System for estimating vehicle power consumption and method for estimating vehicle power consumption
The system addresses inefficiencies in vehicle power consumption estimation by using mean and standard deviation to determine necessary learning, enhancing estimation accuracy through targeted machine learning.
Patent Information
- Authority / Receiving Office
- JP · JP
- Patent Type
- Applications
- Current Assignee / Owner
- TOYOTA JIDOSHA KK
- Filing Date
- 2024-11-07
- Publication Date
- 2026-05-19
AI Technical Summary
Existing vehicle power consumption estimation systems perform unnecessary learning at inappropriate times, leading to inefficiencies.
A system that includes a first storage unit for prediction errors, a second storage unit for input/output pairs, and a learning unit that determines the necessity of learning based on mean and standard deviation of prediction errors, performing learning only when improvement in estimation accuracy is expected.
Accurately determines the need for learning and improves estimation accuracy by selectively performing machine learning on vehicle power consumption data.
Smart Images
Figure 2026082308000001_ABST
Abstract
Description
Technical Field
[0001] The present invention relates to a system for estimating the power consumption of a vehicle and a method for estimating the power consumption of a vehicle.
Background Art
[0002] Patent Document 1 describes a device that acquires appropriate fuel consumption or electricity cost suitable for the driving situation of a vehicle and the driver.
Prior Art Documents
Patent Documents
[0003]
Patent Document 1
Summary of the Invention
Problems to be Solved by the Invention
[0004] However, since the device of Patent Document 1 is set to learn at each elapsed time, learning is performed even at timings when learning is unnecessary. Therefore, an object of the present disclosure is to provide a system for estimating the power consumption of a vehicle that appropriately determines the necessity of learning and performs learning.
Means for Solving the Problems
[0005] The system for estimating the power consumption of a vehicle according to the present disclosure includes a first storage unit that stores a prediction error of the power consumption of the vehicle, a second storage unit that stores pairs of inputs and outputs of a model for estimating the power consumption of the vehicle, and a learning unit that, when the estimation accuracy of the model for estimating the power consumption of the vehicle is poor, determines whether improvement of the estimation accuracy can be expected by learning using the second storage unit, using the average and standard deviation of the prediction error of the power consumption of the vehicle in the first storage unit, and performs learning only when improvement of the estimation accuracy can be expected.
[0006] The above configuration provides a system that appropriately determines the need for learning and estimates the power consumption of the vehicle to be learned. Learning is performed using machine learning.
[0007] The system for estimating the power consumption of the vehicle in this disclosure is: When the mean is small and the standard deviation is small, no improvement in estimation accuracy can be expected. When the mean is large and the standard deviation is small, an improvement in the estimation accuracy can be expected. The method is characterized by determining that no improvement in estimation accuracy can be expected when the mean is small and the standard deviation is large.
[0008] The above configuration is an example of a decision-making method that uses the mean and standard deviation of prediction errors.
[0009] The system for estimating the power consumption of the vehicle in this disclosure is: The first storage unit and the second storage unit are characterized by storing data for each driver or each vehicle.
[0010] The above configuration allows for consideration of variations in power consumption caused by the driver or vehicle.
[0011] The system for estimating the power consumption of the vehicle in this disclosure is: This method is characterized by calculating the mean and standard deviation of the prediction error for the power consumption of the vehicle, excluding abnormal data on the vehicle's power consumption.
[0012] The above configuration provides a system that can more accurately determine the need for learning and estimate the power consumption of the vehicle undergoing learning.
[0013] The method for estimating the power consumption of a vehicle in this disclosure is: A first storage unit for storing prediction errors in the vehicle's power consumption, It comprises a second storage unit that stores input and output pairs for a model that estimates the power consumption of a vehicle, When the estimation accuracy of the model for estimating the power consumption of the vehicle is poor, it is determined using the average and standard deviation of the prediction error of the power consumption of the vehicle in the first storage unit whether the estimation accuracy can be improved by learning using the second storage unit, and only when the improvement of the estimation accuracy is expected is learned. This is a method for estimating the power consumption of a vehicle.
[0014] With the above configuration, a method for estimating the power consumption of a vehicle is provided that appropriately determines the necessity of learning and then learns.
Advantages of the Invention
[0015] According to the present disclosure, a system and the like for estimating the power consumption of a vehicle that appropriately determines the necessity of learning and then learns are provided.
Brief Description of the Drawings
[0016] [Figure 1A] It is the first flowchart of the method for estimating the power consumption of a vehicle according to the embodiment. [Figure 1B] It is the second flowchart of the method for estimating the power consumption of a vehicle according to the embodiment. [Figure 1C] It is the third flowchart of the method for estimating the power consumption of a vehicle according to the embodiment. [Figure 2] It is a diagram showing (A) the flowchart and (B) the prediction error when not learning in the first case of the method for estimating the power consumption of a vehicle according to the embodiment. [Figure 3] It is a diagram showing (A) the flowchart and (B) the prediction error when learning in the case of the method for estimating the power consumption of a vehicle according to the embodiment. [Figure 4] It is a diagram showing (A) the flowchart and (B) the prediction error when not learning in the second case of the method for estimating the power consumption of a vehicle according to the embodiment. [Figure 5A] It is a diagram showing an example of learning data for each driver in the method for estimating the power consumption of a vehicle according to the embodiment. [Figure 5B] It is a diagram showing an example of learning excluding abnormal data in the method for estimating the power consumption of a vehicle according to the embodiment. [Figure 6]It is a block diagram showing the configuration of a system for estimating the power consumption of a vehicle according to an embodiment.
Embodiment for Carrying Out the Invention
[0017] Embodiment Hereinafter, embodiments of the present invention will be described with reference to the drawings. However, the invention according to the claims is not limited to the following embodiments. Also, not all of the configurations described in the embodiments are essential as means for solving the problems. For the sake of clarity of explanation, the following description and drawings have been appropriately omitted and simplified. In each drawing, the same reference numerals are assigned to the same elements, and duplicate explanations are omitted as necessary.
[0018] (Explanation of a method for estimating the power consumption of a vehicle according to an embodiment) FIG. 1A is a first flowchart of a method for estimating the power consumption of a vehicle according to an embodiment. FIG. 1B is a second flowchart of a method for estimating the power consumption of a vehicle according to an embodiment. FIG. 1C is a third flowchart of a method for estimating the power consumption of a vehicle according to an embodiment. A method for estimating the power consumption of a vehicle according to an embodiment will be described while referring to FIGS. 1A to 1C. The method for estimating the power consumption of a vehicle according to an embodiment uses machine learning. The vehicle will be described as an electric vehicle that runs on electricity, but it may also be a vehicle that runs on gasoline or diesel.
[0019] As shown in FIG. 1A, first, input information of the model and weights specific to each vehicle are acquired (step S101). The model uses artificial intelligence (Artificial Intelligence). The machine learning of the model uses deep learning such as a neural network. Therefore, input information of the model and unique weights for each vehicle are acquired. However, since the model only needs to be learnable, it is not limited to deep learning. The input of the model is road information and vehicle information in the driving section, etc.
[0020] Next, the predicted value of power consumption is calculated (step S102). The model output is power consumption. However, while hybrid or electric vehicles output power consumption, gasoline or diesel vehicles output fuel consumption. A method for estimating a vehicle's power consumption is to calculate a predicted value of fuel consumption or electric energy consumption. Thus, it is desirable for the model to be pre-trained using driving data from multiple vehicles. If training is performed using initial weights, it is expected that the model will return prediction results for average driving behavior. In the case of average driving behavior, the prediction error will be small, and it can be determined that training is unnecessary.
[0021] Next, start driving (step S103). Drive while collecting driving data. Next, end driving (step S104). Next, split the driving data (step S105). The driving data may be split as needed.
[0022] Next, the driving data is input to the driver detector (step S106). The driver detector can be used to separate the data into queues based on the driver. The queue is a storage unit or memory unit. An in-vehicle camera or similar can be used to identify the driver. Even with the same vehicle, driving tendencies may differ depending on the driver, allowing for a more accurate determination of the need for learning.
[0023] Next, the driving data is input into the anomaly data detector (step S107). The anomaly data detector can also be used to delete anomaly data. Anomaly data may include driving a route significantly different from the suggested route, traffic congestion due to an accident, or prolonged parking during the drive. By not using anomaly data in the evaluation, the need for learning can be accurately determined.
[0024] As shown in Figure 1B, the next step is to determine whether the data is abnormal or not (step S108). If it is determined to be abnormal data (if YES in step S108), the data is deleted (step S109). If it is not determined to be abnormal data (if NO in step S108), the measured value and predicted value of power consumption are compared for each data point and the prediction error is calculated (step S110).
[0025] After step S110, the calculated prediction error is added to queue 1, which stores the prediction error on a per-vehicle basis (step S111). The prediction error is added to queue 1, which is the first storage unit. The first storage unit has a limited capacity.
[0026] Next, the information to be input to the model and the measured power consumption are added to Queue 2, which stores model input / output pairs on a per-vehicle basis (step S112). The information to be input to the model and the measured power consumption, which is the output, are added to Queue 2, which is the second storage unit. The second storage unit has a limited capacity. The limitations on the capacity of the first and second storage units allow the system to be made smaller. However, the system must learn efficiently. Therefore, it is necessary to select and filter the data to be learned.
[0027] As shown in Figure 1C, the next step is to determine whether queue 1 is full or not (step S113). If queue 1 is not full (NO in step S113), the process is terminated.
[0028] If queue 1 is full (YES in step S113), the mean and standard deviation of the prediction errors stored in queue 1 are calculated (step S114). Next, it is determined whether or not learning is necessary based on the mean and standard deviation (step S115). The thresholds for the mean and standard deviation used as criteria when evaluating the prediction errors are set in advance, and it is determined that learning is necessary if the mean value exceeds the threshold and the standard deviation value falls below the threshold. If learning is not necessary (NO in step S115), the process is terminated.
[0029] If training is needed (YES in step S115), train the model input / output pairs stored in queue 2 (step S116). Then update and save the weights (step S117). Update the weights of the neural network. Then empty queues 1 and 2 (step S118). Finally, terminate the process.
[0030] The above configuration provides a method for appropriately determining the need for learning and estimating the power consumption of a vehicle undergoing learning.
[0031] (Explanation of the case where the first learning step is not performed in the method for estimating the power consumption of a vehicle according to the embodiment) Figure 2 shows (A) a flowchart and (B) a prediction error in the first case where no learning is performed for the method of estimating the power consumption of a vehicle according to the embodiment. The first case where no learning is performed for the method of estimating the power consumption of a vehicle according to the embodiment will be explained with reference to Figure 2.
[0032] As shown in Figure 2(B), if the average and standard deviation of the power consumption prediction error are small, it is determined that there is no need for learning. As shown in Figure 2(A), the process ends by following the NO route in step 115. This applies when, after vehicle-specific adaptation through learning, the pre-trained model exhibits average driving behavior.
[0033] (Explanation of the learning process in the method for estimating the power consumption of a vehicle according to the embodiment) Figure 3 shows (A) a flowchart and (B) the prediction error when learning the method for estimating the power consumption of a vehicle according to the embodiment. The learning process for the method for estimating the power consumption of a vehicle according to the embodiment will be explained with reference to Figure 3.
[0034] As shown in Figure 3(B), if the average power consumption prediction error is large and the standard deviation is small, it is determined that improvement can be expected through learning. As shown in Figure 3(A), the YES route is taken in step 115. This applies to driving behaviors that tend to increase power consumption, such as ultra-high-speed driving, or when battery degradation occurs over time.
[0035] (Explanation of the case where the second learning step is not performed in the method for estimating the power consumption of a vehicle according to the embodiment) Figure 4 shows (A) a flowchart and (B) a prediction error in the case where the second learning is not performed for the method of estimating the power consumption of a vehicle according to the embodiment. The case where the second learning is not performed for the method of estimating the power consumption of a vehicle according to the embodiment will be explained with reference to Figure 4.
[0036] As shown in Figure 4(B), if the average power consumption prediction error is small and the standard deviation is large, it is determined that there is no need for learning because improvement through learning is unlikely. As shown in Figure 4(A), the process ends by following the NO route in step 115. This applies when factors not considered in the model have an impact, making prediction by the model difficult. For example, if learning is set to occur when the prediction error exceeds a threshold, learning will occur frequently in this case, but in this disclosure, it can be determined that there is no prospect of improvement through learning.
[0037] (Description of the learning method using the driver determination device according to the embodiment) Figure 5A shows an example of learning data for each driver according to the embodiment. The learning method using the driver classifier will be explained with reference to Figure 5A.
[0038] Let's assume that the driver determination system separates the driving for each driver. If two drivers, driver A and driver B, are driving the vehicle, driver A stores the prediction error and input / output pairs in separate queues, such as queue 1A and queue 2A, and driver B stores them in queues 1B and queue 2B. As shown in Figure 5A, even if the standard deviation of the prediction error becomes large due to differences in driving tendencies between drivers, and it is determined that there is no need for learning, separating the queues allows for a more appropriate determination of whether learning is necessary.
[0039] (Description of the learning method using the abnormal data detector according to the embodiment) Figure 5B shows an example of learning by excluding abnormal data according to the embodiment. The learning method using the abnormal data detector according to the embodiment will be explained with reference to Figure 5B.
[0040] Abnormal data may include situations where the vehicle travels a route significantly different from the predicted route, such as accidents causing traffic congestion or prolonged parking during the journey. As shown in Figure 5B, even if the presence of abnormal data increases the standard deviation of the prediction error and it is determined that there is no need for training, excluding the abnormal data from the evaluation allows for a more accurate determination of the need for training.
[0041] (Description of a system for estimating the power consumption of a vehicle according to an embodiment) Figure 6 is a block diagram showing the configuration of a system for estimating the power consumption of a vehicle according to the embodiment. The system for estimating the power consumption of a vehicle according to the embodiment will be described with reference to Figure 6.
[0042] The system 600 for estimating the power consumption of a vehicle according to this embodiment comprises a first storage unit 601, a second storage unit 602, and a learning unit 603.
[0043] The first storage unit 601 stores the prediction error of the vehicle's power consumption. The second storage unit 602 stores input and output pairs of a model for estimating the vehicle's power consumption. The first and second storage units 601 and 602 store data for each driver or each vehicle.
[0044] If the estimation accuracy of the model for estimating the vehicle's power consumption is poor, the learning unit 603 determines whether improvement in estimation accuracy can be expected through learning using the second storage unit 602, using the mean and standard deviation of the prediction error of the vehicle's power consumption in the first storage unit 601. If improvement in estimation accuracy can be expected, the learning unit 603 performs machine learning on the vehicle's power consumption.
[0045] The vehicle power consumption estimation system 600 according to this embodiment is comprised of an information processing device. The information processing device includes a processor that executes and processes a program, and a memory that stores the program. The information processing device may consist of one device or multiple devices. The information processing device may also be a cloud server that processes some or all of its functions in a distributed manner.
[0046] It should be noted that the present invention is not limited to the embodiments described above, and can be modified as appropriate without departing from the spirit of the invention. [Explanation of Symbols]
[0047] 600 System for estimating vehicle power consumption, 601 First storage unit, 602 Second storage unit, 603 Learning unit
Claims
1. A first storage unit for storing prediction errors in the vehicle's power consumption, A second storage unit that stores input and output pairs for a model that estimates the vehicle's power consumption, A system for estimating the power consumption of a vehicle, comprising: a learning unit that, if the estimation accuracy of the model for estimating the power consumption of the vehicle is poor, determines whether the estimation accuracy can be improved by learning using the second storage unit, using the mean and standard deviation of the prediction error of the vehicle's power consumption in the first storage unit, and learns only if the estimation accuracy can be improved.
2. When the mean is small and the standard deviation is small, no improvement in estimation accuracy can be expected. When the mean is large and the standard deviation is small, an improvement in the estimation accuracy can be expected. A system for estimating the power consumption of a vehicle according to claim 1, wherein when the mean is small and the standard deviation is large, it is determined that no improvement in the estimation accuracy can be expected.
3. The system for estimating the power consumption of a vehicle according to claim 1, wherein the first storage unit and the second storage unit store data for each driver or for each vehicle.
4. A system for estimating the power consumption of a vehicle according to claim 1, comprising calculating the mean and standard deviation of the prediction error of the power consumption, excluding abnormal data of the power consumption of the vehicle.
5. A first storage unit for storing prediction errors in the vehicle's power consumption, It comprises a second storage unit that stores input and output pairs for a model that estimates the power consumption of a vehicle, A method for estimating the power consumption of a vehicle, wherein if the estimation accuracy of the model for estimating the power consumption of the vehicle is poor, it is determined whether the estimation accuracy can be improved by learning using the second storage unit, using the mean and standard deviation of the prediction error of the vehicle's power consumption in the first storage unit, and learning is performed only when the estimation accuracy can be improved.