Information providing method and device, computer device and storage medium

Abstract

The application relates to an information providing method and device, computer equipment and a storage medium. The method comprises: obtaining information for prediction corresponding to a target time, wherein the information for prediction comprises: a plurality of first vehicle parameters of a target vehicle at the target time, a road surface image related to the target time, and the road surface image comprises a road surface object representing a target road surface; using a target neural network, predicting a prediction result related to the target road surface according to the information for prediction, wherein the prediction result comprises: a road surface adhesion coefficient of the target road surface, a bump degree of the target road surface, and a road surface slope of the target road surface; and providing each item in the prediction result to a corresponding system of the target vehicle respectively, so that the corresponding system is adjusted according to the corresponding item in the prediction result.

Description

TECHNICAL FIELD

[0001] The present application relates to the field of vehicles, in particular to an information providing method and device, computer equipment and storage medium. BACKGROUND

[0002] It is an important function of an auxiliary driving system to predict the related information of a road surface and provide the related information of the road surface to a system of a vehicle so that the corresponding system of the vehicle adjusts according to the related information of the road surface.

[0003] In the related art, the related information of the road surface provided to the system of the vehicle is the type of the road surface. However, the road conditions of a road surface of a type of road surface are divided into various road conditions. For example, the road conditions of a road surface of a snow-covered road surface type are divided into deep snow road conditions, shallow snow road conditions, dry snow road conditions, and the like. The type of the road surface on which the vehicle travels cannot reflect the road conditions of the road surface on which the vehicle travels, and the accuracy of the related information of the road surface provided to the corresponding system of the vehicle is low. The low accuracy of the related information of the road surface on which the vehicle travels provided to the corresponding system of the vehicle makes it difficult for the corresponding system of the vehicle to make a relatively accurate corresponding adjustment. How to improve the accuracy of the related information of the road surface on which the vehicle travels provided to the system of the vehicle becomes a problem to be solved. SUMMARY

[0004] One of the purposes of the present application is to provide an information providing method and device, computer equipment and storage medium to solve the problem of how to improve the accuracy of the related information provided to the system of the vehicle.

[0005] In order to achieve the above-mentioned purpose, the technical solution adopted by the present application is as follows:

[0006] An information providing method, the method comprising:

[0007] obtaining information for prediction corresponding to a target time, the information for prediction comprising: a plurality of first vehicle parameters of a target vehicle at the target time, a road surface image related to the target time, the road surface image comprising a road surface object representing a target road surface;

[0008] using a target neural network to predict a prediction result related to the target road surface according to the information for prediction, the prediction result comprising: a road surface adhesion coefficient of the target road surface, a jolt degree of the target road surface, and a road surface slope of the target road surface;

[0009] providing each item in the prediction result to a corresponding system of the target vehicle respectively, so that the corresponding system adjusts according to the corresponding item in the prediction result.

[0010] Further, providing each of the prediction results to a corresponding system of the target vehicle respectively, so that the corresponding system adjusts according to the corresponding one of the prediction results, comprises:

[0011] providing the road adhesion coefficient of the target road to a traction control system of the target vehicle, so that the traction control system of the target vehicle adjusts the output power of the engine of the target vehicle according to the road adhesion coefficient of the target road;

[0012] providing the bumpiness of the target road to a suspension system of the target vehicle, so that the suspension system of the target vehicle adjusts the damping rate according to the bumpiness of the target road;

[0013] providing the road slope of the target road to a torque distribution system of the target vehicle, so that the torque distribution system of the target vehicle adjusts the engine torque of the target vehicle according to the road slope of the target road.

[0014] Further, predicting, by using a target neural network, prediction results related to the target road according to the information for prediction, comprises:

[0015] generating a hidden state vector of each of the plurality of first vehicle parameters, and extracting a road image feature vector of the road image; generating a target hidden state vector according to the hidden state vector of each of the plurality of first vehicle parameters and the hidden state vector of each of the plurality of second vehicle parameters of the target vehicle at the last time of the target time; generating a feature for prediction according to the target hidden state vector and the road image feature vector, and predicting the prediction results according to the feature for prediction.

[0016] Further, generating a target hidden state vector according to the hidden state vector of each of the plurality of first vehicle parameters and the hidden state vector of each of the plurality of second vehicle parameters of the target vehicle at the last time of the target time comprises:

[0017] determining an attention parameter of each of the plurality of first vehicle parameters according to the hidden state vector of each of the plurality of first vehicle parameters;

[0018] generating an attention hidden state vector of each of the plurality of first vehicle parameters according to the hidden state vector of each of the plurality of first vehicle parameters and the attention parameter of each of the plurality of first vehicle parameters;

[0019] generating a target hidden state vector according to the attention hidden state vector of each of the plurality of first vehicle parameters and the hidden state vector of each of the plurality of second vehicle parameters.

[0020] Further, determining the attention parameter of each first vehicle parameter according to the hidden state vector of the each first vehicle parameter comprises:

[0021] Determining a first attention score of each first vehicle parameter according to the hidden state vector of the each first vehicle parameter;

[0022] For the each first vehicle parameter, determining a second attention score of the first vehicle parameter according to the hidden state vector of the first vehicle parameter and the hidden state vector of the second vehicle parameter corresponding to the first vehicle parameter, wherein the type of the first vehicle parameter is the same as the type of the second vehicle parameter corresponding to the first vehicle parameter;

[0023] Determining the attention parameter of the each first vehicle parameter according to the first attention score of the each first vehicle parameter and the second attention score of the each first vehicle parameter.

[0024] Further, generating the target hidden state vector according to the attention hidden state vector of the each first vehicle parameter and the hidden state vector of the each second vehicle parameter comprises:

[0025] Generating the attention hidden state vector of the each second vehicle parameter according to the hidden state vector of the each second vehicle parameter and the attention parameter of the each second vehicle parameter;

[0026] Generating the target hidden state vector according to the attention hidden state vector of the each first vehicle parameter and the attention hidden state vector of the each second vehicle parameter.

[0027] Further, generating the target hidden state vector according to the attention hidden state vector of the each first vehicle parameter and the attention hidden state vector of the each second vehicle parameter comprises:

[0028] Generating a first vehicle parameter combination vector comprising the attention hidden state vector of the each first vehicle parameter, and generating a second vehicle parameter combination vector comprising the attention hidden state vector of the each second vehicle parameter;

[0029] Adding the first vehicle parameter combination vector and the second vehicle parameter combination vector to obtain the target hidden state vector.

[0030] Further, generating the feature for prediction according to the target hidden state vector and the road image feature vector comprises:

[0031] multiply the target hidden state vector with a vehicle parameter feature weight to obtain a first weighted vector, and multiply the road surface image feature vector with a road surface image feature weight to obtain a second weighted vector;

[0032] generate a feature for prediction according to the first weighted vector and the second weighted vector.

[0033] Further, generating a feature for prediction according to the first weighted vector and the second weighted vector comprises:

[0034] adding the first weighted vector and the second weighted vector to obtain a feature for prediction.

[0035] Further, before obtaining the information for prediction corresponding to the target moment, the method further comprises:

[0036] determining the vehicle parameter feature weight and the road surface image feature weight according to an accuracy of the vehicle parameter feature extraction network in the target neural network on a validation set of the target neural network and an accuracy of the road surface image feature extraction network in the target neural network on the validation set.

[0037] According to the above technical means, the road conditions of the same type of road surface are considered to be divided into multiple types, and different road conditions of the same type of road surface have different road adhesion coefficients, jolts, and road slopes. For example, the road conditions of the snow-covered road surface type are divided into deep snow road surface conditions, shallow snow road surface conditions, and dry snow road surface conditions. The deep snow road conditions, the shallow snow road conditions, and the dry snow road conditions have different road adhesion coefficients, and the deep snow road conditions, the shallow snow road conditions, and the dry snow road conditions have different jolts. The predicted target road adhesion coefficient, the target road jolt, and the target road slope of the target road surface are related to the road condition of the target road surface. Among them, the target road adhesion coefficient can describe the friction between the tire and the target road surface, the target road jolt can reflect the flatness or vibration degree of the target road surface, and the target road slope can reflect the angle of inclination of the target road surface. The accuracy of the related information provided to the system of the vehicle is relatively high. The target road adhesion coefficient, the target road jolt, and the target road slope related to the road condition of the target road surface are respectively provided to the corresponding system of the target vehicle, so that the corresponding system of the target vehicle can adjust according to the information related to the road condition of the target road surface, and the adjustment accuracy is relatively high. The target road adhesion coefficient, the target road jolt, and the target road slope of the target road surface can be provided to the chassis system of the target vehicle, so that the chassis system of the target vehicle adjusts according to the target road adhesion coefficient, the target road jolt, and the target road slope of the target road surface, improves the dynamic stability, safety, and comfort of the target vehicle.

[0038] Different road conditions of the same type of road have different road adhesion coefficients, bump degrees, and road slopes, and the corresponding system is adjusted according to at least one of the road adhesion coefficient, the bump degree, and the road slope, so that the information providing method provided in the embodiment of the present application adjusts the corresponding system in different adjustment modes for different road conditions of the same type of road, and the degree of refinement of the adjustment of the corresponding system in the embodiment of the present application is higher than that of the related art in which the corresponding system is adjusted according to the same type of road for different road conditions of the same type of road.

[0039] Since various road conditions of various types of roads have corresponding road adhesion coefficients, bump degrees, and road slopes, at least one of the road adhesion coefficient, the bump degree, and the road slope can be provided to the corresponding system for various types of roads, the information providing method provided in the embodiment of the present application has strong applicability, and it can also be said that the information providing method provided in the embodiment of the present application covers a wider range of road conditions. The information providing method provided in the embodiment of the present application can avoid the following situation: inaccurate identification of the road type, for example, the target road type is complex, which leads to inaccurate identification of the road type, the target road is a composite road condition of part of the water and part of the silt, which leads to inaccurate identification of the road type, and the corresponding system is only adjusted according to the road type, which leads to low accuracy of the adjustment.

[0040] An information providing device comprises:

[0041] An acquisition unit is configured to acquire information for prediction corresponding to a target time, wherein the information for prediction comprises: a plurality of first vehicle parameters of a target vehicle at the target time, and a road image related to the target time, wherein the road image comprises a road object representing a target road;

[0042] A prediction unit is configured to predict a prediction result related to the target road by using a target neural network according to the information for prediction, wherein the prediction result comprises: a road adhesion coefficient of the target road, a bump degree of the target road, and a road slope of the target road.

[0043] A providing unit is configured to provide each item in the prediction result to a corresponding system of the target vehicle respectively, so that the corresponding system performs corresponding adjustment according to the corresponding item in the prediction result.

[0044] Further, the providing unit is further configured to provide each of the prediction results to a corresponding system of the target vehicle respectively, so that the corresponding system adjusts according to the corresponding prediction result, including: providing the road surface adhesion coefficient of the target road surface to a traction control system of the target vehicle, so that the traction control system of the target vehicle adjusts the output power of the engine of the target vehicle according to the road surface adhesion coefficient of the target road surface; providing the bump degree of the target road surface to a suspension system of the target vehicle, so that the suspension system of the target vehicle adjusts the damping rate according to the bump degree of the target road surface; and providing the road surface slope of the target road surface to a torque distribution system of the target vehicle, so that the torque distribution system of the target vehicle adjusts the engine torque of the target vehicle according to the road surface slope of the target road surface.

[0045] Further, the prediction unit is further configured to generate a hidden state vector of each of the plurality of first vehicle parameters, and extract a road surface image feature vector of the road surface image; generate a target hidden state vector according to the hidden state vector of each of the plurality of first vehicle parameters and the hidden state vector of each of the plurality of second vehicle parameters of the target vehicle at the last time point of the target time point; generate a feature for prediction according to the target hidden state vector and the road surface image feature vector, and predict the prediction result according to the feature for prediction.

[0046] Further, the prediction unit is further configured to determine an attention parameter of each of the plurality of first vehicle parameters according to the hidden state vector of each of the plurality of first vehicle parameters; generate an attention hidden state vector of each of the plurality of first vehicle parameters according to the hidden state vector of each of the plurality of first vehicle parameters and the attention parameter of each of the plurality of first vehicle parameters; and generate a target hidden state vector according to the attention hidden state vector of each of the plurality of first vehicle parameters and the hidden state vector of each of the plurality of second vehicle parameters.

[0047] Further, the prediction unit is further configured to determine a first attention score of each of the plurality of first vehicle parameters according to the hidden state vector of each of the plurality of first vehicle parameters; for each of the plurality of first vehicle parameters, determine a second attention score of the first vehicle parameter according to the hidden state vector of the first vehicle parameter and the hidden state vector of the second vehicle parameter corresponding to the first vehicle parameter, wherein the type of the first vehicle parameter is the same as the type of the second vehicle parameter corresponding to the first vehicle parameter; and determine an attention parameter of each of the plurality of first vehicle parameters according to the first attention score of each of the plurality of first vehicle parameters and the second attention score of each of the plurality of first vehicle parameters.

[0048] Further, the prediction unit is further configured to generate an attention hidden state vector of each second vehicle parameter according to the hidden state vector of the each second vehicle parameter, the attention parameter of the each second vehicle parameter; and generate a target hidden state vector according to the attention hidden state vector of the each first vehicle parameter, the attention hidden state vector of the each second vehicle parameter.

[0049] Further, the prediction unit is further configured to generate a first vehicle parameter combination vector including the attention hidden state vector of the each first vehicle parameter, and generate a second vehicle parameter combination vector including the attention hidden state vector of the each second vehicle parameter; and add the first vehicle parameter combination vector and the second vehicle parameter combination vector to obtain the target hidden state vector.

[0050] Further, the prediction unit is further configured to multiply the target hidden state vector by a vehicle parameter feature weight to obtain a first weighted vector, and multiply the road surface image feature vector by a road surface image feature weight to obtain a second weighted vector; and generate a feature for prediction according to the first weighted vector and the second weighted vector.

[0051] Further, the prediction unit is further configured to add the first weighted vector and the second weighted vector to obtain the feature for prediction.

[0052] Further, the information providing device comprises:

[0053] The weight determination unit is configured to determine a vehicle parameter feature weight and a road surface image feature weight according to an accuracy of a vehicle parameter feature extraction network in the target neural network on a validation set of the target neural network and an accuracy of a road surface image feature extraction network in the target neural network on the validation set before obtaining the information for prediction corresponding to a target moment.

[0054] A computer device comprises:

[0055] The memory and the processor are connected with each other in communication, the memory stores computer instructions, and the processor executes the computer instructions to perform the above method.

[0056] A computer readable storage medium stores computer instructions, and the computer instructions are used to make a computer execute the above method.

[0057] A computer program product comprises computer instructions, and the computer instructions are used to make a computer execute the above method.

[0058] The beneficial effects of the present application are as follows:

[0059] With the target neural network, the road surface adhesion coefficient of the target road surface, the bump degree of the target road surface, and the road surface slope of the target road surface required for corresponding adjustment of the corresponding system of the target vehicle are directly predicted according to the plurality of first vehicle parameters of the target vehicle at the target time and the road surface image related to the target time. The road surface adhesion coefficient of the target road surface, the bump degree of the target road surface, and the road surface slope of the target road surface required for corresponding adjustment of the corresponding system of the target vehicle are directly provided to the corresponding system of the target vehicle, so that the corresponding system of the target vehicle can be adjusted according to the road surface adhesion coefficient of the target road surface, the bump degree of the target road surface, and the road surface slope of the target road surface, so that the driving and riding feelings of the personnel in the target vehicle are more comfortable when the target vehicle travels on the target road surface.

[0060] The information providing method provided by the embodiment of the present application considers that the road conditions of the same type of road surface are divided into multiple types, and the different road conditions of the same type of road surface have different road surface adhesion coefficients, bump degrees, and road surface slopes. For example, the road conditions of the snow-covered road surface type are divided into deep snow road surface conditions, shallow snow road surface conditions, and dry snow road surface conditions. The deep snow road surface conditions, the shallow snow road surface conditions, and the dry snow road surface conditions have different road surface adhesion coefficients, and the deep snow road surface conditions, the shallow snow road surface conditions, and the dry snow road surface conditions have different bump degrees. The road surface adhesion coefficient of the target road surface, the bump degree of the target road surface, and the road surface slope of the target road surface predicted by the present application are related to the road conditions of the target road surface. Among them, the road surface adhesion coefficient of the target road surface can describe the friction between the tire and the target road surface, the bump degree of the target road surface can reflect the flatness or vibration degree of the target road surface, and the road surface slope of the target road surface can reflect the angle of inclination of the target road surface. The accuracy of the related information provided to the system of the vehicle is high. The road surface adhesion coefficient of the target road surface, the bump degree of the target road surface, and the road surface slope of the target road surface related to the road conditions of the target road surface are respectively provided to the corresponding system of the target vehicle, so that the corresponding system of the target vehicle can be adjusted according to the information related to the road conditions of the target road surface, and the adjustment accuracy is high. The road surface adhesion coefficient of the target road surface, the bump degree of the target road surface, and the road surface slope of the target road surface can be provided to the chassis system of the target vehicle, so that the chassis system of the target vehicle can be adjusted according to the road surface adhesion coefficient of the target road surface, the bump degree of the target road surface, and the road surface slope of the target road surface, and the dynamic stability, safety, and comfort of the target vehicle are improved.

[0061] Different road conditions of the same type of road have different road adhesion coefficients, bumpiness degrees and road slopes, and the corresponding system is adjusted according to at least one of the road adhesion coefficients, the bumpiness degrees and the road slopes, so that the information providing method provided in the embodiment of the present application adjusts the corresponding system in different adjustment modes for different road conditions of the same type of road, and the degree of refinement of the adjustment of the corresponding system in the embodiment of the present application is higher than that of the related art in which the corresponding system is adjusted according to the same type of road for different road conditions of the same type of road.

[0062] Since various road conditions of various types of roads have corresponding road adhesion coefficients, bumpiness degrees and road slopes, at least one of the road adhesion coefficients, the bumpiness degrees and the road slopes can be provided to the corresponding system for various types of roads, the information providing method provided in the embodiment of the present application has strong applicability, and can be said to cover a wider range of road conditions. The information providing method provided in the embodiment of the present application can avoid the situation that the accuracy of identifying the road type is inaccurate, for example, the road type of the target road is complex, which leads to inaccurate identification of the road type, the target road is a composite road condition of part of water and part of silt, which leads to inaccurate identification of the road type, and the accuracy of the adjustment of the corresponding system according to the road type is low.

[0063] In the embodiment of the present application, the road adhesion coefficient of the target road, the bumpiness degree of the target road and the road slope of the target road required for directly adjusting the corresponding system of the target vehicle are provided to the corresponding system of the target vehicle. Thus, the corresponding system of the target vehicle does not need to convert the information provided to the corresponding system of the target vehicle into the corresponding parameters required for the corresponding adjustment before the corresponding adjustment, and the corresponding system of the target vehicle directly adjusts according to the corresponding item of the road adhesion coefficient of the target road, the bumpiness degree of the target road and the road slope of the target road. Thus, the efficiency of the corresponding adjustment of the corresponding system of the target vehicle is improved, the overall control efficiency of the target vehicle is improved, and the target vehicle quickly reaches an optimal state. For example, the bumpiness degree of the target road is directly provided to the suspension system of the target vehicle, and the suspension system of the target vehicle can directly adjust according to the bumpiness degree of the target road to provide a more stable driving experience. The road adhesion coefficient of the target road is directly provided to the power distribution system of the target vehicle, and the power distribution system of the target vehicle can directly adjust the driving force according to the road adhesion coefficient of the target road to ensure that the target vehicle maintains good traction and stability. BRIEF DESCRIPTION OF DRAWINGS

[0064] Figure 1 The flowchart of the information providing method provided in the embodiment of the present application is shown in the figure;

[0065] Figure 2 FIG. 1 shows a flowchart of one example of a method for predicting a prediction result related to a target road by combining a hidden state vector of vehicle parameters and a road image feature vector of a road image related to a target time;

[0066] Figure 3 FIG. 2 shows a schematic diagram of one example structure of a target neural network;

[0067] Figure 4 FIG. 3 shows a flowchart of one example of an information providing method provided by an embodiment of the present application.

[0068] Figure 5 FIG. 4 shows a flowchart of one example of generating features for prediction.

[0069] Figure 6 FIG. 5 shows a schematic diagram of a hardware structure of a computer device provided by an embodiment of the present application. DETAILED DESCRIPTION

[0070] Other advantages and effects of the present application can be easily understood by those skilled in the art from the above description of the embodiments of the present application. The present application can also be implemented or applied in other different embodiments, and various modifications or changes can be made to the details of the present application based on different views and applications without departing from the spirit of the present application. It should be understood that the preferred embodiments are only for illustrating the present application, and are not intended to limit the protection scope of the present application.

[0071] It should be noted that the diagrams provided in the following embodiments only schematically illustrate the basic concepts of the present application, and thus only show the components related to the present application in the diagrams, but are not drawn according to the number, shape and size of the components in actual implementation. The shapes, number and proportions of the components in actual implementation can be arbitrarily changed, and the layout pattern of the components can also be more complex.

[0072] Reference Figure 1 FIG. 1 shows a flowchart of one example of an information providing method provided by an embodiment of the present application. It should be noted that steps S101-S102 can be repeatedly performed during driving of a target vehicle.

[0073] In step S101, information for prediction corresponding to a target time is acquired.

[0074] The information for prediction corresponding to the target time includes: a plurality of first vehicle parameters of the target vehicle at the target time, a road image related to the target time.

[0075] The road image related to the target time includes: a road object representing a target road.

[0076] The target road surface can be a road surface on which the target vehicle travels at the target moment.

[0077] In a possible implementation, the plurality of first vehicle parameters of the target vehicle at the target moment include, but are not limited to, a wheel inertia torque of the target vehicle at the target moment, a wheel longitudinal force of the target vehicle at the target moment, a suspension height of the target vehicle at the target moment, a body vertical acceleration of the target vehicle at the target moment, and a yaw angular velocity of the target vehicle at the target moment. That is, the wheel inertia torque of the target vehicle at the target moment, the wheel longitudinal force of the target vehicle at the target moment, the suspension height of the target vehicle at the target moment, the body vertical acceleration of the target vehicle at the target moment, and the yaw angular velocity of the target vehicle at the target moment are respectively taken as the first vehicle parameters of the target vehicle at the target moment.

[0078] In the embodiment of the present application, the first vehicle parameter of the target vehicle at the target moment is collected at the target moment. That is, the collection moment of the first vehicle parameter of the target vehicle at the target moment is the target moment.

[0079] As an example, the plurality of first vehicle parameters of the target vehicle at the target moment include a wheel inertia torque of the target vehicle at the target moment, a wheel longitudinal force of the target vehicle at the target moment, a suspension height of the target vehicle at the target moment, a body vertical acceleration of the target vehicle at the target moment, and a yaw angular velocity of the target vehicle at the target moment.

[0080] In the embodiment of the present application, the target moment and the moment before the target moment are both adjacent two moments for prediction among all moments for prediction.

[0081] In a possible implementation, the moment for prediction is a moment at which a time interval between the moment for prediction and a moment at which the target vehicle starts is a preset time interval or a multiple of the preset time interval.

[0082] In a possible implementation, the road surface image related to the target moment can be a road surface image with a collection moment closest to the target moment.

[0083] In a possible implementation, in order to make the collection moment of the road surface image more accurately correspond to the collection moment of the vehicle parameter of the target vehicle, for the vehicle parameter i of the target vehicle, the collection moment of the vehicle parameter i of the target vehicle can be obtained by correcting an original collection moment of the vehicle parameter i of the target vehicle. The vehicle parameter i of the target vehicle can be any parameter collected by the target vehicle. image The collection moment of the road surface image with the collection moment closest to the original collection moment of the vehicle parameter i, t sensoris a raw collection time of the vehicle parameter i for collecting the sensor output of the vehicle parameter i. v is a vehicle speed of the target vehicle at the raw collection time of the vehicle parameter i, a collection time of the vehicle parameter i of the target vehicle a collection time t of the vehicle parameter i of the target vehicle adjusted a collection time t of the road surface image closest to the collection time of the vehicle parameter i image closer.

[0084] In step S102, a prediction result related to the target road surface is predicted by the target neural network according to the information for prediction corresponding to the target time.

[0085] The prediction result related to the target road surface includes but is not limited to the road surface adhesion coefficient of the target road surface, the bump degree of the target road surface, and the road surface slope of the target road surface.

[0086] In step S102, the information for prediction corresponding to the target time is input to the target neural network, and the target neural network outputs the road surface adhesion coefficient of the target road surface, the bump degree of the target road surface, and the road surface slope of the target road surface.

[0087] The target neural network can be any neural network that can perform a regression task.

[0088] As an example, the target neural network is a convolutional neural network or a deep neural network.

[0089] It should be noted that the target neural network is trained. Before step S101, the target neural network is trained by using a training data set of the target neural network.

[0090] During training of the target neural network, the parameters of the target neural network are updated. In order to obtain the training data set of the target neural network, the vehicle for training drives on each of a plurality of road surfaces such as snow, mud, rock, sand, asphalt, cement, wading, etc.

[0091] For each of the plurality of road surfaces, during driving of the vehicle for training on the road surface, the vehicle for training collects data of a type to which the data in the information for prediction corresponding to the target time belongs. According to the data collected during driving of the vehicle for training on the road surface, the training data of the road surface is determined. The training data of each of the plurality of road surfaces constitutes the training data set of the target neural network.

[0092] As an example, the vehicle for training collects the inertial torque of the wheel of the vehicle for training, the wheel longitudinal force of the vehicle for training, the suspension height of the vehicle for training, the body vertical acceleration of the vehicle for training, the lateral angular velocity of the vehicle for training, and other vehicle end parameters, and collects the road image including the road surface object representing the road surface on which the vehicle for training travels.

[0093] The training data in the training data set of the target neural network includes: a plurality of first vehicle parameters of the target vehicle at the training time, and a road image related to the training time.

[0094] The training data in the training data set of the target neural network is labeled to obtain a labeling result of the training data of the target neural network. The labeling result of the training data of the target neural network is a result expected to be output by the target neural network when the training data is input into the target neural network.

[0095] The labeling result of the training data includes: a labeled road adhesion coefficient, a labeled road bump degree, and a labeled road slope.

[0096] During training of the target neural network, the training data in the training data set of the target neural network is input into the target neural network, and the target neural network outputs a prediction result corresponding to the training data. A loss between the prediction result corresponding to the training data and the labeling result of the training data is calculated, and the loss is used to update the parameters of the target neural network.

[0097] In one possible implementation, the step S102 includes steps S1021-S1023. The steps S1021-S1023 combine the hidden state vector of the vehicle parameter and the road image feature vector of the road image related to the target time to predict the prediction result related to the target road.

[0098] The vehicle parameters of the target vehicle, the road surface image related to the target moment, and the road surface image feature vector of the road surface image related to the target moment are combined to predict the prediction result related to the target road surface. The prediction result related to the target road surface is predicted by using the vehicle parameters of the target vehicle, the road surface image related to the target moment, and the road surface image feature vector of the road surface image related to the target moment. The richness of the information used to predict the road surface on which the vehicle travels is improved. The prediction result related to the target road surface is predicted by using the vehicle parameters of the target vehicle, the road surface image related to the target moment, and the road surface image feature vector of the road surface image related to the target moment. The accuracy of the prediction result related to the target road surface is high. Specifically, the prediction result related to the road surface on which the vehicle travels is predicted by using the rich features, i.e., the hidden state vectors of the vehicle parameters of the target vehicle and the road surface image feature vectors of the road surface image related to the target moment. The accuracy of the prediction result related to the target road surface is high. Thus, the accuracy of the prediction result related to the target road surface is improved, and the accuracy of the corresponding system of the target vehicle is improved.

[0099] Reference Figure 2 The flowchart shows an example of predicting the prediction result related to the target road surface by combining the hidden state vectors of the vehicle parameters and the road surface image feature vectors of the road surface image related to the target moment.

[0100] In this example, the information used for prediction corresponding to the target moment is input into the target neural network. The information used for prediction corresponding to the target moment includes: the plurality of first vehicle parameters of the target vehicle at the target moment, and the road surface image related to the target moment.

[0101] In this example, the target neural network generates the hidden state vector of each of the plurality of first vehicle parameters of the target vehicle at the target moment. The target neural network extracts the road surface image feature vector of the road surface image related to the target moment.

[0102] In this example, the target neural network generates the target hidden state vector according to the hidden state vectors of the plurality of first vehicle parameters of the target vehicle at the target moment and the hidden state vectors of each of the plurality of second vehicle parameters of the target vehicle at the moment before the target moment.

[0103] It should be noted that the hidden state vector of the plurality of second vehicle parameters of the target vehicle at the target time is generated when predicting the prediction result related to the target road surface corresponding to the time before the target time.

[0104] In this example, the target neural network generates a feature for prediction according to the target hidden state vector, the road surface image feature vector of the road surface image related to the target time.

[0105] The prediction result related to the target road surface includes: the road surface adhesion coefficient of the target road surface, the road surface roughness of the target road surface, and the road surface slope of the target road surface.

[0106] Reference Figure 3 which shows a schematic diagram of one example structure of the target neural network.

[0107] The target neural network includes: a vehicle parameter feature extraction network, a road surface image feature extraction network, a feature generation unit for prediction, and a prediction unit.

[0108] The type of the vehicle parameter feature extraction network is: Long Short-Term Memory (LSTM). The type of the road surface image feature extraction network is: convolutional neural network.

[0109] The vehicle parameter feature extraction network receives the corresponding parameter at the target time. The vehicle parameter feature extraction network outputs the hidden state vector of the corresponding parameter at the corresponding time.

[0110] The road surface image feature extraction network receives the corresponding road surface image and outputs the road surface image feature vector of the corresponding road surface image.

[0111] The feature generation unit for prediction generates a feature for prediction according to the hidden state vector output by the vehicle parameter feature extraction network and the road surface image feature vector output by the road surface image feature extraction network.

[0112] The prediction unit can include: a fully connected layer for predicting the road surface adhesion coefficient, the roughness, and the road surface slope. The feature for prediction is input into the fully connected layer for predicting the road surface adhesion coefficient, the roughness, and the road surface slope. The fully connected layer for predicting the road surface adhesion coefficient, the roughness, and the road surface slope performs linear transformation on the feature vector for prediction through a weight matrix and a bias vector. The fully connected layer for predicting the road surface adhesion coefficient, the roughness, and the road surface slope has three output nodes, which are respectively an output node for outputting the road surface adhesion coefficient, an output node for outputting the road surface roughness, and an output node for outputting the road surface slope.

[0113] In step S1021, a hidden state vector of each of the plurality of first vehicle parameters of the target vehicle at the target time is generated, and a road surface image feature vector of the road surface image related to the target time is extracted.

[0114] In step S1021, the hidden state vector of each of the plurality of first vehicle parameters of the target vehicle at the target time is generated by the vehicle parameter feature extraction network in the target neural network.

[0115] It should be noted that the hidden state vector of the plurality of second vehicle parameters of the target vehicle at the last time before the target time is generated when predicting the prediction result of the road surface related to the representation of the road surface object in the road surface image related to the last time before the target time. The prediction result of the road surface related to the representation of the road surface object in the road surface image related to the last time before the target time is generated before step S301. The process of generating the hidden state vector of the plurality of second vehicle parameters of the target vehicle at the last time before the target time is the same as the process of generating the hidden state vector of each of the plurality of first vehicle parameters of the target vehicle at the target time.

[0116] The second vehicle parameter of the target vehicle at the last time before the target time is collected at the last time before the target time, that is, the collection time of the second vehicle parameter of the target vehicle at the last time before the target time is the last time before the target time.

[0117] In the embodiment of the present application, the number of first vehicle parameters included in the plurality of first vehicle parameters is equal to the number of second vehicle parameters included in the plurality of second vehicle parameters.

[0118] The plurality of first vehicle parameters and the plurality of second vehicle parameters correspond one-to-one.

[0119] For one of the plurality of first vehicle parameters, the type of the first vehicle parameter is the same as the type of the second vehicle parameter corresponding to the first vehicle parameter.

[0120] As an example, the plurality of first vehicle parameters of the target vehicle at the target time includes: an inertial torque of a wheel of the target vehicle at the target time, a wheel longitudinal force of the target vehicle at the target time, a suspension height of the target vehicle at the target time, a body vertical acceleration of the target vehicle at the target time, a yaw angular velocity of the target vehicle at the target time. The plurality of second vehicle parameters of the target vehicle at the target time at a time before the target time includes: an inertial torque of a wheel of the target vehicle at the time before the target time, a wheel longitudinal force of the target vehicle at the time before the target time, a suspension height of the target vehicle at the time before the target time, a body vertical acceleration of the target vehicle at the time before the target time, a yaw angular velocity of the target vehicle at the time before the target time. The inertial torque of the wheel of the target vehicle at the target time corresponds to the inertial torque of the wheel of the target vehicle at the time before the target time, the wheel longitudinal force of the target vehicle at the target time corresponds to the wheel longitudinal force of the target vehicle at the time before the target time, the suspension height of the target vehicle at the target time corresponds to the suspension height of the target vehicle at the time before the target time, the body vertical acceleration of the target vehicle at the target time corresponds to the body vertical acceleration of the target vehicle at the time before the target time, and the yaw angular velocity of the target vehicle at the target time corresponds to the yaw angular velocity of the target vehicle at the time before the target time.

[0121] In step S1021, for each of the plurality of first vehicle parameters of the target vehicle at the target time, the vehicle parameter feature extraction network receives the first vehicle parameter, and outputs a hidden state vector of the first vehicle parameter.

[0122] The type of the vehicle parameter feature extraction network is a long short-term memory neural network.

[0123] In one possible implementation, for a first vehicle parameter of the target vehicle at the target time, an input layer of the target neural network encodes the first vehicle parameter to obtain an embedding vector of the first vehicle parameter; the embedding vector of the first vehicle parameter is taken as an input of a hidden unit in the vehicle parameter feature extraction network for generating a hidden state vector of a parameter of a type to which the first vehicle parameter belongs; and the hidden unit for generating the hidden state vector of the parameter of the type outputs the hidden state vector of the first vehicle parameter.

[0124] The generation of the hidden state vector of the first vehicle parameter can be represented as:

[0125] h k =[σ(S o ·s k +S o ·h k-1 +bo )]⊙tanh(c k )

[0126] c k =[σ(S f ·s k +S f ·h k-1 +b f )]⊙c k-1 +[σ(S i ·s k +S i ·h k-1 +b i )]⊙[tanh(S c ·s k +S c ·h k-1 +b c )]

[0127] wherein h k denotes the hidden state vector of the first vehicle parameter, s k denotes the embedding vector of the first vehicle parameter, h k-1 denotes the hidden state vector of the second vehicle parameter corresponding to the first vehicle parameter, c k denotes the cell state at the corresponding moment, c k-1 denotes the cell state of the previous moment at the target moment, S o denotes the weight matrix of the output gate, b o denotes the bias of the output gate, S f denotes the weight matrix of the forget gate, b f denotes the bias of the forget gate, S i denotes the weight matrix of the input gate, b i denotes the bias of the input gate, S c denotes the weight matrix of the cell state, b c denotes the bias of the cell state, σ denotes an activation function, and ⊙ denotes an effective product.

[0128] At step S1021, the road surface image feature extraction network can receive the road surface image related to the target moment, and the road surface image feature extraction network outputs the road surface image feature vector of the road surface image.

[0129] At step S1022, the target hidden state vector is generated according to the hidden state vector of each first vehicle parameter of the target vehicle at the target moment, and the hidden state vector of each second vehicle parameter of the target vehicle at the previous moment of the target moment.

[0130] In a possible implementation of the step S1022, a vehicle parameter combination vector including the hidden state vector of each of the plurality of first vehicle parameters of the target vehicle at the target moment is generated, where the hidden state vector of each of the plurality of first vehicle parameters of the target vehicle at the target moment is respectively as a component of the vehicle parameter combination vector including the hidden state vector of each of the plurality of first vehicle parameters of the target vehicle at the target moment; a vehicle parameter combination vector including the hidden state vector of each of the plurality of second vehicle parameters of the target vehicle at the moment before the target moment is generated, where the hidden state vector of each of the plurality of second vehicle parameters of the target vehicle at the moment before the target moment is respectively as a component of the vehicle parameter combination vector including the hidden state vector of each of the plurality of second vehicle parameters of the target vehicle at the moment before the target moment; and the vehicle parameter combination vector including the hidden state vector of each of the plurality of first vehicle parameters of the target vehicle at the target moment is added to the vehicle parameter combination vector including the hidden state vector of each of the plurality of second vehicle parameters of the target vehicle at the moment before the target moment, to obtain the target hidden state vector.

[0131] In the step S1023, a feature for prediction is generated according to the target hidden state vector and the road surface image feature vector of the road surface image related to the target moment, and a prediction result related to the target road surface is predicted according to the feature for prediction.

[0132] In a possible implementation of step S1023, the target hidden state vector is added to the road image feature vector of the road image related to the target time to obtain the feature for prediction. When the hidden state vector of the vehicle parameter and the road image feature vector of the road image related to the target time are combined by step S1021-step S1023 to predict the prediction result related to the target road, the target neural network can be trained in the following manner: training the target neural network by using a training data set of the target neural network. During the training of the target neural network, the parameters of the target neural network are updated. In order to obtain the training data set of the target neural network, the vehicle for training drives on each of a plurality of roads such as snow, mud, rock, sand, asphalt, cement, wading, etc. For each of the plurality of roads, the vehicle for training collects data corresponding to the type of data in the information for prediction during the driving of the vehicle for training on the road. According to the data collected during the driving of the vehicle for training on the road, the training data of the road is determined. The training data of each of the plurality of roads constitutes the training data set of the target neural network. As an example, the vehicle for training collects the inertial torque of the wheel of the vehicle for training, the wheel longitudinal force of the vehicle for training, the suspension height of the vehicle for training, the body vertical acceleration of the vehicle for training, the lateral angular velocity of the vehicle for training, etc. The vehicle end parameter, and collects the road image including the road object representing the road on which the vehicle for training drives. The training data in the training data set of the target neural network includes: a plurality of first vehicle parameters of the target vehicle at the training time, a plurality of second vehicle parameters of the target vehicle at the last time of the target training time, and a road image related to the training time. The training data in the training data set of the target neural network is labeled to obtain the labeled result of the training data of the target neural network. The labeled result of the training data of the target neural network is the result expected to be output by the target neural network when the training data is input into the target neural network. The labeled result of the training data includes: the labeled road adhesion coefficient, the labeled road bump degree, and the labeled road slope. During the training of the target neural network, the training data in the training data set of the target neural network is input into the target neural network, and the target neural network outputs the prediction result corresponding to the training data. The loss between the prediction result corresponding to the training data and the labeled result of the training data is calculated, and the loss is used to update the parameters of the target neural network.

[0133] In step S103, each of the prediction results is provided to the corresponding system of the target vehicle, respectively, so that the corresponding system of the target vehicle adjusts according to the corresponding item in the prediction result.

[0134] The prediction results related to the target road surface include: the road surface adhesion coefficient of the target road surface, the bumpiness of the target road surface, and the road surface slope of the target road surface. Each of the road surface adhesion coefficient of the target road surface, the bumpiness of the target road surface, and the road surface slope of the target road surface can be provided to the corresponding system of the target vehicle.

[0135] As an example, the bumpiness of the target road surface is directly provided to the suspension system of the target vehicle, and the suspension system of the target vehicle can directly adaptively adjust according to the bumpiness of the target road surface to provide a more stable driving experience. The road surface adhesion coefficient of the target road surface is directly provided to the power distribution system of the target vehicle, and the power distribution system of the target vehicle can directly adjust the driving force according to the road surface adhesion coefficient of the target road surface to ensure that the target vehicle maintains good traction and stability. The road surface slope of the target road surface is directly provided to the power output system of the target vehicle.

[0136] In addition, the prediction results related to the target road surface can also include information required for corresponding adjustment of other corresponding systems of the target vehicle in addition to the road surface adhesion coefficient of the target road surface, the bumpiness of the target road surface, and the road surface slope of the target road surface. For example, the prediction results related to the target road surface can also include the slipperiness of the target road surface. The slipperiness of the target road surface can be provided to the vehicle control system of the target vehicle, and the vehicle control system of the target vehicle can directly automatically adjust the vehicle speed according to the slipperiness of the road surface to reduce the risk of skidding.

[0137] In step S103, the road surface adhesion coefficient of the target road surface, the bumpiness of the target road surface, and the road surface slope of the target road surface can be provided to the chassis system of the target vehicle, so that the suspension of the target vehicle, the braking of the target vehicle, and the traction control of the target vehicle are optimized according to the road surface adhesion coefficient of the target road surface, the bumpiness of the target road surface, and the road surface slope of the target road surface by the chassis system of the target vehicle.

[0138] In one possible implementation, providing each of the prediction results to a corresponding system of the target vehicle respectively, so that the corresponding system of the target vehicle adjusts according to the corresponding prediction result, comprises: providing the road surface adhesion coefficient of the target road to the traction control system of the target vehicle, so that the traction control system of the target vehicle adjusts the output power of the engine of the target vehicle according to the road surface adhesion coefficient of the target road; providing the bump degree of the target road to the suspension system of the target vehicle, so that the suspension system of the target vehicle adjusts the damping rate according to the bump degree of the target road; providing the road surface slope of the target road to the torque distribution system of the target vehicle, so that the torque distribution system of the target vehicle adjusts the engine torque of the target vehicle according to the road surface slope of the target road. The adhesion coefficient represents the friction between the tire and the road surface, and is an important indicator of the vehicle's grip. When the road surface adhesion coefficient of the target road is low (e.g., the target road is wet, or the target road is an icy or snowy road), the traction control (TCS) system of the chassis system of the target vehicle can limit the output power of the engine of the target vehicle to prevent the driven wheels from spinning and ensure smooth acceleration of the target vehicle. When the road surface adhesion coefficient of the target road is high (e.g., the target road is a dry asphalt road), the traction control system allows higher power output to improve acceleration performance. Low adhesion coefficient of the road surface usually means poor grip, when the road surface adhesion coefficient of the target road is low, the suspension system of the chassis system of the target vehicle can choose a softer mode to absorb shocks and improve the persistence of tire contact with the ground. When the road surface adhesion coefficient of the target road is high, the suspension system of the target vehicle can choose a harder mode to increase handling and improve driving stability. The bump degree reflects the unevenness of the road surface, which directly affects the stability and ride comfort of the vehicle. The chassis system can adjust the damping, stiffness, and other parameters of the suspension to enhance the response to shocks and improve the driving experience. When the bump degree of the target road is high (e.g., the target road is bumpy or the target road is a gravel road), the suspension system of the target vehicle can dynamically adjust the damping rate according to the bump degree information to absorb the impact of the road surface and reduce the shaking of the body of the target vehicle, improving the comfort inside the vehicle. When the bump degree of the target road is low, for example, the target road is a flat road, the suspension system of the target vehicle can increase the damping to provide better handling. The road surface slope of the target road provides the angle information of the road surface inclination, which has an important influence on the traction and braking force distribution of the vehicle. When the road surface slope of the target road indicates a large slope, for example, the target vehicle is driving on a steep uphill road, the torque distribution system of the chassis system of the target vehicle will increase the engine torque of the target vehicle and distribute more traction to the driven wheels to prevent the target vehicle from slipping during climbing and ensure stable acceleration performance. When the road surface slope of the target road indicates a downhill, the engine torque of the target vehicle is actively reduced to prevent the speed of the target vehicle from getting out of control.When the road slope of the target road indicates that the target vehicle is descending a steep slope, a hill descent control (HDC) system of the chassis system of the target vehicle can be activated. The hill descent control system will gradually reduce the speed of the vehicle by controlling the brakes to avoid the target vehicle descending too fast due to the steep slope, and to improve the stability and safety of the target vehicle when descending the slope.

[0139] The road adhesion coefficient of the target road, the roughness of the target road, and the road slope of the target road reflecting the road conditions of the target road are provided to the chassis system of the target vehicle, so that the chassis system of the target vehicle can make corresponding adjustments suitable for the road conditions of the target road according to the information reflecting the road conditions of the target road. The dynamic stability, safety and comfort of the target vehicle are improved.

[0140] Reference Figure 4 which shows a flowchart of another information providing method provided by an embodiment of the present application.

[0141] In step S401, information for prediction corresponding to a target time is obtained.

[0142] In step S402, using a target neural network, a prediction result related to a target road is predicted according to the information for prediction corresponding to the target time, including: generating a target hidden state vector according to an attention hidden state vector of each first vehicle parameter of the target vehicle at the target time and a hidden state vector of each second vehicle parameter of the target vehicle at a previous time of the target time.

[0143] Step S4022 includes step S40221-step S40223.

[0144] In step S40221, an attention parameter of each first vehicle parameter of the target vehicle at the target time is determined according to the hidden state vector of each first vehicle parameter of the target vehicle at the target time.

[0145] In step S40222, an attention hidden state vector of each first vehicle parameter of the target vehicle at the target time is generated according to the hidden state vector of each first vehicle parameter of the target vehicle at the target time and the attention parameter of each first vehicle parameter of the target vehicle at the target time.

[0146] In step S40223, a target hidden state vector is generated according to the attention hidden state vector of each first vehicle parameter of the target vehicle at the target time and the hidden state vector of each second vehicle parameter of the target vehicle at the previous time of the target time.

[0147] The attention parameter of the first vehicle parameter of the target vehicle at the target time can reflect the importance of the hidden state vector of the first vehicle parameter of the target vehicle at the target time. The greater the attention parameter of the first vehicle parameter of the target vehicle at the target time, the higher the importance of the hidden state vector of the first vehicle parameter of the target vehicle at the target time. The smaller the attention parameter of the first vehicle parameter of the target vehicle at the target time, the lower the importance of the hidden state vector of the first vehicle parameter of the target vehicle at the target time. Thus, the accuracy of the target hidden state vector can be improved, and the accuracy of the prediction result related to the road surface on which the vehicle travels can be improved.

[0148] In one possible implementation of step S40221, the target neural network comprises a first linear layer. In step S40221, for each first vehicle parameter of the target vehicle at the target time, the hidden state vector of the first vehicle parameter is input to the input first linear layer, the first linear layer applies the first score function to the hidden state vector of the first vehicle parameter, and the first linear layer outputs the first linear layer output of the first vehicle parameter; the first linear layer output of the first vehicle parameter is applied to the activation function corresponding to the first linear layer to obtain the first attention score of the first vehicle parameter. The first attention scores of each first vehicle parameter of the target vehicle at the target time are added to obtain the sum of the first attention scores of each first vehicle parameter of the target vehicle at the target time. For each first vehicle parameter of the target vehicle at the target time, the first attention score of the first vehicle parameter is divided by the sum of the first attention scores of each first vehicle parameter of the target vehicle at the target time to obtain the normalized first attention score of the first vehicle parameter at the target time, and the normalized first attention score of the first vehicle parameter is determined as the attention parameter of the first vehicle parameter.

[0149] The parameters of the first linear layer include a weight matrix of the first linear layer and a bias of the first linear layer.

[0150] The first score function includes a term representing the weight matrix of the first linear layer, a term of the bias of the first linear layer, and a term representing the hidden state vector as an independent variable.

[0151] The activation function corresponding to the first linear layer can be a tanh function.

[0152] The first attention score of a first vehicle parameter obtained by using the first linear layer and the activation function corresponding to the first linear layer can be represented as:

[0153] γ(h i) = tanh(h i .w a +b a )

[0154] wherein y(h i ) denotes the first attention score of the first vehicle parameter, h i denotes the hidden state vector of the first vehicle parameter, w a denotes the weight matrix of the first linear layer, and b a denotes the bias of the first linear layer.

[0155] In another possible implementation of the step S40221, the step S40221 comprises: a step S40221a to a step S40221c.

[0156] In the step S40221a, the first attention score of each first vehicle parameter of the target vehicle at the target time is determined according to the hidden state vector of each first vehicle parameter of the target vehicle at the target time.

[0157] In the step S40221b, for each first vehicle parameter of the target vehicle at the target time, the second attention score of the first vehicle parameter is determined according to the hidden state vector of the first vehicle parameter and the hidden state vector of the second vehicle parameter corresponding to the first vehicle parameter, wherein the type of the first vehicle parameter is the same as the type of the second vehicle parameter corresponding to the first vehicle parameter.

[0158] In the step S40221c, the attention parameter of each first vehicle parameter of the target vehicle at the target time is determined according to the first attention score of each first vehicle parameter of the target vehicle at the target time and the second attention score of each first vehicle parameter of the target vehicle at the target time.

[0159] The step S40221 considers the importance of the hidden state vector of the first vehicle parameter of the target vehicle at the target time in the case of considering the hidden state vector of the second vehicle parameter corresponding to the first vehicle parameter. The accuracy of the attention parameter of each first vehicle parameter of the target vehicle at the target time is improved.

[0160] In one possible implementation of the step S40221a, in the step S40221a, for each first vehicle parameter of the target vehicle at the target time, the hidden state vector of the first vehicle parameter is input into an input first linear layer, the first linear layer applies a first score function to the hidden state vector of the first vehicle parameter, and the first linear layer outputs a first linear layer output of the first vehicle parameter; an activation function corresponding to the first linear layer is applied to the first linear layer output of the first vehicle parameter to obtain the first attention score of the first vehicle parameter.

[0161] In a possible implementation of step S40221b, the target neural network comprises: a second linear layer. In step S40221b, for each first vehicle parameter of the target vehicle at the target time, the hidden state vector of the second vehicle parameter corresponding to the first vehicle parameter is pooled to obtain a pooling result of the hidden state vector of the second vehicle parameter corresponding to the first vehicle parameter; the hidden state vector of the first vehicle parameter and the pooling result of the hidden state vector of the second vehicle parameter corresponding to the first vehicle parameter are input into the second linear layer, the second linear layer applies a second score function to the hidden state vector of the first vehicle parameter and the pooling result of the hidden state vector of the second vehicle parameter corresponding to the first vehicle parameter, and the second linear layer outputs a second linear layer output of the first vehicle parameter; and an activation function corresponding to the second linear layer is applied to the second linear layer output of the first vehicle parameter to obtain a second attention score of the first vehicle parameter.

[0162] The parameters of the second linear layer include: a first weight matrix, a second weight matrix, and a bias of the second linear layer.

[0163] The second score function includes: a term representing the first weight matrix, a term representing the second weight matrix, a bias term of the second linear layer, and a term representing the hidden state vector as an independent variable.

[0164] The activation function corresponding to the second linear layer can be a tanh function.

[0165] The second attention score of the first vehicle parameter obtained by using the second linear layer and the activation function corresponding to the second linear layer can be represented as:

[0166]

[0167] wherein, the second attention score of the first vehicle parameter is represented by the hidden state vector of the first vehicle parameter is represented by the hidden state vector of the second vehicle parameter corresponding to the first vehicle parameter is represented by w1 represents the first weight matrix, w2 represents the second weight matrix, b represents the bias of the second linear layer, and pool represents pooling.

[0168] In a possible implementation of the step S40221c, for each first vehicle parameter of the target vehicle at the target moment, an exponential function with base e is applied to the second attention score of the first vehicle parameter, i.e. the second attention score of the first vehicle parameter is taken as the value of the independent variable of the exponential function with base e, to obtain a second function value corresponding to the first vehicle parameter. For each first vehicle parameter of the target vehicle at the target moment, an exponential function with base e is applied to the first attention score of the first vehicle parameter, i.e. the first attention score of the first vehicle parameter is taken as the value of the independent variable of the exponential function with base e, to obtain a first function value corresponding to the first vehicle parameter. The first function values corresponding to each first vehicle parameter of the target vehicle at the target moment are added to obtain a sum of the first function values corresponding to each first vehicle parameter of the target vehicle at the target moment. For each first vehicle parameter of the target vehicle at the target moment, the second function value corresponding to the first vehicle parameter is divided by the sum of the first function values corresponding to each first vehicle parameter of the target vehicle at the target moment, to obtain an attention parameter of the first vehicle parameter.

[0169] The attention parameter of a first vehicle parameter can be represented as:

[0170]

[0171] wherein α t represents the attention parameter of the first vehicle parameter, exp represents an exponential function with base e, represents the second attention score of the first vehicle parameter, γ(h k ) represents the first attention score of the kth first vehicle parameter of the target vehicle at the target moment.

[0172] In the step S40222, for each first vehicle parameter of the target vehicle at the target moment, the attention parameter of the first vehicle parameter is multiplied by the hidden state vector of the first vehicle parameter to obtain an attention hidden state vector of the first vehicle parameter. The ith component of the attention hidden state vector of the first vehicle parameter is the product of the ith component of the hidden state vector of the first vehicle parameter and the attention parameter of the first vehicle parameter.

[0173] In a possible implementation of the S40223, at the step S40223, the vehicle parameter combination vector including the attention hidden state vectors of each first vehicle parameter of the target vehicle at the target time is generated, where the attention hidden state vectors of each first vehicle parameter of the target vehicle at the target time are respectively taken as components of the vehicle parameter combination vector including the attention hidden state vectors of each first vehicle parameter of the target vehicle at the target time; the vehicle parameter combination vector including the hidden state vectors of each second vehicle parameter of the target vehicle at the target time is generated, where the hidden state vectors of each second vehicle parameter of the target vehicle at the target time are respectively taken as components of the vehicle parameter combination vector including the hidden state vectors of each second vehicle parameter of the target vehicle at the target time; and the vehicle parameter combination vector including the attention hidden state vectors of each first vehicle parameter of the target vehicle at the target time is added to the vehicle parameter combination vector including the hidden state vectors of each second vehicle parameter of the target vehicle at the target time, to obtain the target hidden state vector.

[0174] In another possible implementation of the S40223, the S40223 includes the S40223a.

[0175] At the step S40223a, the attention hidden state vectors of each second vehicle parameter of the target vehicle at the target time are generated according to the hidden state vectors of each second vehicle parameter of the target vehicle at the target time, and the attention parameters of each second vehicle parameter of the target vehicle at the target time; and the target hidden state vector is generated according to the attention hidden state vectors of each first vehicle parameter of the target vehicle at the target time and the attention hidden state vectors of each second vehicle parameter of the target vehicle at the target time.

[0176] The attention parameters of each second vehicle parameter of the target vehicle at the target time are generated according to the hidden state vectors of each second vehicle parameter of the target vehicle at the target time.

[0177] It should be noted that the hidden state vectors of each second vehicle parameter of the target vehicle at the target time and the attention parameters of each second vehicle parameter of the target vehicle at the target time are generated when predicting the prediction result of the road surface related to the road surface object representation in the road surface image related to the last time of the target time. The prediction result of the road surface related to the road surface object representation in the road surface image related to the last time of the target time is before the step S401.

[0178] Step S40223a considers the importance of the hidden state vector of the target vehicle's second vehicle parameter at the target time for predicting the prediction results related to the target road surface. The magnitude of the attention parameter of the target vehicle's second vehicle parameter at the previous time step reflects the importance of the hidden state vector of the target vehicle's second vehicle parameter at the target time step. A larger attention parameter indicates higher importance of the hidden state vector of the target vehicle's second vehicle parameter at the previous time step, and a smaller attention parameter indicates lower importance of the hidden state vector of the target vehicle's second vehicle parameter at the previous time step. Therefore, the accuracy of the target hidden state vector can be improved, thereby improving the accuracy of predicting the prediction results related to the road surface on which the vehicle is traveling.

[0179] The process of determining the attention parameter of each second vehicle parameter of the target vehicle at the previous time based on the hidden state vector of each second vehicle parameter of the target vehicle at the previous time is similar to the process of determining the attention parameter of each first vehicle parameter of the target vehicle at the previous time based on the hidden state vector of each first vehicle parameter among multiple first vehicle parameters of the target vehicle at the previous time in step S40221.

[0180] Based on the hidden state vector of each second vehicle parameter of the target vehicle at the previous time of the target time, the implementation of the attention parameter of each second vehicle parameter of the target vehicle at the previous time of the target time can be any possible implementation of step S40221.

[0181] The target time is denoted as time t, the previous time of the target time is denoted as time t-1, and the previous time of time t-1 is denoted as time t-2.

[0182] By replacing the first vehicle parameter of the target vehicle at time t in one possible implementation of step S40221 with the second vehicle parameter of the target vehicle at time t-1, an implementation of determining the attention parameter of each second vehicle parameter of the target vehicle at the time before the target time can be obtained based on the hidden state vector of each second vehicle parameter of the target vehicle at the time before the target time.

[0183] Another possible implementation of step S40221 is that the first vehicle parameter of the target vehicle at time t is replaced by the second vehicle parameter of the target vehicle at time t-1, and the second vehicle parameter of the target vehicle at time t-1 is replaced by the vehicle parameter of the target vehicle at time t-2, so as to obtain another implementation of determining the attention parameter of each second vehicle parameter of the target vehicle at the last time of the target time according to the hidden state vector of each second vehicle parameter of the target vehicle at the last time of the target time.

[0184] The attention hidden state vector of each second vehicle parameter of the target vehicle at the last time of the target time is generated according to the attention parameter of each second vehicle parameter of the target vehicle at the last time of the target time and the hidden state vector of each second vehicle parameter of the target vehicle at the last time of the target time.

[0185] For each second vehicle parameter of the target vehicle at the last time of the target time, the attention hidden state vector of the second vehicle parameter is obtained by multiplying the attention parameter of the second vehicle parameter and the hidden state vector of the second vehicle parameter.

[0186] In one possible implementation of step 40223a, a first vehicle parameter combination vector including the attention hidden state vectors of each first vehicle parameter of the target vehicle at the target time is generated, wherein the attention hidden state vectors of each first vehicle parameter of the target vehicle at the target time are respectively taken as components of the first vehicle parameter combination vector; a second vehicle parameter combination vector including the attention hidden state vectors of each second vehicle parameter of the target vehicle at the last time of the target time is generated, wherein the attention hidden state vectors of each second vehicle parameter of the target vehicle at the last time of the target time are respectively taken as components of the second vehicle parameter combination vector; and the first vehicle parameter combination vector and the second vehicle parameter combination vector are added to obtain the target hidden state vector.

[0187] Reference Figure 5 FIG. 4 shows a flowchart of one example of obtaining the target hidden state vector.

[0188] In this example, time t is the target time, and time t-1 is the last time of the target time.

[0189] In this example, the hidden state vector of each second vehicle parameter of the target vehicle at the time t-1 is obtained by respectively inputting each second vehicle parameter of the target vehicle at the time t-1 into the target neural network.

[0190] It should be noted that the hidden state vector of each second vehicle parameter of the target vehicle at the time t-1, the attention parameter of each second vehicle parameter of the target vehicle at the time t-1 are generated when predicting the prediction result related to the road surface object represented in the road surface image related to the time t-1.

[0191] In this example, the hidden state vector of each first vehicle parameter of the target vehicle at the time t is obtained by respectively inputting each first vehicle parameter of the target vehicle at the time t into the target neural network.

[0192] In this example, the hidden state vector of each second vehicle parameter of the target vehicle at the time t-1 is pooled to obtain the pooling result of the hidden state vector of each second vehicle parameter of the target vehicle at the time t-1. According to the pooling result of the hidden state vector of each second vehicle parameter of the target vehicle at the time t-1 and the hidden state vector of each first vehicle parameter of the target vehicle at the time t, the attention hidden state vector of each first vehicle parameter of the target vehicle at the time t is obtained.

[0193] In this example, the target hidden state vector is generated according to the attention hidden state vector of each first vehicle parameter of the target vehicle at the time t and the hidden state vector of each second vehicle parameter of the target vehicle at the time t-1.

[0194] In one possible implementation of the step S40223, the target hidden state vector is multiplied by the vehicle parameter feature weight to obtain a first weighted vector, and the road surface image feature vector of the road surface image related to the target time is multiplied by the road surface image feature weight to obtain a second weighted vector; the feature used for prediction is generated according to the first weighted vector and the second weighted vector.

[0195] The i-th component in the first weighted vector is the product of the i-th component in the target hidden state vector and the vehicle parameter feature weight.

[0196] The i-th component in the second weighted vector is the product of the i-th component in the road surface image feature vector of the road surface image related to the target time and the road surface image feature weight.

[0197] In a possible implementation, generating the feature for prediction according to the first weight vector and the second weight vector comprises: adding the first weight vector and the second weight vector to obtain the feature for prediction.

[0198] In a possible implementation, before step S401, the vehicle parameter feature weight and the road surface image feature weight are determined according to an accuracy of the vehicle parameter feature extraction network in the target neural network on a validation set of the target neural network, and an accuracy of the road surface image feature extraction network in the target neural network on the validation set of the target neural network.

[0199] The number of times that the prediction output of the vehicle parameter feature extraction network on the validation data is accurate is divided by the number of times that the validation data is input to the vehicle parameter feature extraction network.

[0200] The number of times that the prediction output of the road surface image feature extraction network on the validation data is accurate is divided by the number of times that the validation data is input to the vehicle parameter feature extraction network.

[0201] In a possible implementation, it is determined whether a similarity between a prediction result output by the vehicle parameter feature extraction network in the target neural network and a vehicle parameter feature label output corresponding to the validation data when the validation data is input to the target neural network is greater than a similarity threshold, if yes, the prediction of the vehicle parameter feature extraction network on the validation data is accurate, and if no, the prediction of the vehicle parameter feature extraction network on the validation data is inaccurate. The vehicle parameter feature label output corresponding to the validation data is a result expected to be output by the vehicle parameter feature extraction network when the validation data is input to the vehicle parameter feature extraction network. It is determined whether a similarity between a prediction result output by the road surface image feature extraction network in the target neural network and a road surface image feature label output corresponding to the validation data when the validation data is input to the target neural network is greater than a similarity threshold, if yes, the prediction of the road surface image feature extraction network on the validation data is accurate, and if no, the prediction of the road surface image feature extraction network on the validation data is inaccurate.

[0202] The vehicle parameter feature weight θ can be calculated by using the following formula:

[0203]

[0204] wherein, A S represents the accuracy of the vehicle parameter feature extraction network on the validation set of the target neural network, A i represents the accuracy of the road surface image feature extraction network on the validation set of the target neural network.

[0205] The road surface image feature weight μ can be calculated by using the following formula:

[0206]

[0207] In step S403, each item in the prediction result is provided to a corresponding system of the target vehicle respectively, so that the corresponding system of the target vehicle adjusts according to the corresponding item in the prediction result.

[0208] The process of step S403 refers to the process of step S303.

[0209] The embodiments of the present application also provide an information providing device, which is installed on a vehicle, and is used to implement the above-mentioned method embodiments and preferred embodiments, and will not be described here. As used below, the term "unit" can be a combination of software and / or hardware that implements a predetermined function. Although the devices described in the following embodiments are preferably implemented in software, hardware, or a combination of software and hardware can also be implemented and conceived. The devices in the embodiments of the present application are presented in the form of functional units, which refer to ASIC (Application Specific Integrated Circuit) circuits, processors and memories that execute one or more software or fixed programs, and / or other devices that can provide the above-mentioned functions.

[0210] The information providing device comprises:

[0211] An acquisition unit is configured to acquire information for prediction corresponding to a target time, the information for prediction comprising: a plurality of first vehicle parameters of the target vehicle at the target time, a road surface image related to the target time, the road surface image comprising a road surface object representing a target road surface;

[0212] A prediction unit is configured to predict, by using a target neural network, a prediction result related to the target road surface according to the information for prediction, the prediction result comprising: a road surface adhesion coefficient of the target road surface, a bump degree of the target road surface, and a road surface slope of the target road surface.

[0213] A providing unit is configured to provide each item in the prediction result to a corresponding system of the target vehicle respectively, so that the corresponding system adjusts according to the corresponding item in the prediction result.

[0214] In a possible implementation, the providing unit is further configured to provide each of the prediction results to a corresponding system of the target vehicle respectively, so that the corresponding system adjusts according to the corresponding prediction result, including: providing the road surface adhesion coefficient of the target road surface to a traction control system of the target vehicle, so that the traction control system of the target vehicle adjusts the output power of the engine of the target vehicle according to the road surface adhesion coefficient of the target road surface; providing the bump degree of the target road surface to a suspension system of the target vehicle, so that the suspension system of the target vehicle adjusts the damping rate according to the bump degree of the target road surface; and providing the road surface slope of the target road surface to a torque distribution system of the target vehicle, so that the torque distribution system of the target vehicle adjusts the engine torque of the target vehicle according to the road surface slope of the target road surface.

[0215] In a possible implementation, the prediction unit is further configured to generate a hidden state vector of each of the plurality of first vehicle parameters, and extract a road surface image feature vector of the road surface image; generate a target hidden state vector according to the hidden state vector of each of the plurality of first vehicle parameters and the hidden state vector of each of the plurality of second vehicle parameters of the target vehicle at the last moment of the target moment; generate a feature for prediction according to the target hidden state vector and the road surface image feature vector, and predict the prediction result according to the feature for prediction.

[0216] In a possible implementation, the prediction unit is further configured to determine an attention parameter of each of the plurality of first vehicle parameters according to the hidden state vector of each of the plurality of first vehicle parameters; generate an attention hidden state vector of each of the plurality of first vehicle parameters according to the hidden state vector of each of the plurality of first vehicle parameters and the attention parameter of each of the plurality of first vehicle parameters; and generate a target hidden state vector according to the attention hidden state vector of each of the plurality of first vehicle parameters and the hidden state vector of each of the plurality of second vehicle parameters.

[0217] In a possible implementation, the prediction unit is further configured to determine a first attention score of each of the plurality of first vehicle parameters according to the hidden state vector of each of the plurality of first vehicle parameters; for each of the plurality of first vehicle parameters, determine a second attention score of the first vehicle parameter according to the hidden state vector of the first vehicle parameter and the hidden state vector of the second vehicle parameter corresponding to the first vehicle parameter, where the type of the first vehicle parameter is the same as the type of the second vehicle parameter corresponding to the first vehicle parameter; and determine an attention parameter of each of the plurality of first vehicle parameters according to the first attention score of each of the plurality of first vehicle parameters and the second attention score of each of the plurality of first vehicle parameters.

[0218] In a possible implementation, the prediction unit is further configured to generate an attention hidden state vector of each second vehicle parameter according to the hidden state vector of the second vehicle parameter, the attention parameter of the second vehicle parameter; and generate a target hidden state vector according to the attention hidden state vector of each first vehicle parameter, the attention hidden state vector of each second vehicle parameter.

[0219] In a possible implementation, the prediction unit is further configured to generate a first vehicle parameter combination vector including the attention hidden state vector of each first vehicle parameter, and generate a second vehicle parameter combination vector including the attention hidden state vector of each second vehicle parameter; and add the first vehicle parameter combination vector and the second vehicle parameter combination vector to obtain a target hidden state vector.

[0220] In a possible implementation, the prediction unit is further configured to multiply the target hidden state vector by a vehicle parameter feature weight to obtain a first weighted vector, and multiply the road surface image feature vector by a road surface image feature weight to obtain a second weighted vector; and generate a feature for prediction according to the first weighted vector and the second weighted vector.

[0221] In a possible implementation, the prediction unit is further configured to add the first weighted vector and the second weighted vector to obtain a feature for prediction.

[0222] In a possible implementation, the information providing apparatus includes:

[0223] The weight determination unit is configured to determine a vehicle parameter feature weight and a road surface image feature weight according to an accuracy of a vehicle parameter feature extraction network in the target neural network on a validation set of the target neural network, and an accuracy of a road surface image feature extraction network in the target neural network on the validation set, before obtaining the information for prediction corresponding to a target moment.

[0224] Reference Figure 6 , Figure 6Fig. 1 is a schematic diagram of a hardware structure of a computer device according to an embodiment of the present application. The computer device includes one or more processors 10, a memory 20, and interfaces for connecting various components, including high-speed interfaces and low-speed interfaces. The various components communicate with each other using different buses, and can be mounted on a common main board or otherwise mounted as needed. The processor can process instructions executed within the computer device, including instructions stored in the memory or on the memory to display graphical information of a GUI on an external input / output device, such as a display device coupled to the interface. In some alternative embodiments, multiple processors and / or buses can be used with multiple memories and multiple memory, if needed. Also, multiple vehicles can be connected, each providing part of the necessary operations (e.g., as a server array, a set of blade servers, or a multi-processor system). The processor 10 can be a central processing unit, a network processing unit, or a combination thereof. The processor 10 can further include a hardware chip. The hardware chip can be an application specific integrated circuit, a programmable logic device, or a combination thereof. The programmable logic device can be a complex programmable logic device, a field programmable logic gate array, a generic array logic, or any combination thereof. The memory 20 stores instructions executable by the at least one processor 10 to cause the at least one processor 10 to perform the method according to the embodiments described above. The memory 20 can include a program storage area and a data storage area, where the program storage area can store an operating system, application programs required by at least one function, and the data storage area can store data created according to the use of the vehicle, etc. In addition, the memory 20 can include a high-speed random access memory, and can also include a non-transitory memory, such as at least one magnetic disk storage device, a flash memory device, or other non-transitory solid-state memory device. In some alternative embodiments, the memory 20 can optionally include a memory disposed remotely with respect to the processor 10, which can be connected to the computer device through a network. Examples of the network include, but are not limited to, the Internet, an intranet, a local area network, a mobile communication network, and a combination thereof. The memory 20 can include a volatile memory, such as a random access memory, and can also include a non-volatile memory, such as a flash memory, a hard disk, or a solid state disk, and can also include a combination of the above types of memories. The computer device further includes an input device 30 and an output device 40. The processor 10, the memory 20, the input device 30, and the output device 40 can be connected by a bus or otherwise. The input device 30 can receive input digital or character information, and generate key signal inputs related to the user settings and function control of the computer device, such as a touch screen, a keypad, a mouse, a trackpad, a touchpad, a joystick, one or more mouse buttons, a trackball, a joystick, etc.The output device 40 can include a display device, an auxiliary lighting device (e.g., an LED), a haptic feedback device (e.g., a vibration motor), and the like. The display device described above includes, but is not limited to, a liquid crystal display, a light emitting diode, a display, and a plasma display. In some alternative embodiments, the display device can be a touch screen.

[0225] The embodiments of the present application also provide a computer readable storage medium, the method according to the embodiments of the present application can be implemented in hardware, firmware, or recorded in a storage medium, or stored in a remote storage medium or a non-transitory machine readable storage medium and downloaded to a local storage medium by network, so that the method described herein can be processed by such software on a storage medium using a general purpose computer, a special purpose processor, or programmable or special purpose hardware. Among them, the storage medium can be a magnetic disk, an optical disk, a read-only memory, a random access memory, a flash memory, a hard disk or a solid state disk, etc.; further, the storage medium can also include a combination of the above types of memories. It can be understood that the computer, processor, microprocessor controller or programmable hardware includes a storage component that can store or receive software or computer code, when the software or computer code is accessed and executed by the computer, processor or hardware, the method shown in the above embodiments is implemented.

[0226] Some of the embodiments of the present application can be applied as a computer program product, for example, computer program instructions, when executed by a computer, the operation of the computer can call or provide the method and / or technical solutions according to the present application. Those skilled in the art should understand that the form of computer program instructions in computer readable medium includes but is not limited to source file, executable file, installation package file, etc., and accordingly, the way of computer program instructions executed by computer includes but is not limited to: the computer directly executes the instructions, or the computer executes the corresponding compiled program after compiling the instructions, or the computer reads and executes the instructions, or the computer reads and installs the instructions and then executes the corresponding installed program. Here, the computer readable medium can be any available computer readable storage medium or communication medium accessible to the computer.

[0227] The above embodiments are only preferred embodiments for fully illustrating the present application, and the protection scope of the present application is not limited thereto. Any equivalent replacement or transformation of the present application made by those skilled in the art based on the present application is within the protection scope of the present application.

Claims

1. An information providing method characterized by comprising: The method comprises: obtaining information for prediction corresponding to a target time, the information for prediction comprising: a plurality of first vehicle parameters of a target vehicle at the target time, a road surface image related to the target time, the road surface image comprising a road surface object representing a target road surface; using a target neural network, predicting a prediction result related to the target road surface according to the information for prediction, the prediction result comprising: a road surface adhesion coefficient of the target road surface, a bump degree of the target road surface, and a road surface slope of the target road surface; the using the target neural network to predict the prediction result related to the target road surface according to the information for prediction comprises: generating a hidden state vector of each of the plurality of first vehicle parameters, and extracting a road surface image feature vector of the road surface image; generating a target hidden state vector according to the hidden state vector of each of the plurality of first vehicle parameters and a hidden state vector of each of a plurality of second vehicle parameters of the target vehicle at a previous time of the target time; generating a feature for prediction according to the target hidden state vector and the road surface image feature vector, and predicting the prediction result according to the feature for prediction; providing each of the prediction result to a corresponding system of the target vehicle respectively, so that the corresponding system adjusts according to the corresponding item of the prediction result.

2. The method of claim 1, wherein: providing each of the prediction result to a corresponding system of the target vehicle respectively, so that the corresponding system adjusts according to the corresponding item of the prediction result comprises: providing the road surface adhesion coefficient of the target road surface to a traction control system of the target vehicle, so that the traction control system of the target vehicle adjusts an output power of an engine of the target vehicle according to the road surface adhesion coefficient of the target road surface; providing the bump degree of the target road surface to a suspension system of the target vehicle, so that the suspension system of the target vehicle adjusts a damping rate according to the bump degree of the target road surface; providing the road surface slope of the target road surface to a torque distribution system of the target vehicle, so that the torque distribution system of the target vehicle adjusts an engine torque of the target vehicle according to the road surface slope of the target road surface.

3. The method of claim 1, wherein: generating the target hidden state vector according to the hidden state vector of each of the plurality of first vehicle parameters and the hidden state vector of each of the plurality of second vehicle parameters of the target vehicle at the previous time of the target time comprises: determining an attention parameter of each of the plurality of first vehicle parameters according to the hidden state vector of each of the plurality of first vehicle parameters; generating an attention hidden state vector of each of the plurality of first vehicle parameters according to the hidden state vector of each of the plurality of first vehicle parameters and the attention parameter of each of the plurality of first vehicle parameters; generating the target hidden state vector according to the attention hidden state vector of each of the plurality of first vehicle parameters and the hidden state vector of each of the plurality of second vehicle parameters.

4. The method of claim 3, wherein: Based on the hidden state vector of each of the plurality of first vehicle parameters, the attention parameters for each first vehicle parameter are determined, including: Based on the hidden state vector of each of the plurality of first vehicle parameters, a first attention score is determined for each of the first vehicle parameters; For each first vehicle parameter, a second attention score for the first vehicle parameter is determined based on the hidden state vector of the first vehicle parameter and the hidden state vector of the second vehicle parameter corresponding to the first vehicle parameter, wherein the type of the first vehicle parameter is the same as the type of the second vehicle parameter corresponding to the first vehicle parameter. The attention parameter of each first vehicle parameter is determined based on the first attention score and the second attention score of each first vehicle parameter.

5. The method of claim 4, wherein: Generating the target hidden state vector based on the attention hidden state vector of each first vehicle parameter and the hidden state vector of each second vehicle parameter includes: Based on the hidden state vector of each second vehicle parameter and the attention parameter of each second vehicle parameter, generate the attention hidden state vector of each second vehicle parameter; A target hidden state vector is generated based on the attention hidden state vector of each first vehicle parameter and the attention hidden state vector of each second vehicle parameter.

6. The method of claim 5, wherein: Generating the target hidden state vector based on the attention hidden state vector of each first vehicle parameter and the attention hidden state vector of each second vehicle parameter includes: Generate a first vehicle parameter combination vector that includes the attention hidden state vector of each first vehicle parameter, and generate a second vehicle parameter combination vector that includes the attention hidden state vector of each second vehicle parameter; The first vehicle parameter combination vector is added to the second vehicle parameter combination vector to obtain the target hidden state vector.

7. The method according to any one of claims 3-6, characterized by: Based on the target hidden state vector and the road surface image feature vector, the features generated for prediction include: The target hidden state vector is multiplied by the vehicle parameter feature weights to obtain a first weighted vector, and the road surface image feature vector is multiplied by the road surface image feature weights to obtain a second weighted vector. Features for prediction are generated based on the first weighted vector and the second weighted vector.

8. The method of claim 7, wherein: Based on the first weighted vector and the second weighted vector, the features generated for prediction include: The first weighted vector and the second weighted vector are added together to obtain the features used for prediction.

9. The method of claim 7, wherein: Before acquiring the information for prediction corresponding to the target time, the method further includes: Based on the accuracy of the vehicle parameter feature extraction network in the target neural network on the validation set and the accuracy of the road image feature extraction network in the target neural network on the validation set, the weights of the vehicle parameter features and the weights of the road image features are determined.

10. An information providing apparatus characterized by comprising: The device includes: An acquisition unit is configured to acquire information for prediction corresponding to a target time, the information for prediction including a plurality of first vehicle parameters of a target vehicle at the target time, and a road surface image related to the target time, the road surface image including a road surface object representing a target road surface; A prediction unit is configured to predict, by using a target neural network, a prediction result related to the target road surface according to the information for prediction, the prediction result including a road surface adhesion coefficient of the target road surface, a bump degree of the target road surface, and a road surface slope of the target road surface, the prediction by using the target neural network according to the information for prediction including generating a hidden state vector of each of the plurality of first vehicle parameters, and extracting a road surface image feature vector of the road surface image, generating a target hidden state vector according to the hidden state vector of each of the plurality of first vehicle parameters and a hidden state vector of each of a plurality of second vehicle parameters of the target vehicle at a previous time of the target time, and generating a feature for prediction according to the target hidden state vector and the road surface image feature vector, and predicting the prediction result according to the feature for prediction; A providing unit is configured to provide each of the prediction result to a corresponding system of the target vehicle respectively, so that the corresponding system adjusts according to the corresponding item of the prediction result.

11. A computer device installed on a vehicle, characterized by, The method comprises the following steps: A memory and a processor are connected with each other in communication, the memory stores computer instructions, and the processor executes the computer instructions to perform the method in any one of claims 1 to 9.

12. A computer-readable storage medium, characterized in that, The computer readable storage medium stores computer instructions for causing a computer to perform the method in any one of claims 1 to 9.

13. A computer program product, characterised in that, The computer instructions are configured to cause a computer to perform the method in any one of claims 1 to 9.

Images