Methods, devices and equipment related to vehicle remaining mileage
The remaining mileage prediction model constructed by deep reinforcement learning and value iteration methods solves the problem of inaccurate prediction of remaining mileage during vehicle operation, achieves accurate prediction under the influence of multiple factors, reduces the difficulty of sample acquisition, and improves the accuracy and generalization ability of the model.
Patent Information
- Application Number
- CN202310252127.1
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- Filing Date
- 2023-03-15
- Publication Date
- 2026-11-17
- Estimated Expiration
- 2043-03-15
AI Technical Summary
In existing technologies, methods for predicting the remaining mileage of a vehicle are not accurate enough. In particular, due to the influence of various factors during vehicle operation, traditional statistical or machine learning models cannot effectively represent the remaining mileage in the actual driving process, and the cost of obtaining training samples is high, resulting in poor model performance.
By employing the Deep Reinforcement Learning (DQN) algorithm combined with the Value Iteration (VFI) method, a remaining mileage prediction model is constructed. This model utilizes the state features and final state features of the vehicle during its journey, constructs a loss function based on Bellman error, and iteratively trains the network parameters to achieve accurate prediction of the remaining mileage.
It improves the accuracy of remaining mileage prediction and the model's generalization ability, reduces the difficulty of obtaining training samples, and can more accurately predict the vehicle's remaining mileage under different driving conditions.
Smart Images

Figure CN116777033B_ABST
Abstract
Description
Technical Field
[0001] This application relates to, but is not limited to, the field of vehicle technology, and in particular to a method, apparatus, and device related to the remaining mileage of a vehicle. Background Technology
[0002] When driving, drivers need to pay attention to the parameters displayed on the vehicle's dashboard, especially the remaining range. For gasoline-powered vehicles, the remaining range refers to the distance the vehicle can travel continuously with the remaining fuel; for electric vehicles, it refers to the distance the vehicle can travel continuously with the remaining battery power. The accuracy and stability of the remaining range helps drivers use the vehicle rationally and accurately grasp its status, providing a reliable basis for route planning and switching driving styles, thus reducing range anxiety.
[0003] However, in related technologies, the remaining range is usually predicted based on a mapping relationship between the current remaining battery power (or current remaining fuel) and the remaining range. This mapping relationship is more suitable for predicting the remaining range of newly manufactured vehicles, thus leading to inaccuracies in the remaining range predicted using this mapping relationship. Therefore, it is necessary to study methods for estimating the remaining range displayed by the vehicle. Summary of the Invention
[0004] In view of this, embodiments of this application provide at least one method, apparatus, and device related to the remaining mileage of a vehicle.
[0005] The technical solution of this application embodiment is implemented as follows:
[0006] In a first aspect, embodiments of this application provide a training method for a vehicle remaining mileage prediction model. The method includes: training the remaining mileage prediction model using the start-state features and end-state features of each training sample in a training sample set, obtaining a first remaining mileage expectation corresponding to the start-state features and a second remaining mileage expectation corresponding to the end-state features of each training sample, wherein the start-state features and the end-state features are features corresponding to the start-state data and end-state data of a vehicle traveling a certain distance, respectively; determining a loss function for the remaining mileage prediction model based on the mileage traveled in each training sample corresponding to the start-state features and the end-state features, the first remaining mileage expectation, and the second remaining mileage expectation; and adjusting the network parameters of the remaining mileage prediction model using the loss function so that the loss function of the detection result output by the adjusted remaining mileage prediction model satisfies a convergence condition, wherein the remaining mileage prediction model after satisfying the convergence condition is used to predict the remaining mileage of the vehicle.
[0007] Secondly, embodiments of this application provide a method for predicting the remaining mileage of a vehicle. The method includes: acquiring the current state data of the vehicle; determining the current state features of the current state data; and inputting the current state features into the aforementioned trained remaining mileage prediction model to obtain the current remaining mileage of the vehicle.
[0008] Thirdly, embodiments of this application provide a training apparatus for a vehicle remaining mileage prediction model. The apparatus includes: a training module, used to train the remaining mileage prediction model using the start state features and end state features of each training sample in a training sample set, to obtain a first remaining mileage expectation corresponding to the start state feature and a second remaining mileage expectation corresponding to the end state feature of each training sample, wherein the start state feature and the end state feature are features corresponding to the start state data and end state data of a vehicle traveling a certain distance, respectively; a first determining module, used to determine the loss function of the remaining mileage prediction model based on the mileage traveled in each training sample corresponding to the start state feature and the end state feature, the first remaining mileage expectation, and the second remaining mileage expectation; and an adjustment module, used to adjust the network parameters of the remaining mileage prediction model using the loss function, so that the loss function of the detection result output by the adjusted remaining mileage prediction model satisfies the convergence condition, wherein the remaining mileage prediction model after satisfying the convergence condition is used to predict the remaining mileage of the vehicle.
[0009] Fourthly, embodiments of this application provide a computer device, including a memory and a processor, wherein the memory stores a computer program that can run on the processor, and the processor executes the program to implement some or all of the steps in the above-described method.
[0010] In this embodiment, firstly, the remaining mileage prediction model is trained using the start-state features and end-state features of each training sample in the training sample set, thereby obtaining the first remaining mileage expectation corresponding to the start-state feature and the second remaining mileage expectation corresponding to the end-state feature for each training sample; then, based on the mileage traveled, the first remaining mileage expectation, and the second remaining mileage expectation in each training sample, the loss function of the remaining mileage prediction model is determined; finally, the network parameters of the remaining mileage prediction model are adjusted using the loss function to make the remaining mileage prediction model converge.
[0011] The solution provided in this application has the following effects:
[0012] Firstly, since the remaining mileage of a vehicle is affected by a variety of factors before it has been driven or even before the destination and route planning are known, and the driver's driving situation will also be different under the influence of different factors, the remaining mileage expected in this application embodiment is more representative of the remaining mileage in the driver's actual use process than the remaining mileage after one drive.
[0013] Secondly, traditional methods for determining remaining mileage based on statistical or machine learning models evaluate the model's performance by comparing the actual remaining mileage with the predicted remaining mileage. During training, the remaining mileage from a specific real-world occurrence is typically used as the actual remaining mileage. This means that if the actual remaining mileage is inaccurate, the final trained model will also be biased. However, in this embodiment, a remaining mileage prediction model based on Deep QNetwork (DQN) combined with Value Function Iteration (VFI) is used. The loss function can be determined based on the Bellman error. Here, because the DQN algorithm is trained using VFI, the expected remaining mileage is continuously updated during the iteration process. Since there exists an actual expected remaining mileage that satisfies the Bellman equation, the Bellman error decreases continuously through model training, making the final expected remaining mileage infinitely close to the actual expected remaining mileage, thereby improving the model's accuracy.
[0014] Thirdly, in the embodiment of this application, the remaining mileage prediction model based on the DQN algorithm combined with VFI is adopted. The training process is in the form of value iteration. Therefore, the samples used for training do not have to be the samples obtained after the battery is depleted as mentioned in the related technology, but can be any driving mileage, which facilitates the collection of sample data.
[0015] It should be understood that the above general description and the following detailed description are merely exemplary and explanatory, and are not intended to limit the technical solutions of this disclosure. Attached Figure Description
[0016] The accompanying drawings, which are incorporated in and constitute a part of this specification, illustrate embodiments consistent with this application and, together with the specification, serve to explain the technical solutions of this application.
[0017] Figure 1 A schematic diagram illustrating the implementation process of a training method for a vehicle remaining mileage prediction model provided in this application embodiment;
[0018] Figure 2 A schematic diagram illustrating the implementation process of a method for determining start state characteristics and end state characteristics provided in an embodiment of this application;
[0019] Figure 3 A schematic diagram illustrating the implementation flow of another method for determining start state characteristics and end state characteristics provided in an embodiment of this application;
[0020] Figure 4 A schematic diagram illustrating the implementation process of a vehicle remaining mileage prediction method provided in this application embodiment;
[0021] Figure 5 A schematic diagram of the composition structure of a training device for a vehicle remaining mileage prediction model provided in an embodiment of this application;
[0022] Figure 6 This is a schematic diagram of the hardware entity of a computer device provided in an embodiment of this application. Detailed Implementation
[0023] To make the objectives, technical solutions, and advantages of this application clearer, the technical solutions of this application are further described in detail below with reference to the accompanying drawings and embodiments. The described embodiments should not be regarded as limitations on this application. All other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.
[0024] In the following description, references are made to “some embodiments,” which describe a subset of all possible embodiments. However, it is understood that “some embodiments” may be the same subset or different subsets of all possible embodiments and may be combined with each other without conflict.
[0025] The terms “first / second / third” are used merely to distinguish similar objects and do not represent a specific ordering of objects. It is understood that “first / second / third” may be interchanged in a specific order or sequence where permitted, so that the embodiments of this application described herein can be implemented in an order other than that illustrated or described herein.
[0026] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this application pertains. The terminology used herein is for descriptive purposes only and is not intended to limit the scope of this application.
[0027] Related technologies have been studied to some extent regarding the remaining mileage, including the following solutions:
[0028] Option 1: Calculate the remaining range based on the battery's remaining State of Charge (SOC) and the maximum range multiplied by the SOC.
[0029] Option 2: Correct the results of Option 1 using the battery's terminal voltage, SOC, and filtering to obtain the remaining range;
[0030] Option 3: Based on Option 2, add information such as speed, battery voltage / current, and energy consumption from the current moment and the previous period to predict the remaining range. The prediction method is a statistical or machine learning model.
[0031] Option 4: Based on Option 3, and with high-precision map navigation available, add features that predict future driving routes to predict the remaining mileage.
[0032] Among these options, Schemes 1 and 2 only utilize information such as SOC and terminal voltage, neglecting data like speed and energy consumption. This prevents them from accurately determining the vehicle's remaining range based on its current state, resulting in inaccurate estimates. In contrast, Schemes 3 and 4 incorporate more dimensions of information, including speed, energy consumption, voltage, and current, providing a more realistic representation of the vehicle's current state. By using a model to predict the remaining range, they achieve greater accuracy.
[0033] However, Schemes 3 and 4 typically evaluate the model's prediction quality by comparing the actual remaining mileage with the predicted remaining mileage. But before a vehicle has traveled, and even before knowing its destination and route planning, its remaining mileage is affected by various factors. Furthermore, the driver's driving behavior varies under different influences. Therefore, the expected remaining mileage is more representative of the remaining mileage in actual driving than the actual remaining mileage after a single trip. Thus, if statistical or machine learning models are used for training, since these methods construct a loss function by calculating the difference between the predicted and actual remaining mileage, and then iteratively update the model's parameters, the remaining mileage output by the trained model will closely approximate the actual remaining mileage. During training, the remaining mileage from a single real-world occurrence (i.e., the remaining mileage after the vehicle's battery is depleted) is typically taken as the actual remaining mileage. This means that if the remaining mileage is inaccurate, the final trained model will also be inaccurate. Even if there is no inaccuracy, the remaining mileage is only from a single occurrence and cannot represent the expected actual remaining mileage. Therefore, the remaining mileage output by the model after training will only be closer to the actual remaining mileage, rather than representing the remaining mileage in real-world use. At the same time, such training will significantly increase the model's variance, thereby affecting the model's performance.
[0034] Furthermore, obtaining the remaining mileage is very costly (the vehicle's battery must be completely depleted, otherwise it cannot represent the remaining mileage), and the variance of the model cannot be reduced by increasing the sample size (because after each time the battery is depleted, even if the battery is recharged to the previous value, the state of the vehicle is different in the two instances, so the sample size cannot be increased). As a result, this method cannot fit models with complex data dimensions, especially deep learning models, and its performance on actual models is also unsatisfactory.
[0035] Based on this, embodiments of this application provide a training method for a vehicle remaining mileage prediction model, which can be executed by a processor of a computer device. The computer device refers to a device with data processing capabilities, such as a server, laptop, tablet, desktop computer, smart TV, set-top box, or mobile device (e.g., mobile phone, portable video player, personal digital assistant, dedicated messaging device, portable gaming device). Figure 1 This application provides a schematic diagram illustrating the implementation process of a training method for a vehicle remaining mileage prediction model, as shown in the embodiments below. Figure 1 As shown, the method includes the following steps S101 to S103:
[0036] Step S101: Using the start state features and end state features of each training sample in the training sample set, train the remaining mileage prediction model to obtain the first remaining mileage expectation corresponding to the start state feature and the second remaining mileage expectation corresponding to the end state feature of each training sample, wherein the start state feature and the end state feature are the features corresponding to the start state data and the end state data of the vehicle traveling a certain distance.
[0037] Here, the vehicle can be a gasoline-powered car, a pure electric vehicle, a hybrid electric vehicle, a fuel cell vehicle, etc. The distance traveled by the vehicle can be any distance, such as a 5-kilometer journey from position A to position B. The start state data refers to the vehicle's state data at the beginning of this journey. For example, if the vehicle's starting position is position A, then the start state data is the vehicle's state data at position A. The end state data refers to the vehicle's state data at the end of this journey. For example, if the vehicle's ending position is position B, then the end state data is the vehicle's state data at position B. State data can include weather conditions, vehicle and energy consumption status, driver status, road conditions, etc. Weather conditions include sunny, rainy, foggy, snowy, etc. Vehicle and energy consumption status includes: the wear and tear of internal vehicle parts, whether different functions are activated, electrical performance parameters, energy consumption over different distances, such as whether autonomous driving is activated, battery charge, voltage, average energy consumption over 10 kilometers (km), average energy consumption over 5 kilometers, etc. Driver condition includes whether they are drowsy or alert. Road condition includes whether the road is flat or steep.
[0038] The start-state features and end-state features are the features corresponding to the start-state data and end-state data of the vehicle after traveling a certain distance. In other words, they are vectors that the model can use, obtained by transforming the start-state data and end-state data. The transformation method for the start-state data and end-state data depends on the data format of the start-state data and end-state data. In some embodiments, the data format of the start-state data and end-state data can be extracted and processed numerical values, mainly targeting vehicle and energy consumption status data. The numerical values can include two types of numbers: the first type is a specific number representing a certain magnitude, such as the battery voltage of 36 volts; the second type is an encoding representing a certain state, such as 1 for the autonomous driving function not being activated and 0 for the autonomous driving function being activated. This application does not limit the method of extracting and processing the raw data to obtain numerical values; relevant technologies can be consulted for implementation. In some embodiments, the data format of the start-state data and end-state data can be images, such as an image of the driver captured by a camera to determine the driver's state (fatigue or alertness); or an image of the road ahead captured by a camera to determine the road condition (flat or steep).
[0039] When the start and end state data are in the form of the first type of numerical values, normalization can be used to obtain the start and end state features respectively. When the start and end state data are in the form of the second type of numerical values, they can be obtained by direct concatenation. For example, the code for activating the autonomous driving function is 1, and the code for the tire pressure system not issuing a fault alarm is 0, then the concatenated state features are [1,0]. When the start and end state data are in the form of images, a classification network can be used to classify and process them to obtain the start and end state features respectively. This application does not limit the method for obtaining the start and end state features.
[0040] As mentioned above, the expected remaining mileage better represents the remaining mileage in actual application. The expected remaining mileage is essentially the expected value obtained after different driving modes. It's equivalent to a virtual driving scenario: assuming the driver completes the journey in one mode, returns to their previous state, and then drives in other modes multiple times, the expected remaining mileage is obtained by calculating the expected value. This is a recursive, iterative problem, and reinforcement learning is precisely the method used to solve such recursive, iterative problems. It can be seen that the remaining mileage problem is well-suited to the research scope of reinforcement learning. Multiple studies have shown that problems solvable using reinforcement learning methods achieve better results than those not solved using reinforcement learning. Therefore, this application's embodiments employ reinforcement learning to solve the remaining mileage problem.
[0041] The DQN algorithm, a method within reinforcement learning, uses a deep neural network to replace the value function. The network's input is state information, and its output is the value for each state. Since many state parameters are used to determine remaining mileage, including weather conditions, vehicle and energy consumption status, driver status, and road conditions, using a deep neural network to fit the value function can improve the model's generalization ability. Therefore, the remaining mileage prediction model in this embodiment can be a DQN-based model. By utilizing the DQN algorithm, the relationship between state parameters and the expected value (remaining mileage) can be better matched. This embodiment does not limit the type of remaining mileage prediction model; other models capable of achieving the above functions can also be used.
[0042] It should be noted that, typically, the value function defined by the DQN algorithm is Q(s,a), where a represents an action. However, the remaining mileage problem in this embodiment is a task of estimating the remaining mileage (i.e., value) based on the state, and therefore does not involve actions. Therefore, the value function of the remaining mileage model based on the DQN algorithm used in this embodiment is Q(s), which can also be denoted as V(s).
[0043] In this embodiment, the input to the remaining mileage prediction model is state features, and the output is the expected remaining mileage. When the starting state features are input into the remaining mileage prediction model, the output is a first expected remaining mileage corresponding to the starting state features; when the ending state features are input into the remaining mileage prediction model, the output is a second expected remaining mileage corresponding to the ending state features.
[0044] In some embodiments, the network parameter of the remaining mileage prediction model is θ. It takes start-state features and end-state features as input, and outputs a first expected remaining mileage and a second expected remaining mileage, respectively. That is, the model structures and network parameters for outputting the first and second expected remaining mileage are identical. In some embodiments, the network parameter θ of the remaining mileage prediction model can be adjusted to θ'. The remaining mileage prediction model with network parameter θ takes start-state features as input and outputs a first expected remaining mileage, while the remaining mileage prediction model with network parameter θ' takes end-state features as input and outputs a second expected remaining mileage. That is, the model structures for outputting the first and second expected remaining mileage are identical, but the network parameters are different. This application does not limit this specific implementation.
[0045] Step S102: Based on the driving mileage, the first expected remaining mileage, and the second expected remaining mileage corresponding to the start state feature and the end state feature in each training sample, determine the loss function of the remaining mileage prediction model;
[0046] Here, the mileage in each training sample corresponds to the start state feature and the end state feature. For example, if the start state feature and the end state feature are the features of the 5-kilometer distance traveled by the vehicle from position A to position B, then the mileage is 5 kilometers.
[0047] Since the true V function (i.e., the expected remaining mileage) satisfies the Bellman Equation: V(S b If ) = r + V(S'), then the expected remaining mileage before the vehicle starts driving = the actual mileage traveled in one segment + the expected remaining mileage after the vehicle starts driving. Where S... b S and S' represent the states of the vehicle before and after traveling a distance r, respectively.
[0048] The difference between the two sides of the above Bellman equation V(S)b The error )-(r+V(S')) is called the Bellman error or temporal difference (TD). The true V is the sum of all S's. b The Bellman error on S' is zero. Based on this idea, the remaining mileage prediction model can construct a loss function based on the aforementioned Bellman error, and use VFI (Variable Frequency Indicator) for iterative training to bring the Bellman error close to zero, thereby obtaining the expected remaining mileage closest to the true V and improving the model's accuracy.
[0049] In some embodiments, the expression for the loss function can be the following formula (1):
[0050] L(θ b )=∑(V(S b / θ b )-(r+V(S' / θ'))) 2 (1);
[0051] Where V(S|θ) is the neural network, and S is the input variable of the neural network (including the initial state feature S). b and the final state feature S'), θ is the parameter of the neural network (θ). b θ' represents the parameters of the neural network corresponding to the starting state after traveling a distance r, and θ' represents the parameters of the neural network corresponding to the ending state after traveling a distance r.
[0052] During iterative training using the optimizer's optimization algorithm, θ' is updated after a certain number of training samples. In some embodiments, θ' is synchronized with θ. b Perform a clone update, i.e., θ' equals θ. b In some embodiments, θ' = (1-λ)θ' + λθ b , θ b After substituting into the formula, we obtain θ', which is then updated.
[0053] In formula (1), r+V(S' / θ') is the target value, obtained by training V(S) b / θ b θ in ) b Value, such that V(S) b / θ b Try to get as close as possible to the target value of r+V(S' / θ').
[0054] In some embodiments, the optimization algorithm of the optimizer may include adaptive moment estimation (Adam), root mean square propagation (RMSProp), gradient descent, etc. The embodiments of this application do not limit the type of optimization algorithm.
[0055] The embodiments of this application do not limit the type of loss function, and can also be other regression loss functions that can evaluate the above-mentioned Bellman error, such as L1 loss.
[0056] Step S103: Using the loss function, adjust the network parameters of the remaining mileage prediction model so that the loss function of the detection result output by the adjusted remaining mileage prediction model satisfies the convergence condition, wherein the remaining mileage prediction model after satisfying the convergence condition is used to predict the remaining mileage of the vehicle.
[0057] Here, the network parameters are adjusted by the loss function to make the remaining mileage prediction model converge and complete the training of the model.
[0058] In some embodiments, the first expected remaining mileage is the expected remaining mileage output using the remaining mileage prediction model with first network parameters, and the second expected remaining mileage is the expected remaining mileage output using the remaining mileage prediction model with second network parameters.
[0059] Correspondingly, the implementation of step S103 may include the following steps S1031 and S1032:
[0060] Step S1031: Based on the loss function, by taking the derivative of the first network parameters, the adjusted first network parameters of the remaining mileage prediction model with the first network parameters are obtained;
[0061] Here, the first network parameter can be θ from formula (1) above. b The second network parameter can be θ' in the above formula (1).
[0062] In implementation, a batch of data can be taken first and substituted into the above formula (1). Then, θ in formula (1) can be adjusted. b Taking the derivative, we get θ b Value, to update V(S) b / θ b The network parameters in ) are finally θ b Substitute the value into the following formula (2):
[0063] θ'=(1-λ)θ'+λθ b (2);
[0064] According to the above formula (2), θ' is calculated and substituted into the above formula (1). Then, other batches of data are used for training until the model converges, and the network parameters corresponding to the model after training are obtained, that is, the adjusted first network parameters.
[0065] In some embodiments, θ can also be obtained by training using multiple batches of data. b Then update θ'. In some embodiments, θ' can also be equal to θ. b Instead of using the formula (2) above to calculate, θ' is updated.
[0066] Step S1032: Based on the adjusted first network parameters, adjust the network parameters of the remaining mileage prediction model with the first network parameters so that the loss function of the detection result output by the adjusted remaining mileage prediction model with the first network parameters satisfies the convergence condition.
[0067] In this embodiment, firstly, the remaining mileage prediction model is trained using the start-state features and end-state features of each training sample in the training sample set, thereby obtaining the first remaining mileage expectation corresponding to the start-state feature and the second remaining mileage expectation corresponding to the end-state feature for each training sample; then, based on the mileage traveled, the first remaining mileage expectation, and the second remaining mileage expectation in each training sample, the loss function of the remaining mileage prediction model is determined; finally, the network parameters of the remaining mileage prediction model are adjusted using the loss function to make the remaining mileage prediction model converge.
[0068] The solution provided in this application has the following effects:
[0069] Firstly, since the remaining mileage of a vehicle is affected by a variety of factors before it has been driven or even before the destination and route planning are known, and the driver's driving situation will also be different under the influence of different factors, the remaining mileage expected in this application embodiment is more representative of the remaining mileage in the driver's actual use process than the remaining mileage after one drive.
[0070] Secondly, traditional methods for determining remaining mileage based on statistical or machine learning models evaluate model performance by comparing actual remaining mileage with predicted remaining mileage. During training, the remaining mileage from a single real-world occurrence is typically used as the actual remaining mileage. This means that if the actual remaining mileage is inaccurate, the final trained model will also be inaccurate. However, in this embodiment, a remaining mileage prediction model based on the DQN algorithm combined with VFI is used, where the loss function can be determined based on the Bellman error. Here, because the DQN algorithm is trained using VFI, the expected remaining mileage is continuously updated during iteration. Since there exists an actual expected remaining mileage that satisfies the Bellman equation, through continuous model training, the Bellman error decreases, making the final expected remaining mileage infinitely close to the actual expected remaining mileage, thereby improving the model's accuracy.
[0071] Thirdly, in the embodiment of this application, the remaining mileage prediction model based on the DQN algorithm combined with VFI is adopted. The training process is in the form of value iteration. Therefore, the samples used for training do not have to be the samples obtained after the battery is depleted as mentioned in the related technology, but can be any driving mileage, which facilitates the collection of sample data.
[0072] In some embodiments, the method further includes the following steps S201 and S202:
[0073] Step S201: Obtain a training sample set; wherein the training samples in the training sample set include the start state data, the mileage traveled, and the end state data of the vehicle after traveling a certain distance;
[0074] In some embodiments, the state data in the start state data and the end state data includes weather conditions, vehicle and energy consumption conditions, driver conditions, and road conditions.
[0075] Correspondingly, the implementation of step S201, "obtaining the training sample set," includes the following steps:
[0076] Obtain a training sample set; wherein the training samples in the training sample set include: data on weather conditions, vehicle and energy consumption conditions, driver status and road conditions at the beginning of a vehicle's journey, the mileage traveled, and data on weather conditions, vehicle and energy consumption conditions, driver status and road conditions at the end of the journey.
[0077] Step S202: Based on the start state data and end state data of each training sample in the training sample set, determine the start state feature and end state feature of each training sample respectively.
[0078] In some embodiments, such as Figure 2 As shown, when both the start state data and the end state data are of the first type, the implementation of step S202 includes the following steps S301 to S303:
[0079] Step S301: For the start state data and end state data of each training sample, determine the state data of the same type in the training sample set;
[0080] Here, the first type of data can be the numerical value of the first type mentioned above, that is, a specific number representing a certain magnitude, such as the battery voltage of 36 volts. State data of the same type refers to the same data item in the state data of the training sample set. For example, if the initial state data of the training samples includes whether autonomous driving is enabled, battery level, voltage, average energy consumption over 10km, and average energy consumption over 5km, then the initial state data of each training sample in the training sample set includes whether autonomous driving is enabled, battery level, voltage, average energy consumption over 10km, and average energy consumption over 5km. Therefore, all the battery level data items in the training sample set belong to the same type of state data, or all the whether autonomous driving is enabled data items in the training sample set belong to the same type of state data.
[0081] Step S301 involves determining the state data of the same type for both the start state data and the end state data.
[0082] Step S302: For the start state data and end state data of each training sample, normalize the state data of the same type in the training sample set to obtain the start state number set and end state number set of each training sample.
[0083] Here, normalization, also known as min-max standardization or deviation standardization, is used to limit the original data to a certain range. Normalizing the state data of the same type in the training sample set yields normalized values. In practice, each value in each training sample is normalized with the state data of the same type in the training sample set, resulting in a new set of data. This effectively redefines all values and limits them to a certain range. The initial state data of each training sample, after normalization, yields the initial state set; the final state data, after normalization, yields the final state set.
[0084] Step S303: For each training sample, combine the numbers in the start state number set and the end state number set to obtain the start state feature and end state feature of each training sample.
[0085] The following example uses the set of starting state numbers = {0.5, 0.3, 0, 1} as an example. The resulting starting state feature after combination is [0.5, 0.3, 0, 1].
[0086] In this embodiment of the application, when both the start state data and the end state data are first type data (the first type of numerical value), the start state features and end state features can be obtained through normalization processing.
[0087] In some embodiments, both the start state data and the end state data are second-type data, and the remaining mileage prediction model includes a classification subnetwork and a prediction subnetwork based on the DQN algorithm, correspondingly, as shown below. Figure 3 As shown, the implementation of step S202 includes the following steps S501 and S502:
[0088] Step S501: Use the classification sub-network to classify the start state data and end state data of each training sample in the training sample set to obtain the classification result of the start state data and the classification result of the end state data of each training sample.
[0089] Here, the second type of data can be images. The classification sub-network can be a network used for image classification, such as the Visual Geometry Group (VGG) network, Googlenet, Residual Network 50 (ResNet50), MobileNetV2, etc. The classification result refers to the score of each category after classification processing. For example, if the object to be classified is an image of a driver to determine whether the driver is fatigued, the classification sub-network will output a score indicating whether the driver is fatigued or not. For example, the score for driver fatigue is 0.9, and the score for not being fatigued is 0.1.
[0090] Step S502: For each training sample, combine the classification results of the start state data and the classification results of the end state data to obtain the start state features and end state features of each training sample.
[0091] The following example illustrates the classification results output by the classification subnetwork: the score for fatigue is 0.9 and the score for non-fatigue is 0.1. The combined classification results yield the state features [0.9, 0.1].
[0092] Correspondingly, the implementation of step S101 may include:
[0093] The prediction sub-model based on the DQN algorithm is trained using the start state features and end state features of each training sample in the training sample set, so as to obtain the first remaining mileage expectation corresponding to the start state feature and the second remaining mileage expectation corresponding to the end state feature of each training sample.
[0094] In this embodiment of the application, when both the start state data and the end state data are second-type data (images), the classification results are output through the classification sub-network, and the start state features and end state features can be obtained by combining the classification results.
[0095] In some embodiments, the start state data and end state data may simultaneously include first type data and second type data. Therefore, step S202, "Based on the start state data and end state data of each training sample in the training sample set, determine the start state features and end state features of each training sample respectively," may include: obtaining the state features corresponding to the first type data and the second type data according to the aforementioned method for determining start state features and end state features when the start state data and end state data are first type data or second type data; and then concatenating the state features of each part to obtain the final start state feature or end state feature. In some embodiments, to improve the accuracy of the model, dimensionality reduction processing may be performed on the concatenated state features after concatenating each part.
[0096] This application also provides a method for training a vehicle remaining mileage prediction model, the method comprising:
[0097] Using the DQN method, a remaining mileage model V(s) (i.e., the remaining mileage prediction model mentioned above) is trained based on arbitrary start-up state data, end-up state data, and mileage data. V(s) is defined as the remaining mileage of the vehicle based on all possible travel paths of the current state vector. In reinforcement learning or dynamic programming, V(s) is generally referred to as the value function.
[0098] In the above V(s), s is a state vector (e.g., remaining battery power, whether the battery level is new, how many kilometers the battery has traveled, etc.), and s_t represents the state vector at a certain time t, i.e., it is acquired at time t. s can be a vector obtained after preprocessing the original data, such as feature extraction, principal component analysis (PCA) dimensionality reduction; or a variable obtained by training a sub-model based on other tasks (e.g., images, maps, autoencoder models, etc.). V can be a deep neural network based on s.
[0099] Training process:
[0100] Define a state vector s, obtain multiple segments of car trip data, excluding data from when the battery is depleted, use the result of any single trip as the dataset, define a loss function and an optimizer, divide the dataset into multiple batches of fixed size, and randomly shuffle the dataset;
[0101] For each batch of data:
[0102] Calculate the Bellman error for a batch of data, and update the model parameters based on the forward and backward feedback mechanism of the neural network.
[0103] It should be noted that the DQN algorithm defined in this embodiment may differ from the definitions of some other DQN algorithms. Some DQN algorithms define the value function as Q(s,a), where a represents an action. Since the remaining mileage problem in this embodiment is a task of estimating remaining mileage (value) based on data or state, there are no actions involved.
[0104] Furthermore, similar to the DQN algorithm is the Value Function Iteration (VFI) method in dynamic programming. This method also uses the Bellman equation to build the model, but in terms of model updates, it iteratively updates the V function using the right-hand side of the Bellman equation until the V function converges. This method generally requires the assumption that the V function is a convex function (a linear function, for example) to achieve convergence. Since the s-state used to predict the remaining mileage may have a wide range of dimensions in actual modeling, involving various in-cabin and out-of-cabin information, driving history, and map planning fusion information, it is difficult to guarantee that the scope of the Convex function can fit the true V function of the remaining mileage. Therefore, in the embodiments of this application, a deep neural network is used to fit the V function, and deep learning methods are used to optimize the neural network, making the model more generalizable.
[0105] The method provided in this application embodiment has the following effects:
[0106] First, by utilizing reinforcement learning methods (DQN algorithm combined with VFI), it is possible to use data from any driving path, rather than data from driving paths that terminate due to energy depletion. This solves the problem of increased data collection difficulty in reality, where driving is required to terminate due to energy depletion.
[0107] Second, in methods that estimate the remaining mileage, if supervised learning algorithms are used for training with data from arbitrary driving paths, the discrepancy between the estimated and actual remaining mileage can lead to deviations or increased errors in the final model output. Unlike supervised learning, reinforcement learning algorithms do not require knowledge of the actual remaining mileage, thus improving the model's accuracy.
[0108] This application also provides a method for predicting the remaining mileage of a vehicle, such as... Figure 4 As shown, the method includes the following steps S701 to S703:
[0109] Step S701: Obtain the current status data of the vehicle;
[0110] Step S702: Determine the current state characteristics of the current state data;
[0111] Step S703: Input the current state features into the trained remaining mileage prediction model to obtain the current remaining mileage of the vehicle.
[0112] In some embodiments, the method further includes the following steps S801 to S803:
[0113] Step S801: Output the location information of at least one charging device within the current remaining mileage range;
[0114] Step S802: Based on the location information of the at least one charging device, determine and output the target charging device among the at least one charging device;
[0115] Here, the target charging device can be the charging device closest to the vehicle.
[0116] In some embodiments, the target charging device may also be the charging station with the fastest charging efficiency. Correspondingly, before determining and outputting the target charging device, the method further includes: obtaining the charging efficiency of the at least one charging device; and determining and outputting the target charging device among the at least one charging device based on the charging efficiency of the at least one charging device.
[0117] Step S803: Automatically navigate the vehicle to the target charging device.
[0118] Based on the foregoing embodiments, this application provides a training device for a vehicle remaining mileage prediction model. The device includes various units and modules included in each unit, which can be implemented by a processor in a computer device; of course, it can also be implemented by specific logic circuits. In the implementation process, the processor can be a central processing unit (CPU), a microprocessor unit (MPU), a digital signal processor (DSP), or a field programmable gate array (FPGA), etc.
[0119] Figure 5 This is a schematic diagram of the composition of a training device for a vehicle remaining mileage prediction model provided in an embodiment of this application, as shown below. Figure 5 As shown, the device 600 includes: a training module 610, a first determination module 620, and an adjustment module 630, wherein:
[0120] The training module 610 is used to train the remaining mileage prediction model by using the start state features and end state features of each training sample in the training sample set, and to obtain the first remaining mileage expectation corresponding to the start state feature and the second remaining mileage expectation corresponding to the end state feature of each training sample, wherein the start state feature and the end state feature are the features corresponding to the start state data and the end state data of the vehicle traveling a certain distance.
[0121] The first determining module 620 is used to determine the loss function of the remaining mileage prediction model based on the driving mileage corresponding to the start state feature and the end state feature in each training sample, the first remaining mileage expectation and the second remaining mileage expectation;
[0122] The adjustment module 630 is used to adjust the network parameters of the remaining mileage prediction model using the loss function, so that the loss function of the detection result output by the adjusted remaining mileage prediction model satisfies the convergence condition, wherein the remaining mileage prediction model after satisfying the convergence condition is used to predict the remaining mileage of the vehicle.
[0123] In some embodiments, the first remaining mileage expectation is the remaining mileage expectation output by the remaining mileage prediction model with first network parameters, and the second remaining mileage expectation is the remaining mileage expectation output by the remaining mileage prediction model with second network parameters. Correspondingly, the adjustment module 630 includes:
[0124] The derivative submodule, based on the loss function, obtains the adjusted first network parameters of the remaining mileage prediction model by differentiating the first network parameters with the first network parameters;
[0125] The adjustment submodule is used to adjust the network parameters of the remaining mileage prediction model with the first network parameters based on the adjusted first network parameters, so that the loss function of the detection result output by the adjusted remaining mileage prediction model with the first network parameters satisfies the convergence condition.
[0126] In some embodiments, the apparatus further includes: an acquisition module, configured to acquire a training sample set; wherein the training samples in the training sample set include the start state data, the mileage traveled, and the end state data of the vehicle after traveling a certain distance; and a second determination module, configured to determine the start state features and end state features of each training sample based on the start state data and end state data of each training sample in the training sample set.
[0127] In some embodiments, when both the start state data and the end state data are of the first type, the second determining module includes: a first determining submodule, configured to determine state data of the same type in the training sample set for each training sample's start state data and end state data respectively; a normalization submodule, configured to normalize the state data of the same type in the training sample set for each training sample's start state data and end state data respectively, to obtain a start state number set and an end state number set for each training sample; and a first combining submodule, configured to combine the numbers in the start state number set and the end state number set for each training sample respectively, to obtain a start state feature and an end state feature for each training sample.
[0128] In some embodiments, when both the start state data and the end state data are second type data, the remaining mileage prediction model includes a classification sub-network and a prediction sub-network based on the DQN algorithm. Correspondingly, the second determining module includes: a classification sub-module, used to classify the start state data and end state data of each training sample in the training sample set using the classification sub-network, to obtain the classification result of the start state data and the classification result of the end state data of each training sample; a second combination sub-module, used to combine the classification result of the start state data and the classification result of the end state data for each training sample, to obtain the start state feature and the end state feature of each training sample; the training module is further used to: train the prediction sub-model based on the DQN algorithm using the start state feature and the end state feature of each training sample in the training sample set, to obtain the first expected remaining mileage corresponding to the start state feature and the second expected remaining mileage corresponding to the end state feature of each training sample.
[0129] In some embodiments, the state data in the start state data and the end state data includes weather state, vehicle and energy consumption state, driver state, and road state. Correspondingly, the acquisition module is further configured to acquire a training sample set; wherein, the training samples in the training sample set include: data on weather state, vehicle and energy consumption state, driver state, and road state at the beginning of a vehicle's journey of a certain distance, the mileage traveled, and data on weather state, vehicle and energy consumption state, driver state, and road state at the end of the journey.
[0130] The descriptions of the above apparatus embodiments are similar to those of the above method embodiments, and have similar beneficial effects. The functions or modules included in the apparatus provided in this disclosure can be used to execute the methods described in the above method embodiments. For technical details not disclosed in the apparatus embodiments of this application, please refer to the descriptions of the method embodiments of this application for understanding.
[0131] If the technical solution of this application involves personal information, the product using this technical solution has clearly informed the user of the personal information processing rules and obtained the user's voluntary consent before processing the personal information. If the technical solution of this application involves sensitive personal information, the product using this technical solution has obtained the user's separate consent before processing the sensitive personal information, and also meets the requirement of "express consent". For example, at personal information collection devices such as cameras, clear and prominent signs are set up to inform users that they have entered the scope of personal information collection and that personal information will be collected. If an individual voluntarily enters the collection scope, it is deemed that they have agreed to the collection of their personal information; or on the personal information processing device, with clear signs / information informing users of the personal information processing rules, authorization is obtained from the user through pop-up information or by asking the user to upload their personal information; wherein, the personal information processing rules may include information such as the personal information processor, the purpose of personal information processing, the processing method, and the types of personal information processed.
[0132] It should be noted that, in the embodiments of this application, if the training method for the vehicle remaining mileage prediction model described above is implemented as a software functional module and sold or used as an independent product, it can also be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the embodiments of this application, or the part that contributes to the related technology, can be embodied in the form of a software product. This software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the methods described in the various embodiments of this application. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), magnetic disks, or optical disks. Thus, the embodiments of this application are not limited to any specific hardware, software, or firmware, or any combination of hardware, software, and firmware.
[0133] This application provides a computer device including a memory and a processor. The memory stores a computer program that can run on the processor. When the processor executes the program, it implements some or all of the steps in the above-described method.
[0134] This application provides a computer-readable storage medium storing a computer program thereon, which, when executed by a processor, implements some or all of the steps in the above-described method. The computer-readable storage medium can be transient or non-transient.
[0135] This application provides a computer program including computer-readable code, wherein when the computer-readable code is executed in a computer device, a processor in the computer device performs some or all of the steps in the above-described method.
[0136] This application provides a computer program product, which includes a non-transitory computer-readable storage medium storing a computer program. When the computer program is read and executed by a computer, it implements some or all of the steps in the above-described method. This computer program product can be implemented specifically through hardware, software, or a combination thereof. In some embodiments, the computer program product is specifically embodied as a computer storage medium; in other embodiments, the computer program product is specifically embodied as a software product, such as a software development kit (SDK), etc.
[0137] It should be noted that the descriptions of the various embodiments above tend to emphasize the differences between them, while their similarities or commonalities can be referred to interchangeably. The descriptions of the above embodiments of the device, storage medium, computer program, and computer program product are similar to the descriptions of the above method embodiments and have similar beneficial effects. For technical details not disclosed in the embodiments of the device, storage medium, computer program, and computer program product of this application, please refer to the descriptions of the method embodiments of this application for understanding.
[0138] It should be noted that, Figure 6 This application provides a hardware entity diagram of a computer device as an embodiment of the present application, such as... Figure 6 As shown, the hardware entity of the computer device 700 includes: a processor 701, a communication interface 702, and a memory 703, wherein:
[0139] Processor 701 typically controls the overall operation of computer device 700.
[0140] Communication interface 702 enables computer devices to communicate with other terminals or servers over a network.
[0141] The memory 703 is configured to store instructions and applications executable by the processor 701, and can also cache data to be processed or already processed (e.g., image data, audio data, voice communication data, and video communication data) in the processor 701 and various modules in the computer device 700. It can be implemented using flash memory or random access memory (RAM). Data transfer between the processor 701, the communication interface 702, and the memory 703 can be performed via bus 704.
[0142] It should be understood that the phrase "one embodiment" or "an embodiment" throughout the specification means that a specific feature, structure, or characteristic related to the embodiment is included in at least one embodiment of this application. Therefore, "in one embodiment" or "in an embodiment" appearing throughout the specification does not necessarily refer to the same embodiment. Furthermore, these specific features, structures, or characteristics can be combined in any suitable manner in one or more embodiments. It should be understood that in the various embodiments of this application, the sequence numbers of the above steps / processes do not imply a sequential order of execution; the execution order of each step / process should be determined by its function and internal logic, and should not constitute any limitation on the implementation process of the embodiments of this application. The sequence numbers of the above embodiments of this application are merely descriptive and do not represent the superiority or inferiority of the embodiments.
[0143] It should be noted that, in this document, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus. Unless otherwise specified, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes that element.
[0144] In the several embodiments provided in this application, it should be understood that the disclosed devices and methods can be implemented in other ways. The device embodiments described above are merely illustrative. For example, the division of units is only a logical functional division, and in actual implementation, there may be other division methods, such as: multiple units or components can be combined, or integrated into another system, or some features can be ignored or not executed. In addition, the coupling, direct coupling, or communication connection between the various components shown or discussed can be through some interfaces, and the indirect coupling or communication connection between devices or units can be electrical, mechanical, or other forms.
[0145] The units described above as separate components may or may not be physically separate. The components shown as units may or may not be physical units. They may be located in one place or distributed across multiple network units. Some or all of the units may be selected to achieve the purpose of this embodiment according to actual needs.
[0146] In addition, each functional unit in the various embodiments of this application can be integrated into one processing unit, or each unit can be a separate unit, or two or more units can be integrated into one unit; the integrated unit can be implemented in hardware or in the form of hardware plus software functional units.
[0147] Those skilled in the art will understand that all or part of the steps of the above method embodiments can be implemented by hardware related to program instructions. The aforementioned program can be stored in a computer-readable storage medium. When the program is executed, it performs the steps of the above method embodiments. The aforementioned storage medium includes various media that can store program code, such as mobile storage devices, read-only memory (ROM), magnetic disks, or optical disks.
[0148] Alternatively, if the integrated units described above are implemented as software functional modules and sold or used as independent products, they can also be stored in a computer-readable storage medium. Based on this understanding, the technical solution of this application, in essence or the part that contributes to related technologies, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the methods described in the various embodiments of this application. The aforementioned storage medium includes various media capable of storing program code, such as mobile storage devices, ROM, magnetic disks, or optical disks.
[0149] The above description is merely an embodiment of this application, but the scope of protection of this application is not limited thereto. Any changes or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in this application should be included within the scope of protection of this application.
Claims
1. A training method for a vehicle remaining mileage prediction model, characterized in that, include: The remaining mileage prediction model is trained using the start state features and end state features of each training sample in the training sample set, and the first remaining mileage expectation corresponding to the start state feature and the second remaining mileage expectation corresponding to the end state feature of each training sample are obtained respectively. The start state feature and the end state feature are the features corresponding to the start state data and the end state data of the vehicle traveling a certain distance. Based on the mileage, the first expected remaining mileage, and the second expected remaining mileage corresponding to the start state feature and the end state feature in each training sample, the loss function of the remaining mileage prediction model is determined; the loss function is determined based on Bellman error; the Bellman error is the difference between the first expected remaining mileage and the sum of the mileage and the second expected remaining mileage. The network parameters of the remaining mileage prediction model are adjusted using the aforementioned loss function so that the loss function of the detection results output by the adjusted remaining mileage prediction model satisfies the convergence condition. The remaining mileage prediction model that satisfies the convergence condition is used to predict the remaining mileage of the vehicle.
2. The method of claim 1, wherein, The first expected remaining mileage is the expected remaining mileage output using the remaining mileage prediction model with the first network parameters, and the second expected remaining mileage is the expected remaining mileage output using the remaining mileage prediction model with the second network parameters. Correspondingly, adjusting the network parameters of the remaining mileage prediction model using the loss function, so that the loss function of the detection result output by the adjusted remaining mileage prediction model satisfies the convergence condition, includes: Based on the loss function, by taking the derivative of the first network parameters, the adjusted first network parameters of the remaining mileage prediction model with the first network parameters are obtained; Based on the adjusted first network parameters, the network parameters of the remaining mileage prediction model with the first network parameters are adjusted so that the loss function of the detection result output by the adjusted remaining mileage prediction model with the first network parameters satisfies the convergence condition.
3. The method according to claim 1 or 2, characterized in that, Also includes: Obtain a training sample set; wherein, the training samples in the training sample set include the start state data, the mileage traveled, and the end state data of the vehicle after traveling a certain distance; Based on the start state data and end state data of each training sample in the training sample set, the start state features and end state features of each training sample are determined respectively.
4. The method according to claim 3, characterized in that, When both the start state data and the end state data are of the first type, determining the start state feature and end state feature of each training sample based on the start state data and end state data of each training sample in the training sample set includes: For each training sample's start state data and end state data, determine the state data belonging to the same type in the training sample set; For the start state data and end state data of each training sample, the state data of the same type in the training sample set are normalized to obtain the start state number set and end state number set of each training sample. For each training sample, the numbers in the start state number set and the end state number set are combined to obtain the start state feature and end state feature of each training sample.
5. The method according to claim 3, characterized in that, When both the start state data and the end state data are of the second type, the remaining mileage prediction model includes a classification subnetwork and a prediction subnetwork based on the DQN algorithm. Correspondingly, The step of determining the start-state features and end-state features of each training sample based on the start-state data and end-state data of each training sample in the training sample set includes: The classification subnetwork is used to classify the start state data and end state data of each training sample in the training sample set, respectively, to obtain the classification result of the start state data and the classification result of the end state data of each training sample. For each training sample, the classification results of the start state data and the classification results of the end state data are combined to obtain the start state features and end state features of each training sample. The remaining mileage prediction model is trained using the start-state features and end-state features of each training sample in the training sample set, to obtain the first expected remaining mileage corresponding to the start-state features and the second expected remaining mileage corresponding to the end-state features of each training sample, including: The prediction sub-model based on the DQN algorithm is trained using the start state features and end state features of each training sample in the training sample set, so as to obtain the first remaining mileage expectation corresponding to the start state feature and the second remaining mileage expectation corresponding to the end state feature of each training sample.
6. The method according to claim 3, characterized in that, The state data in the start state data and the end state data include weather conditions, vehicle and energy consumption conditions, driver conditions, and road conditions. Correspondingly, obtaining the training sample set includes: Obtain a training sample set; wherein the training samples in the training sample set include: data on weather conditions, vehicle and energy consumption conditions, driver status and road conditions at the beginning of a vehicle's journey, the mileage traveled, and data on weather conditions, vehicle and energy consumption conditions, driver status and road conditions at the end of the journey.
7. A method for predicting the remaining mileage of a vehicle, characterized in that, include: Obtain the current status data of the vehicle; Determine the current state characteristics of the current state data; The current state features are input into the trained remaining mileage prediction model according to any one of claims 1 to 6 to obtain the current remaining mileage of the vehicle.
8. The method according to claim 7, characterized in that, Also includes: Output the location information of at least one charging device within the current remaining mileage range; Based on the location information of the at least one charging device, a target charging device is determined and output among the at least one charging device; The vehicle is automatically navigated to the target charging device.
9. A training device for a vehicle remaining mileage prediction model, characterized in that, include: The training module is used to train the remaining mileage prediction model using the start state features and end state features of each training sample in the training sample set, and to obtain the first remaining mileage expectation corresponding to the start state feature and the second remaining mileage expectation corresponding to the end state feature of each training sample, wherein the start state feature and the end state feature are the features corresponding to the start state data and the end state data of the vehicle traveling a certain distance. The first determining module is used to determine the loss function of the remaining mileage prediction model based on the driving mileage, the first remaining mileage expectation, and the second remaining mileage expectation corresponding to the start state feature and the end state feature in each training sample; the loss function is determined based on Bellman error; the Bellman error is the difference between the first remaining mileage expectation and the sum of the driving mileage and the second remaining mileage expectation. An adjustment module is used to adjust the network parameters of the remaining mileage prediction model using the loss function, so that the loss function of the detection result output by the adjusted remaining mileage prediction model satisfies the convergence condition, wherein the remaining mileage prediction model after satisfying the convergence condition is used to predict the remaining mileage of the vehicle.
10. A computer device comprising a memory and a processor, the memory storing a computer program executable on the processor, characterized in that, When the processor executes the program, it implements the steps of the method according to any one of claims 1 to 8.
Citation Information
Patent Citations
DQN-based vehicle automatic driving path planning method and device
CN110646009A
Vehicle endurance mileage prediction method and device, vehicle and storage medium
CN112549970A
