Apparatus for estimating vehicle weight and method therefor

1 . An apparatus for estimating a vehicle weight, the apparatus comprising:

a memory storing computer-executable instructions; and

at least one processor configured to access the memory and execute the instructions;

wherein the at least one processor is configured to;

apply a real velocity and real acceleration to a trained real-to-virtual transformation model to obtain a virtual output set at a target time point, the virtual output set including a virtual gradient of a vehicle, a virtual velocity of the vehicle, and virtual longitudinal acceleration of the vehicle;

apply the virtual output set at the target time point and a real wheel torque at the target time point to a trained mass estimation model to obtain weight information of the vehicle;

determine a first loss function for training the real-to-virtual transformation model, based on at least one of weight information of the vehicle prior to the target time point, longitudinal acceleration of the vehicle prior to the target time point, a velocity of the vehicle prior to the target time point, or a gradient of the vehicle prior to the target time point, or any combination thereof;

determine a second loss function for training the mass estimation model, based on weight information of the vehicle prior to the target time point, a ratio between the weight information of the vehicle and a real wheel torque of the vehicle prior to the target time point, and a resistance force of the vehicle prior to the target time point;

include longitudinal dynamics information in the second loss function in the process of training the mass estimation model; and

predict a maintenance timing suitable for a part influenced by a weight among expendables of the vehicle, which require combined maintenance, using the real-to-virtual transformation model and the mass estimation model;

wherein the virtual gradient is configured to indicate a gradient of the vehicle in a simulation environment; and

wherein the virtual velocity is configured to indicate a velocity of the vehicle in the simulation environment; and

wherein the virtual longitudinal acceleration is configured to indicate longitudinal acceleration of the vehicle in the simulation environment; and

wherein the real-to-virtual transformation model includes:

a first sub-model configured to obtain an initial virtual velocity from the real velocity;

a second sub-model configured to obtain the virtual longitudinal acceleration from the real velocity; and

a third sub-model configured to obtain the virtual gradient from a gradient of the vehicle.

2 . The apparatus of claim 1 , wherein the at least one processor is configured to apply the real velocity to the first sub-model to obtain the initial virtual velocity, to apply the real velocity to the second sub-model to obtain the virtual longitudinal acceleration, and to determine the virtual velocity, based on numerical integration of the virtual longitudinal acceleration and the initial virtual velocity.

3 . The apparatus of claim 2 , wherein the at least one processor is configured to determine a following gradient of the vehicle, based on the real acceleration and the virtual longitudinal acceleration, and applies the following gradient to the third sub-model to obtain the virtual gradient.

4 . The apparatus of claim 1 , wherein the first sub-model includes a deep convolutional neural network (DCNN), a flatten network, and a fully connected neural network (FCNN), wherein the second sub-model includes a recurrent neural network (RNN) including at least one long short term memory (LSTM) block, or wherein the third sub-model includes a DCNN and an FCNN.

5 . The apparatus of claim 1 , wherein the at least one processor is configured to train each of the first sub-model, the second sub-model, and the third sub-model included in the real-to-virtual transformation model, based on the first loss function; and

wherein the first loss function includes at least one of a mean square error based on the weight information of the vehicle, a mean square error based on the longitudinal acceleration of the vehicle, a mean square error based on the velocity of the vehicle, or a mean square error based on the gradient of the vehicle, or any combination thereof.

6 . The apparatus of claim 1 , wherein the second loss function includes at least one of a mean square error based on the weight information of the vehicle, a mean square error based on the ratio between the weight information of the vehicle and the real wheel torque of the vehicle, or a mean square error based on the resistance force of the vehicle, or any combination thereof.

7 . The apparatus of claim 1 , wherein the mass estimation model includes an FCNN associated with obtaining the weight information of the vehicle and a resistance force of the vehicle from the virtual output set at the target time point and the real wheel torque at the target time point.

8 . The apparatus of claim 1 , wherein the at least one processor is configured to:

train the mass estimation model, before training the real-to-virtual transformation model;

apply the real velocity and the real acceleration to the real-to-virtual transformation model to obtain a first temporary output at a time point different from the target time point, based on that the training of the mass estimation model is completed;

apply the first temporary output to the trained mass estimation model to obtain a second temporary output different from the first temporary output; and

apply the first temporary output and the second temporary output to a loss function of the real-to-virtual transformation model to train the real-to-virtual transformation model.

9 . A method for estimating a vehicle weight, the method comprising:

applying, by a processor, a real velocity and real acceleration to a trained real-to-virtual transformation model to obtain a virtual output set at a target time point, the virtual output set including a virtual gradient of a vehicle, a virtual velocity of the vehicle, and virtual longitudinal acceleration of the vehicle;

applying, by the processor, the virtual output set at the target time point and a real wheel torque at the target time point to a trained mass estimation model to obtain weight information of the vehicle;

determining a first loss function for training the real-to-virtual transformation model, based on at least one of weight information of the vehicle prior to the target time point, longitudinal acceleration of the vehicle prior to the target time point, a velocity of the vehicle prior to the target time point, or a gradient of the vehicle prior to the target time point, or any combination thereof; and

determining a second loss function for training the mass estimation model, based on weight information of the vehicle prior to the target time point, a ratio between the weight information of the vehicle and a real wheel torque of the vehicle prior to the target time point, and a resistance force of the vehicle prior to the target time point;

including longitudinal dynamics information in the second loss function in the process of training the mass estimation model; and

predicting a maintenance timing suitable for a part influenced by a weight among expendables of the vehicle, which require combined maintenance, using the real-to-virtual transformation model and the mass estimation model;

wherein the virtual gradient is configured to indicate a gradient of the vehicle in a simulation environment; and

wherein the virtual velocity is configured to indicate a velocity of the vehicle in the simulation environment; and

wherein the virtual longitudinal acceleration is configured to indicate longitudinal acceleration of the vehicle in the simulation environment; and

wherein the real-to-virtual transformation model includes:

a first sub-model configured to obtain an initial virtual velocity from the real velocity;

a second sub-model configured to obtain the virtual longitudinal acceleration from the real velocity; and

a third sub-model configured to obtain the virtual gradient from a gradient of the vehicle.

10 . The method of claim 9 , wherein obtaining the virtual output set includes:

applying the real velocity to the first sub-model to obtain the initial virtual velocity;

applying the real velocity to the second sub-model to obtain the virtual longitudinal acceleration; and

determining the virtual velocity, based on numerical integration of the virtual longitudinal acceleration and the initial virtual velocity.

11 . The method of claim 10 , further comprising:

determining a following gradient of the vehicle, based on the real acceleration and the virtual longitudinal acceleration; and

applying the following gradient to the third sub-model to obtain the virtual gradient.

12 . The method of claim 9 , wherein the first sub-model includes a deep convolutional neural network (DCNN), a flatten network, and a fully connected neural network (FCNN), wherein the second sub-model includes a recurrent neural network (RNN) including at least one long short term memory (LSTM) block, or wherein the third sub-model includes a DCNN and an FCNN.

13 . The method of claim 9 , wherein determining the first loss function includes:

training each of the first sub-model, the second sub-model, and the third sub-model included in the real-to-virtual transformation model, based on the first loss function; and

wherein the first loss function includes at least one of a mean square error based on the weight information of the vehicle, a mean square error based on the longitudinal acceleration of the vehicle, a mean square error based on the velocity of the vehicle, or a mean square error based on the gradient of the vehicle, or any combination thereof.

14 . The method of claim 9 , wherein the second loss function includes at least one of a mean square error based on the weight information of the vehicle, a mean square error based on the ratio between the weight information of the vehicle and the real wheel torque of the vehicle, or a mean square error based on the resistance force of the vehicle, or any combination thereof.

15 . The method of claim 9 , wherein the mass estimation model includes an FCNN associated with obtaining the weight information of the vehicle and a resistance force of the vehicle from the virtual output set at the target time point and the real wheel torque at the target time point.

16 . The method of claim 9 , further comprising:

training, by the processor, the mass estimation model, before training the real-to-virtual transformation model;

applying the real velocity and the real acceleration to the real-to-virtual transformation model to obtain a first temporary output at a time point different from the target time point, based on that the training of the mass estimation model is completed;

applying the first temporary output to the trained mass estimation model to obtain a second temporary output different from the first temporary output; and

applying the first temporary output and the second temporary output to a loss function of the real-to-virtual transformation model to train the real-to-virtual transformation model.