identifier from the recorded signal of the target application to obtain the recorded data of the target application.

[0107] Optionally, the second signal identifier includes but is not limited to a navigation application identifier, an automatic driving control application identifier, an environment information display application identifier, a user portrait display application identifier, etc. The recorded data of the target application includes but is not limited to: navigation record information, automatic driving record information, environment record information, and user portrait record information.

[0108] S220. Based on the multiple data items in the preset instance format, respectively extract data values ​​corresponding to the multiple data items from the user behavior guidance data, the device state change data, and the record data of the target application.

[0109] The preset construction format can be understood as a data integration method, and the preset construction format includes multiple data items. Optionally, the multiple data items include but are not limited to parameters such as event name, entity, predicate, occurrence date, occurrence time, timestamp, event type, vehicle, and other attributes.

[0110] In order to improve data analysis efficiency, after obtaining user behavior guidance data, device status change data and record data of the target application, data values ​​corresponding to multiple data items are extracted from these three types of data, so that only these data values ​​can be processed later to improve data processing efficiency.

[0111] S230: Fill the data values ​​corresponding to the multiple data items into the preset instance format to obtain multiple event instances.

[0112] In this embodiment, the preset construction format can be specifically arranged in the order of parameters such as event name, entity, predicate, occurrence date, occurrence time, timestamp, event type, vehicle and other attributes. Then, according to this arrangement order, the data values ​​corresponding to multiple data items can be filled into the preset instance format to obtain multiple event instances.

[0113] S240: Configure corresponding event identifiers for each of the multiple event instances to obtain a current event instance including a user behavior guidance event, a device state change event, and a target application record event.

[0114] In order to facilitate the identification of each event instance, multiple event instances are randomly configured with corresponding event identifiers to form a complete current event instance.

[0115] S250. According to preset rules, a graph is constructed for user behavior guidance events, device state change events, and recorded events of the target application included in the current event instance to form a current event graph of the vehicle to be predicted.

[0116] Among them, S250 is similar to S120 and will not be elaborated here.

[0117] S260. Based on the data encoding method corresponding to the transformer encoder in the target prediction model, encode the current event graph to obtain a variety of encoding features corresponding to the current event graph.

[0118] Among them, the transformer encoder is similar to the architecture of other transformer language models. It consists of multiple attention mechanisms and can be used to encode input sequences.

[0119] Specifically, the data encoding method includes a first multi-head self-attention processing mechanism and a first feature processing method. Accordingly, the specific implementation of S260 includes but is not limited to the following methods: based on the first multi-head self-attention processing mechanism corresponding to the first multi-head self-attention network in the transformer encoder, multi-head self-attention processing is performed on the current event map to obtain the first multi-head self-attention feature; based on the first feature processing method corresponding to the feedforward neural network in the transformer encoder, feature transformation and feature extraction are performed on the first multi-head self-attention feature to obtain a variety of encoding features corresponding to the current event map.

[0120] Among them, the first multi-head self-attention processing mechanism corresponding to the first multi-head self-attention network can enable the transformer encoder to pay attention to the relationship between different positions in the input sequence when encoding, so as to better understand the entire sequence. The first feature processing method corresponding to the feedforward neural network can perform pattern classification processing on the first multi-head self-attention feature to obtain multiple encoding features.

[0121] S270. Based on the data decoding method corresponding to the autoregressive decoder in the target prediction model, a plurality of coding features corresponding to the current event map are decoded to obtain the driving intention data of the vehicle to be predicted.

[0122] Among them, the role of the autoregressive decoder is to generate the next text or token.

[0123] Specifically, the data decoding method includes a second multi-head self-attention processing mechanism, a multi-head self-attention mechanism, and a second feature processing method. Accordingly, the specific implementation of S270 includes but is not limited to the following methods: based on the second multi-head self-attention processing mechanism corresponding to the second multi-head self-attention network in the autoregressive decoder, a plurality of encoding features are subjected to multi-head self-attention processing to obtain a second multi-head self-attention feature; based on the multi-head self-attention mechanism corresponding to the multi-head attention network in the autoregressive decoder, a multi-head self-attention feature is subjected to multi-head attention processing to obtain a multi-head attention feature; based on the second feature processing method corresponding to the feedforward neural network in the autoregressive decoder, a multi-head attention feature is subjected to feature transformation and feature extraction to obtain a predicted driving text, and the predicted driving text is used as the driving intention data of the vehicle to be predicted.

[0124] Among them, the second multi-head self-attention processing mechanism corresponding to the second multi-head self-attention network can enable each attention mechanism to optimize different feature parts of each text, thereby balancing the deviations that may be caused by the same attention mechanism.

[0125] Therefore, by utilizing the data encoding method and data decoding method included in the graph content analysis method, it is possible to predict various driving intentions such as navigation destination, navigation route, and driving status, thereby meeting the user's needs for prediction of various driving intentions and ensuring the accuracy of driving intention predictions.

[0126] In yet another embodiment of the present disclosure, before predicting driving intention, the parameters of the pre-trained base model are fine-tuned using historical event instances of the reference vehicle and the reference driving intention to obtain a target prediction model.

[0127] Figure 3 A flowchart of a method for fine-tuning a pre-trained base model provided in an embodiment of the present disclosure is shown.

[0128] like Figure 3 As shown, the fine-tuning method of the pre-trained base model may include the following steps.

[0129] S310. Construct a graph of historical event instances of the reference vehicle within a preset time period before the historical moment to form a historical event graph of the reference vehicle, wherein the historical event instances include a user behavior guidance reference, a device state change reference, and a record reference of the target application.

[0130] The reference vehicle is any vehicle that plans a navigation path within a historical period of time.

[0131] The user behavior guidance reference may be understood as an event instance that guides the pre-trained base model to predict driving intention. Optionally, the user behavior guidance reference includes but is not limited to voice guidance events and gesture guidance events.

[0132] The device state change reference can be understood as an event instance of a reference to different devices of the vehicle. Optionally, the device state change reference includes but is not limited to: a door state change event, a seat belt state change event, a light state change event, and a throttle state change event.

[0133] The target application's record reference can be understood as an event instance recorded by the driving intention recording application on the reference vehicle. The target application's record reference includes but is not limited to: navigation record events, autonomous driving record events, environment record events, and user portrait record events.

[0134] In order to facilitate the analysis of user behavior guidance reference, device state change reference and target application record reference, the data of these three dimensions can be feature constructed in the time dimension, space dimension and event dimension, so as to facilitate the determination of historical event instances based on the data of the three dimensions after feature construction. It should be noted that the feature construction method here is similar to the feature construction method in the above embodiment, and will not be repeated here.

[0135] In this embodiment, the specific implementation method of S310 includes but is not limited to the following methods: sorting the user behavior guidance reference, the device state change reference and the record reference of the target application in chronological order to obtain a sorted historical event instance; according to the identification information corresponding to each event instance in the sorted historical event instance, splicing each event instance in the sorted historical event instance one by one to form a reference vehicle historical event map.

[0136] It should be noted that the specific method of determining the user behavior guidance reference, the device state change reference, and the target application record reference in this embodiment is similar to the above embodiment and will not be described in detail here. In addition, the method of sequentially splicing each event instance in the sorted historical event instance is similar to the above embodiment and will not be described in detail here.

[0137] S320: Based on the historical event graph and the reference driving intention of the reference vehicle, supervised training is performed on the pre-trained base model to obtain a supervised fine-tuning model.

[0138] It is understandable that the pre-trained base model is obtained through pre-training on massive data. It has certain general capabilities, which is equivalent to a huge knowledge base with "unknown potential". The pre-trained base model is fine-tuned with parameters using training data from a specific field to process data from a specific field, so that the target prediction model can understand and generate conversations or query responses related to the user's driving intentions.

[0139] S330: Based on the evaluation statement corresponding to the reference driving intention, the historical event map and the reference driving intention, the supervised fine-tuning model is trained to obtain a reward model.

[0140] The evaluation statement refers to the manual evaluation of the reference driving intention, and the evaluation statement can be a relative ranking. Specifically, the historical event graph is used as the input data of the supervised fine-tuning model, and the evaluation statement corresponding to the reference driving intention and the reference driving intention are used as the output data of the supervised fine-tuning model. The supervised fine-tuning model is trained using the input data and the output data to obtain a reward model (RM).

[0141] S340, using the reward model and the proximal strategy optimization method, the supervised fine-tuning model is optimized to obtain a target prediction model.

[0142] Among them, the proximal policy optimization method can specifically be a proximal policy optimization algorithm (PPO). Specifically, by using the reward model and the proximal policy optimization method to optimize the supervised fine-tuning model, the reward model can output results that are more in line with user expectations. Therefore, when the target prediction model is obtained by this method to predict driving intention data, the prediction results can be more in line with the prediction results expected by the user.

[0143] Therefore, the compatibility of the event graph is used to accommodate user behavior guidance references, device state change references, and target application record references, and then the historical event graph formed by events in these three dimensions is used to fine-tune the parameters of the pre-trained base model, so as to obtain a target prediction model that can predict results that are more in line with user expectations.

[0144] The present disclosure also provides a driving intention prediction device for implementing the above-mentioned driving intention prediction method. Figure 4 Provide explanation.

[0145] Figure 4 A schematic structural diagram of a driving intention prediction device provided by an embodiment of the present disclosure is shown.

[0146] like Figure 4 As shown, the driving intention prediction device 400 may include:

[0147] The acquisition module 410 is used to acquire the current event instance of the vehicle to be predicted, wherein the current event instance includes the user behavior guidance event, the device state change event, and the target application record event at the current moment and within a preset time period before the current moment. The current event instance of the vehicle to be predicted is acquired, wherein the current event instance includes the user behavior guidance event, the device state change event, and the target application record event;

[0148] A forming module 420 is used to construct a graph for the user behavior guidance event, the device state change event, and the recorded event of the target application included in the current event instance according to a preset rule to form a current event graph of the vehicle to be predicted;

[0149] The prediction module 430 is used to analyze the current event graph using the graph content analysis method included in the target prediction model to obtain the driving intention data of the vehicle to be predicted.

[0150] The driving intention prediction device of the disclosed embodiment includes: obtaining the current event instance of the vehicle to be predicted, wherein the current event instance includes the user behavior guidance events, device state change events, and target application record events at the current moment and within a preset time period before the current moment; constructing a graph of the user behavior guidance events, device state change events, and target application record events according to preset rules to form a current event graph of the vehicle to be predicted; parsing the current event graph using the graph content parsing method included in the target prediction model to obtain the driving intention data of the vehicle to be predicted. In this way, it is possible to achieve the fusion of user behavior events, device state change events, and application record events to predict driving intention and make driving intention recommendations, which expands the functions of intelligent recommendation technology and also meets the user's various intelligent recommendation needs.

[0151] In some embodiments of the present disclosure, the graph content parsing method includes a data encoding method and a data decoding method; accordingly, the prediction module 430 includes:

[0152] An encoding unit, configured to encode the current event graph based on a data encoding method corresponding to a transformer encoder in the target prediction model, to obtain a plurality of encoding features corresponding to the current event graph;

[0153] A decoding unit is used to decode the multiple coding features corresponding to the current event map based on the data decoding method corresponding to the autoregressive decoder in the target prediction model to obtain the driving intention data of the vehicle to be predicted.

[0154] In some embodiments of the present disclosure, the data encoding method includes a first multi-head self-attention processing mechanism and a first feature processing method; accordingly, the encoding unit is specifically used to:

[0155] Based on a first multi-head self-attention processing mechanism corresponding to a first multi-head self-attention network in the transformer encoder, performing multi-head self-attention processing on the current event graph to obtain a first multi-head self-attention feature;

[0156] Based on the first feature processing method corresponding to the feedforward neural network in the transformer encoder, feature transformation and feature extraction are performed on the first multi-head self-attention feature to obtain a plurality of encoding features corresponding to the current event map.

[0157] In some embodiments of the present disclosure, the data decoding method includes a second multi-head self-attention processing mechanism, a multi-head self-attention mechanism and a second feature processing method; accordingly, the decoding unit is specifically used to:

[0158] Based on a second multi-head self-attention processing mechanism corresponding to the second multi-head self-attention network in the autoregressive decoder, multi-head self-attention processing is performed on the multiple encoding features to obtain second multi-head self-attention features;

[0159] Based on the multi-head self-attention mechanism corresponding to the multi-head attention network in the autoregressive decoder, performing multi-head attention processing on the second multi-head self-attention feature to obtain a multi-head attention feature;

[0160] Based on the second feature processing method corresponding to the feedforward neural network in the autoregressive decoder, feature transformation and feature extraction are performed on the multi-head attention features to obtain predicted driving text, and the predicted driving text is used as the driving intention data of the vehicle to be predicted.

[0161] In some embodiments of the present disclosure, the acquisition module 410 is specifically used to:

[0162] Obtaining user behavior guidance data, device state change data, and target application record data of the vehicle to be predicted;

[0163] Based on multiple data items in a preset instance format, respectively extracting data values ​​corresponding to the multiple data items from the user behavior guidance data, the device state change data, and the record data of the target application;

[0164] Filling the data values ​​respectively corresponding to the multiple data items into the preset instance format to obtain multiple event instances;

[0165] The multiple event instances are respectively configured with corresponding event identifiers to obtain a current event instance including the user behavior guidance event, the device state change event, and the target application's recorded event.

[0166] In some embodiments of the present disclosure, the forming module 420 is specifically used for:

[0167] Sorting the user behavior guidance event, the device state change event, and the target application record event included in the current event instance in chronological order to obtain a sorted current event instance;

[0168] According to the identification information respectively corresponding to each event instance in the sorted current event instance, each event instance in the sorted current event instance is spliced ​​one by one to form the current event map of the vehicle to be predicted.

[0169] In some embodiments of the present disclosure, the user behavior guidance event includes one or more combinations of the following: voice guidance event, gesture guidance event;

[0170] The device state change event includes one or more of the following combinations: a door state change event, a seat belt state change event, a light state change event, and a throttle state change event;

[0171] The recorded events of the target application include one or more of the following combinations: navigation recorded events, autonomous driving recorded events, environment recorded events, and user portrait recorded events.

[0172] In some embodiments of the present disclosure, the device further includes:

[0173] A historical event graph formation module is used to construct a graph of historical event instances of a reference vehicle within a preset time period before a historical moment to form a historical event graph of the reference vehicle, wherein the historical event instances include a user behavior guidance reference, a device state change reference, and a target application record reference;

[0174] A supervised fine-tuning module, configured to perform supervised training on the pre-trained base model based on the historical event graph and the reference driving intention of the reference vehicle to obtain a supervised fine-tuning model;

[0175] A training module, used for training the supervised fine-tuning model based on the evaluation statement corresponding to the reference driving intention, the historical event map and the reference driving intention to obtain a reward model;

[0176] The optimization module is used to optimize the supervised fine-tuning model using the reward model and the proximal strategy optimization method to obtain the target prediction model.

[0177] It should be noted that Figure 4 The driving intention prediction module 400 shown can execute Figure 1-Figure 3 The various steps in the method embodiment shown in the figure are implemented Figure 1-Figure 3 The various processes and effects in the method embodiment shown are not described in detail here.

[0178] Figure 5 A schematic diagram of the structure of a driving intention prediction device provided by an embodiment of the present disclosure is shown. The driving intention prediction device may be a controller of a target vehicle.

[0179] like Figure 5 As shown, the driving intention prediction device may include a controller 501 and a memory 502 storing computer program instructions.

[0180] Specifically, the controller 501 may include a central processing unit (CPU), or an application specific integrated circuit (ASIC), or may be configured to implement one or more integrated circuits of the embodiments of the present application.

[0181] The memory 502 may include a large capacity memory for information or instructions. By way of example and not limitation, the memory 502 may include a hard disk drive (HDD), a floppy disk drive, a flash memory, an optical disk, a magneto-optical disk, a tape, or a universal serial bus (USB) drive or a combination of two or more of these. Where appropriate, the memory 502 may include a removable or non-removable (or fixed) medium. Where appropriate, the memory 502 may be inside or outside the integrated gateway device. In a particular embodiment, the memory 502 is a non-volatile solid-state memory. In a particular embodiment, the memory 502 includes a read-only memory (ROM). Where appropriate, the ROM may be a mask-programmed ROM, a programmable ROM (PROM), an erasable PROM (EPROM), an electrically erasable PROM (Electrically Erasable Programmable ROM, EEPROM), an electrically rewritable ROM (EAROM) or a flash memory, or a combination of two or more of these.

[0182] The controller 501 reads and executes the computer program instructions stored in the memory 502 to perform the steps of the driving intention prediction method provided in the embodiment of the present disclosure.

[0183] In one example, the driving intention prediction device may further include a transceiver 503 and a bus 504. Figure 5 As shown, the controller 501, the memory 502 and the transceiver 503 are connected via a bus 504 and communicate with each other.

[0184] The bus 504 includes hardware, software, or both. For example, but not limitation, the bus may include an Accelerated Graphics Port (AGP) or other graphics bus, an Extended Industry Standard Architecture (EISA) bus, a Front Side Bus (FSB), a Hyper Transport (HT) interconnect, an Industrial Standard Architecture (ISA) bus, an InfiniBand interconnect, a Low Pin Count (LPC) bus, a memory bus, a MicroChannel Architecture (MCA) bus, a Peripheral Component Interconnect (PCI) bus, a PCI-Express (PCI-X) bus, a Serial Advanced Technology Attachment (SATA) bus, a Video Electronics Standards Association Local Bus (VLB) bus, or other suitable buses or a combination of two or more of these. Where appropriate, the bus 504 may include one or more buses. Although embodiments of the present application describe and illustrate a particular bus, the present application contemplates any suitable bus or interconnect.

[0185] The following is an embodiment of a computer-readable storage medium provided in an embodiment of the present disclosure. The computer-readable storage medium and the driving intention prediction method of the above-mentioned embodiments belong to the same inventive concept. For details not described in detail in the embodiment of the computer-readable storage medium, reference can be made to the embodiment of the above-mentioned driving intention prediction method.

[0186] This embodiment provides a storage medium containing computer executable instructions. When the computer executable instructions are executed by a computer processor, they are used to execute a driving intention prediction method. The method includes:

[0187] Obtaining a current event instance of the vehicle to be predicted, wherein the current event instance includes user behavior guidance events, device state change events, and target application record events at the current moment and within a preset time period before the current moment;

[0188] According to preset rules, a graph is constructed for the user behavior guidance event, the device state change event, and the recorded event of the target application included in the current event instance to form a current event graph of the vehicle to be predicted;

[0189] The current event graph is analyzed using the graph content analysis method included in the target prediction model to obtain the driving intention data of the vehicle to be predicted.

[0190] Of course, the storage medium containing computer executable instructions provided in the embodiment of the present disclosure, whose computer executable instructions are not limited to the above method operations, can also execute related operations in the driving intention prediction method provided in any embodiment of the present disclosure.

[0191] Through the above description of the implementation methods, technicians in the relevant field can clearly understand that the present disclosure can be implemented with the help of software and necessary general hardware, and of course it can also be implemented by hardware, but in many cases the former is a better implementation method. Based on this understanding, the technical solution of the present disclosure can essentially or contribute to the prior art in the form of a software product, which can be stored in a computer-readable storage medium, such as a computer floppy disk, read-only memory (ROM), random access memory (RAM), flash memory (FLASH), hard disk or optical disk, etc., including a number of instructions to enable a computer cloud platform (which can be a personal computer, server, or network cloud platform, etc.) to execute the driving intention prediction method provided by each embodiment of the present disclosure.

[0192] It should be noted that, in this article, relational terms such as "first" and "second" are only used to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Moreover, the term "comprising" is intended to cover non-exclusive inclusion, so that a process, method, article or device including a series of elements includes not only those elements, but also other elements not explicitly listed, or also includes elements inherent to such process, method, article or device.

[0193] The above are only specific embodiments of the present disclosure, so that those skilled in the art can understand or implement the present disclosure. Various modifications to these embodiments will be apparent to those skilled in the art, and the general principles defined herein can be implemented in other embodiments without departing from the spirit or scope of the present disclosure. Therefore, the present disclosure will not be limited to these embodiments herein, but will conform to the widest scope consistent with the principles and novel features disclosed herein.

Claims

1. A driving intention prediction method, characterized in that: include: Obtaining a current event instance of the vehicle to be predicted, wherein the current event instance includes user behavior guidance events, device state change events, and target application record events at the current moment and within a preset time period before the current moment; According to preset rules, a graph is constructed for the user behavior guidance event, the device state change event, and the recorded event of the target application included in the current event instance to form a current event graph of the vehicle to be predicted; The current event graph is analyzed using the graph content analysis method included in the target prediction model to obtain the driving intention data of the vehicle to be predicted.

2. The method according to claim 1, characterized in that The graph content parsing method includes a data encoding method and a data decoding method; the graph content parsing method using the target prediction model to parse the current event graph to obtain the driving intention data of the vehicle to be predicted includes: Based on the data encoding method corresponding to the transformer encoder in the target prediction model, the current event graph is encoded to obtain a plurality of encoding features corresponding to the current event graph; Based on the data decoding method corresponding to the autoregressive decoder in the target prediction model, a plurality of encoding features corresponding to the current event map are decoded to obtain the driving intention data of the vehicle to be predicted.

3. The method according to claim 2, characterized in that The data encoding method includes a first multi-head self-attention processing mechanism and a first feature processing method; the data encoding method corresponding to the transformer encoder in the target prediction model encodes the current event graph to obtain a plurality of encoding features corresponding to the current event graph, including: Based on a first multi-head self-attention processing mechanism corresponding to a first multi-head self-attention network in the transformer encoder, performing multi-head self-attention processing on the current event graph to obtain a first multi-head self-attention feature; Based on the first feature processing method corresponding to the feedforward neural network in the transformer encoder, feature transformation and feature extraction are performed on the first multi-head self-attention feature to obtain a plurality of encoding features corresponding to the current event map.

4. The method according to claim 2, characterized in that: The data decoding method includes a second multi-head self-attention processing mechanism, a multi-head attention mechanism and a second feature processing method; the data decoding method corresponding to the autoregressive decoder in the target prediction model decodes the multiple encoding features corresponding to the current event map to obtain the driving intention data of the vehicle to be predicted, including: Based on a second multi-head self-attention processing mechanism corresponding to the second multi-head self-attention network in the autoregressive decoder, multi-head self-attention processing is performed on the multiple encoding features to obtain second multi-head self-attention features; Performing multi-head attention processing on the second multi-head self-attention feature based on the multi-head attention mechanism corresponding to the multi-head attention network in the autoregressive decoder to obtain a multi-head attention feature; Based on the second feature processing method corresponding to the feedforward neural network in the autoregressive decoder, the multi-head attention feature is transformed and extracted to obtain a predicted driving text, and the predicted driving text is used as the driving intention data of the vehicle to be predicted.

5. The method according to claim 1, characterized in that The obtaining of the current event instance of the vehicle to be predicted includes: Obtaining user behavior guidance data, device state change data, and target application record data of the vehicle to be predicted; Based on multiple data items in a preset instance format, respectively extracting data values ​​corresponding to the multiple data items from the user behavior guidance data, the device state change data, and the record data of the target application; Filling the data values ​​respectively corresponding to the multiple data items into the preset instance format to obtain multiple event instances; The multiple event instances are respectively configured with corresponding event identifiers to obtain a current event instance including the user behavior guidance event, the device state change event, and the target application's recorded event.

6. The method according to claim 1, characterized in that The method of constructing a graph of the user behavior guidance event, the device state change event, and the target application record event contained in the current event instance according to a preset rule to form a current event graph of the vehicle to be predicted includes: Sorting the user behavior guidance event, the device state change event, and the target application record event included in the current event instance in chronological order to obtain a sorted current event instance; According to the identification information respectively corresponding to each event instance in the sorted current event instance, each event instance in the sorted current event instance is spliced ​​one by one to form the current event map of the vehicle to be predicted.

7. The method according to claim 1, characterized in that The user behavior guidance event includes one or more of the following combinations: voice guidance event, gesture guidance event; The device state change event includes one or more of the following combinations: a door state change event, a seat belt state change event, a light state change event, and a throttle state change event; The recorded events of the target application include one or more of the following combinations: navigation recorded events, autonomous driving recorded events, environment recorded events, and user portrait recorded events.

8. The method according to claim 1, characterized in that Before obtaining the current event instance of the vehicle to be predicted, the method further includes: Constructing a graph of historical event instances of a reference vehicle within a preset time period before a historical moment to form a historical event graph of the reference vehicle, wherein the historical event instances include a user behavior guidance reference, a device state change reference, and a target application record reference; Based on the historical event graph and the reference driving intention of the reference vehicle, supervised training is performed on the pre-trained base model to obtain a supervised fine-tuning model; Based on the evaluation statement corresponding to the reference driving intention, the historical event map and the reference driving intention, the supervised fine-tuning model is trained to obtain a reward model; The supervised fine-tuning model is optimized using the reward model and the proximal strategy optimization method to obtain the target prediction model.

9. A driving intention prediction device, characterized in that: include: An acquisition module, used to acquire a current event instance of the vehicle to be predicted, wherein the current event instance includes a user behavior guidance event, a device state change event, and a target application record event at the current moment and within a preset time period before the current moment; A formation module, configured to construct a graph for the user behavior guidance event, the device state change event, and the target application record event contained in the current event instance according to preset rules, so as to form a current event graph of the vehicle to be predicted; The prediction module is used to analyze the current event graph using the graph content analysis method included in the target prediction model to obtain the driving intention data of the vehicle to be predicted.

10. A driving intention prediction device, characterized in that: include: processor; A memory for storing executable instructions; The processor is used to read the executable instructions from the memory and execute the executable instructions to implement the method described in any one of claims 1 to 8.

11. A computer-readable storage medium, characterized in that: The storage medium stores a computer program, and when the computer program is executed by a processor, the processor implements the method according to any one of claims 1 to 7.

12. A vehicle, characterized in that: include: The driving intention prediction device as claimed in claim 9; or The driving intention prediction device as claimed in claim 10; or The computer readable storage medium of claim 11.

Images