A method and system for predicting the energy of electric vehicles
By dividing the trip progress cutoff ratio in electric vehicles and constructing future sliding window features, the energy consumption rate is predicted using the LightGBM regressor, which solves the problems of real-time performance and accuracy of electric vehicle energy forecasting and realizes dynamic assessment of electric vehicle energy management.
Patent Information
- Application Number
- CN202611236246.8
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2026-08-14
- Publication Date
- 2026-11-17
AI Technical Summary
Existing electric vehicle energy forecasting methods fail to effectively unify window-level instantaneous energy consumption prediction with trip-level remaining energy prediction, making it impossible to achieve real-time, dynamic assessment of remaining battery energy during vehicle travel and thus failing to meet the practical application needs of vehicles during travel.
The journey is divided into observed segments and untraveled segments by using the journey progress cut-off ratio. By constructing the future window features of the future sliding window, the instantaneous energy consumption rate is predicted using the LightGBM regressor, and the remaining energy forecast is obtained by accumulating the data level by level. By combining the weights of different information levels and the calibration results, a closed-loop linkage between rolling prediction and estimation is achieved.
It improves the real-time performance and accuracy of remaining battery energy forecasting during vehicle operation, meets the dynamic power assessment needs during vehicle operation, and enhances the real-time performance and accuracy of electric vehicle energy management.
Smart Images

Figure CN123116207A_ABST
Abstract
Description
Technical Field
[0001] This invention belongs to the field of energy management technology and relates to a method and system for predicting the energy of electric vehicles. Background Technology
[0002] Vehicle energy management systems not only need to achieve online perception of instantaneous energy consumption rate at the second level, but also need to dynamically assess the remaining battery energy during vehicle operation to support range display, route planning, and energy management decisions. Existing research mainly falls into two technical approaches. The first approach establishes a regression model using exogenous features such as kinematic characteristics, road gradient, ambient temperature, and accessory power to obtain a relatively stable estimate of instantaneous energy consumption rate within a segment. The second approach employs a two-stage method using trip-level regression, deep neural networks, or driving features, assuming that future driving information can be fully acquired or backfilled post-trip, thus regressing the remaining energy based on a known complete future trajectory.
[0003] The existing electric vehicle energy forecasting has the following problems: the window-level instantaneous energy consumption prediction task and the trip-level remaining energy prediction task are separated from each other, a unified trip data division rule has not been established, a complete evaluation system for rolling prediction under the setting cut-off ratio during vehicle travel has not been formed, and the perception results of the vehicle's local window energy consumption status cannot be effectively linked and transformed into a dynamic rolling assessment of the remaining battery energy, making it difficult to meet the actual on-board application needs of real-time forecasting of remaining energy during vehicle travel. Summary of the Invention
[0004] In view of the shortcomings of the existing technology, the purpose of this invention is to provide a method and system for predicting the energy of electric vehicles.
[0005] To achieve the above objectives, the present invention adopts the following technical solution: This invention provides a method for predicting the energy consumption of electric vehicles, comprising the following steps: cutting off a certain percentage of the journey progress in the predicted journey. place, with Corresponding window index The journey is divided into observed and untraveled segments, with the observed segments as the boundary. Initial statistical conditions for the cutoff points are obtained based on the observed segments. The fusion weights are determined according to whether journey information is acquired and the I3 level. The calibration results are obtained on the target platform validation set to determine whether calibration has been completed. The calibration results are used to determine the target level from information availability levels I0-I4 according to the deployment grading strategy. Based on the target level and the initial conditions of the observed statistics, the cutoff position is constructed. Then, the future window features of each future sliding window are used; the pre-trained window-level regression mapping is then used, with the future window features as input, to perform instantaneous energy consumption rate rolling prediction for each future sliding window, thus obtaining the instantaneous energy consumption rate rolling prediction value for each future sliding window. The instantaneous energy consumption rate prediction values of each future sliding window are rolled over. and the driving distance inside the window Converted to the corresponding window consumes energy. Cut off the location All subsequent sliding windows will consume energy. The remaining energy after the cut-off point is obtained by rolling and accumulating. The This is the forecast result for the remaining energy of the current journey.
[0006] Furthermore, the future window feature is the B1 hysteresis enhancement feature set; the B1 hysteresis enhancement feature set has a dimension of 44; the B1 hysteresis enhancement feature set includes the exogenous feature set B0 and the instantaneous energy consumption rate hysteresis term within the journey; the exogenous feature set B0 includes kinematic features, road gradient, specific power VSP, ambient temperature, and accessory power features.
[0007] Furthermore, the window-level regression is mapped to a LightGBM regressor f; the hyperparameters of the LightGBM regressor f include: maximum number of leaves per tree, maximum tree depth, learning rate, number of weak learners, and random seed; the maximum number of leaves per tree is 50, the maximum tree depth is 10, the learning rate is 0.05, the number of weak learners is 500, and the random seed is 42.
[0008] Furthermore, the travel progress cutoff ratio The percentage is 30-70%; the stroke progress cutoff ratio It can be any one of 30%, 50%, or 70%; The window index for:
[0009] in, The percentage of time the trip is cut off. This represents the total number of sliding windows within the travel period. This is the floor symbol.
[0010] Furthermore, in the information availability levels I0-I4, I0 is the Trip mean baseline level; in the information availability levels I0-I4, I1 is the Markov total blind level; in the information availability levels I0-I4, I2 is the navigation slope semi blind level; in the information availability levels I0-I4, I3 is the GPS-Trip mean integration level; and in the information availability levels I0-I4, I4 is the full information upper bound level.
[0011] Furthermore, the trip information includes navigation path information or GPS gradient information; The deployment hierarchy strategy is as follows: obtain navigation path information or GPS slope information, and fuse weights at I3 levels. The calibration is performed on the target platform validation set. I3 is the target level, otherwise I0 is the target level.
[0012] Furthermore, when blindly pushing future sliding windows at levels I1-I3, the actual instantaneous energy consumption rate of the last two sliding windows before the cutoff position is used. Perform initialization; then use the predicted values from this window to backfill the instantaneous energy consumption rate lag term; The instantaneous energy consumption rate lag term is:
[0013] in, The sliding window number. This is the lag order.
[0014] Furthermore, the state transition probability of I1 is:
[0015] in, Given the starting speed state, To reach the speed state; Remaining energy after the cutoff point of level I3 for:
[0016] in, To integrate weights, The remaining energy after the cutoff point at the I2 level. The remaining energy after the cutoff point of level I0.
[0017] Furthermore, the cutting position The previous process also included preprocessing the real vehicle data; the preprocessing included: resampling the original non-uniformly spaced timestamps according to a 1-second equal-interval time grid to obtain the battery output energy within the sliding window. and regenerative braking energy ; The battery outputs energy for:
[0018] The braking energy recovery for:
[0019] in, For battery power, This represents the window duration.
[0020] The present invention also provides an electric vehicle energy forecasting system, including a segmentation module: used for cutting off a certain percentage of the journey progress in the predicted journey. place, with Corresponding window index The journey is divided into observed and untraveled segments, and the initial statistical conditions for the cutoff point are obtained based on the observed segments. The determination module is used to determine whether journey information has been acquired and the I3 level fusion weights. The calibration results are obtained on the target platform validation set to determine the target level from information availability levels I0-I4 according to the deployment grading strategy. The construction module is used to construct the cutoff location based on the target level and the observed statistical initial conditions. The following steps are performed on the future window features of each future sliding window; the prediction module is used to perform instantaneous energy consumption rate rolling prediction for each future sliding window by taking the pre-trained window-level regression mapping as input, using the future window features as input, to obtain the instantaneous energy consumption rate rolling prediction value for each future sliding window. ; Conversion module: used to convert the instantaneous energy consumption rate prediction values of each future sliding window into rolling prediction values. and the driving distance inside the window Converted to the corresponding window consumes energy. Accumulation module: used to calculate the cut-off position. All subsequent sliding windows will consume energy. The remaining energy after the cut-off point is obtained by rolling and accumulating. The This is the forecast result for the remaining energy of the current journey.
[0021] Compared with the prior art, the present invention has the following beneficial technical effects: This invention provides an electric vehicle energy forecasting method. It divides the travel into observed and untraveled segments at a predetermined cutoff ratio location and constructs initial statistical conditions for the observed segments. Based on the travel information acquisition and fusion weight calibration results, it selects information to obtain the target level and constructs feature data for subsequent future sliding windows at the cutoff position. It then uses a window-level regression model to predict the instantaneous energy consumption rate of each future window, and obtains the predicted value of the remaining travel energy through window energy consumption conversion and step-by-step accumulation. This method links the window-level instantaneous energy consumption prediction with the travel-level remaining energy estimation in a closed loop. It constructs and blindly extrapolates the corresponding future window features according to the I0-I4 multi-gradient information level matching, closely aligning with the real-time performance and accuracy of dynamic forecasting of remaining battery energy during vehicle operation, as there is no prior information about future operating conditions. Attached Figure Description
[0022] Figure 1 This is a flowchart of an electric vehicle energy forecasting method according to the present invention; Figure 2 This is a diagram illustrating the overall framework of a two-level rolling energy forecasting method for electric vehicle energy forecasting according to the present invention, driven by an information hierarchy. Figure 3 This is a flowchart of the preprocessing process in an embodiment of the present invention; Figure 4 This is a comparison chart of the availability levels of trip-level information in embodiments of the present invention; Figure 5 This is a comparison chart of measured and predicted sample sequences of residual energy I3★ in an embodiment of the present invention (I3★ indicates...). The deployment results of I3 under calibrated conditions), where a is the verification set and b is the test set; Figure 6 This is a graph showing the change of remaining energy over distance / time in an embodiment of the present invention; Figure 7 The figures are a scatter plot of the energy consumption prediction error and a time series fitting curve of the measured energy consumption versus the predicted energy consumption for the window-level B1+LightGBM model in the embodiments of the present invention. a is a scatter plot of the prediction error for the validation set samples, b is a scatter plot of the prediction error for the test set samples, c is a time series fitting curve of the measured and predicted values for the validation set, and d is a time series fitting curve of the measured and predicted values for the test set. Figure 8 This is a comparison diagram of cross-vehicle external measurement and LOVO information hierarchy in an embodiment of the present invention (I3★ indicates...). (I3 deployment results under calibrated conditions) Figure 9 This is a distribution diagram of the absolute error difference between pairing I0 and I3★ in an embodiment of the present invention (I3★ represents...). (I3 deployment results under calibrated conditions) Figure 10 This is a block diagram of the internal structure of a computer device according to an embodiment of the present invention. Detailed Implementation
[0023] To enable those skilled in the art to better understand the present invention, the technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings of the embodiments of the present invention. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort should fall within the scope of protection of the present invention.
[0024] Example 1 This invention provides a method for predicting the energy of electric vehicles, such as... Figure 1 As shown: the percentage of trip progress cutoff in the predicted trip. place, with Corresponding window index The journey is divided into observed and untraveled segments, with the observed segments as the boundary. Initial statistical conditions for the cutoff points are obtained based on the observed segments. The fusion weights are determined according to whether journey information is acquired and the I3 level. Whether calibration is completed on the target platform validation set and obtaining calibration results, including navigation path information or GPS slope information.
[0025] The calibration results determine the target level from information availability levels I0-I4 according to the deployment grading strategy; based on the target level and the initial conditions of the observed statistics, the cutoff location is constructed. Then, the future window features of each future sliding window, and the cut-off position. Preprocess the real vehicle data, such as Figure 3 As shown, the preprocessing includes: resampling the original non-uniformly spaced timestamps according to a 1-second equal-interval time grid to obtain the battery output energy within the sliding window. and regenerative braking energy In this embodiment, a subset of pure electric vehicles in Ann Arbor from the Vehicle Energy Dataset (VED) is used, including three pure electric vehicles: EV455, EV10, and EV541. After Trip segmentation, 20-second sliding window, and EC limiting, EV455 has 279 valid Trips and 16,193 valid window samples. Within each micro-fragment, the time scale is aligned using the piecewise cubic Hermite interpolation (PCHIP) method.
[0026] The deployment tiered strategy is as follows: when navigation or GPS path information is available and weights are fused... After calibration on the target platform validation set, I3 level was selected as the target level, and the remaining energy after the cutoff point at I3 level was output. Otherwise, select level I0 as the target level and output the remaining energy after the cutoff point at level I0. .
[0027] Battery output energy for:
[0028] Braking energy recovery for:
[0029] in, For battery power, This represents the window duration.
[0030]
[0031] in, This refers to the battery terminal voltage. Current, in discharge state Positive, under regenerative braking state It is negative.
[0032] Using the pre-trained window-level regression map and the future window features as input, instantaneous energy consumption rate is predicted sequentially for each future sliding window, resulting in the predicted instantaneous energy consumption rate for each future sliding window. .
[0033] The future window features are the B1 hysteresis-enhanced feature set, which includes the exogenous feature set B0 and the instantaneous energy consumption rate hysteresis term within the journey. The exogenous feature set B0 includes kinematic features, road gradient, specific power VSP, ambient temperature and accessory power features, with a dimension of 42. The exogenous feature set B0 is further enhanced by adding the EC hysteresis term to form the B1 hysteresis-enhanced feature set, with a dimension of 44.
[0034] EC lag term is:
[0035] in, The sliding window number. This is the lag order.
[0036] The window-level regression map f uses LightGBM, which is trained offline once on the EV455 run-length training set. All information levels I0-I4 share the same window-level regressor f. The main hyperparameters of LightGBM are shown in Table 1. Table 1 Key Hyperparameters of LightGBM
[0037] The EV455 test suite's upper window level B1+LightGBM evaluation result is: MAE = 2.89 kWh·(100km) - ¹, RMSE=4.09, R²=0.869, MAPE=14.6%, the scatter plot of the prediction error of the validation set samples is shown below. Figure 7 As shown in Figure a, the scatter plot of the prediction error of the test set samples is as follows. Figure 7 As shown in b, the time-series fitting curves of the measured and predicted values of the validation set are as follows: Figure 7 As shown in c, the time-series fitting curves of the measured and predicted values of the test set are as follows: Figure 7 As shown in d.
[0038] Travel-level forecasting in travel progress cutoff ratio To cut off at a certain point, determine the cut-off window index using the following formula:
[0039] in, The percentage of time the trip is cut off. This represents the total number of sliding windows within the travel period. This is the floor symbol.
[0040] Trip progress cutoff ratio The percentage can be any one of 30%, 50%, or 70%, with 50% being the default evaluation scenario.
[0041] Trips are divided into micro-fragments based on the Best framework with continuous parking exceeding 180 seconds as the segmentation condition. Sliding windows are constructed only within continuous driving segments with a window length of 20 seconds and an overlap rate of 0.7.
[0042] Savitzky-Golay filtering was used to smooth the GPS elevation sequence, and the road slope G was calculated.
[0043] in, and This represents the vehicle altitude at adjacent locations. The horizontal distance between the two points is the semi-sine of the sine.
[0044] Specific power Used to characterize the instantaneous power demand of a vehicle:
[0045] Where v is the vehicle speed (unit: m·s) - ¹), where a is acceleration (unit: m·s²). - ²), , , , These are calibration coefficients, which are 0.156461, 0.002002, 0.000493, and 1.532, respectively.
[0046] The instantaneous energy consumption rate prediction values of each future sliding window are rolled over. and the driving distance inside the window Converted to the corresponding window consumes energy. Cut off the location All subsequent sliding windows will consume energy. The remaining energy after the cut-off point is obtained by rolling and accumulating. The The remaining energy forecast for the current journey is shown in the graph below, where remaining energy changes with distance / time. Figure 6 As shown.
[0047] Instantaneous energy consumption rate rolling forecast
[0048]
[0049] in, To output energy to the battery, The distance traveled by vehicles within the window is given. To reduce abnormal integral interference, a [5, 50] limit is applied to EC, eliminating approximately 24.3% of abnormal windows.
[0050] Rolling forecast of instantaneous energy consumption rate for each future sliding window Converting to the corresponding window consumes energy .
[0051]
[0052] Remaining energy after the cut-off point for:
[0053] like Figure 2 As shown, among the information availability levels I0-I4, I0 is the Trip mean baseline level, I1 is the Markov fully blind level, I2 is the navigation slope semi-blind level, I3 is the GPS-Trip mean integration level, and I4 is the full information upper bound level.
[0054] I0 represents the Trip mean baseline level, and the predicted remaining energy value is:
[0055] in, The average instantaneous energy consumption rate of the observed sliding window before the window is cut off. This represents the remaining driving distance.
[0056] I1 represents the Markov fully blind level. Future kinematic features are generated using a 9-level velocity-state Markov model, and the future features are synthesized using the median template of the training set's velocity states. The state transition probability is calculated using the following formula:
[0057] in, Given the starting speed state, To reach the speed state.
[0058] I2 is a semi-blind navigation slope level, which introduces navigation path and GPS slope information on the basis of I1.
[0059] I3 is the GPS-Trip mean fusion level, which is a weighted fusion of the predicted values of I2 and I0:
[0060] in, To integrate weights, The remaining energy after the cutoff point at the I2 level. In this embodiment, the remaining energy after the cutoff point of level I0 is... , At that time, the MAE of the test set changed by approximately 0.011 kWh.
[0061] I4 is the upper bound of the full information level, which uses real future vehicle state information for prediction and is only used as a theoretical upper bound.
[0062] When constructing future window features for levels I1-I3, the instantaneous energy consumption rate lag term is backfilled according to the following rules to obtain the true instantaneous energy consumption rate of the last two sliding windows before the cutoff position. Perform initialization; then use the predicted values from this window to backfill the instantaneous energy consumption rate lag term; The instantaneous energy consumption rate lag term is:
[0063] in, The sliding window number. This is the lag order.
[0064] The information level comparison results on the EV455 test set (57 trips, effective trip count N=53 under 50% cutoff) are shown in Table 2 and [other tables]. Figure 4 As shown.
[0065] Table 2 Comparison of Information Availability Levels at 50% Cutoff Points
[0066] Table 2 shows that MAE decreases in a stepwise manner from I1 to I2 to I3 (0.304 → 0.179 → 0.096 kWh); I2 alone (0.179 kWh) is still higher than I0 (0.148 kWh), indicating that the semi-blind noise can be amplified when navigation gradient is superimposed on Markov velocity error; I3 passes through =0.40 After fusion repair of I2, MAE decreased to 0.096 kWh, R²=0.990, as Figure 5 a and such Figure 5 As shown in b. For paired tests of N=53 Trips, Δ= 0.052 kWh, Wilcoxon one-sided p=0.003, I3 is significantly better than I0 in the domain, such as Figure 9 As shown in a.
[0067] exist At the three cutoff points of 30%, 50%, and 70%, the MAE of the I3 level test set were 0.103 kWh, 0.096 kWh, and 0.094 kWh, respectively, and the R² were 0.991, 0.990, and 0.974, respectively, indicating that the prediction accuracy of the I3 level further improves as the proportion of the observed journey increases.
[0068] The cross-vehicle external test results for the EV10 (192 trips, 8536 effective sliding windows) are shown in Table 3 and [other tables]. Figure 8 As shown.
[0069] Table 3 Comparison of trip-level cross-vehicle information hierarchy (50% cut-off)
[0070] The I1→I2→I3 stepwise acceleration still holds true for the EV10 under cross-vehicle external testing (0.356→0.214→0.105 kWh); however, there is no significant difference between I3 and I0 on the EV10 (Δ= 0.007 kWh, 95% CI [0.022, 0.009], p = 0.299), as Figure 9 As shown in b, the advantage of I3 over I0 is vehicle-dependent. The window-level cross-vehicle external MAE is 2.89 kWh / (100 km). - ¹Increased to 3.51 kWh·(100 km) - ¹, R² decreased from 0.869 to 0.833, but it can still support stroke-level cascading rolling.
[0071] Under the Leave-One-Vehicle Cross-Validation (LOVO) of three BEVs in VED, the three-fold equally weighted average I3 level MAE is 0.129 ± 0.022 kWh. The information level ladder can be reproduced in multiple folds, but the advantage of I3 over I0 is vehicle-dependent, and it still needs to be calibrated on the target platform validation set during deployment. Then enable I3.
[0072] Mean absolute error (MAE), root mean square error (RMSE), coefficient of determination (R²), and mean absolute percentage error (MAPE) are used as evaluation indicators and are calculated according to the following formulas:
[0073]
[0074]
[0075]
[0076] in, These are measured values. For predicted values, is the measured mean, and n is the sample size.
[0077] Example 2 The present invention provides an electric vehicle energy forecasting system, comprising a division module, a determination module, a construction module, a prediction module, a conversion module, and an accumulation module.
[0078] Segmentation module: Used to cut off the percentage of travel progress in the predicted travel route. place, with Corresponding window index The journey is divided into observed and untraveled segments, and the initial statistical conditions for the cutoff point are obtained based on the observed segments. The determination module is used to determine whether journey information has been acquired and the I3 level fusion weights. The calibration results are obtained on the target platform validation set to determine the target level from information availability levels I0-I4 according to the deployment grading strategy. The construction module is used to construct the cutoff location based on the target level and the observed statistical initial conditions. The following steps are performed on the future window features of each future sliding window; the prediction module is used to perform instantaneous energy consumption rate rolling prediction for each future sliding window by taking the pre-trained window-level regression mapping as input, using the future window features as input, to obtain the instantaneous energy consumption rate rolling prediction value for each future sliding window. ; Conversion module: used to convert the instantaneous energy consumption rate prediction values of each future sliding window into rolling prediction values. and the driving distance inside the window Converted to the corresponding window consumes energy. Accumulation module: used to calculate the cut-off position. All subsequent sliding windows will consume energy. The remaining energy after the cut-off point is obtained by rolling and accumulating. The This is the forecast result for the remaining energy of the current journey.
[0079] The electric vehicle energy forecasting system provided by this invention can implement the same method steps as the above method, so it will not be described again.
[0080] Example 3 Computer equipment can specifically be a terminal or a server. For example... Figure 10As shown, the computer device includes a processor, memory, network interface, display, and input devices connected via a system bus. The processor provides computing and control capabilities. The memory includes non-volatile storage media and internal memory. The non-volatile storage media stores the operating system and computer programs. The internal memory provides an environment for the operation of the operating system and computer programs in the non-volatile storage media. The network interface is used to communicate with external terminals via a network connection. When the computer program is executed by the processor, it implements the aforementioned electric vehicle energy prediction method. The display screen can be a liquid crystal display (LCD) or an e-ink display. The input devices can be a touch layer covering the display screen, buttons, a trackball, or a touchpad mounted on the computer device casing, or an external keyboard, touchpad, or mouse.
[0081] As will be understood by those skilled in the art, computer equipment Figure 10 The structure shown is merely a block diagram of a portion of the structure related to the present application and does not constitute a limitation on the computer device to which the present application is applied. Specific computing devices may include more or fewer components than those shown in the figure, or combine certain components, or have different component arrangements.
[0082] Example 4 A computer device includes a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to implement the steps of the above-described method.
[0083] In one embodiment, a computer-readable storage medium is provided having a computer program stored thereon, which, when executed by a processor, implements the steps of the above-described method.
[0084] The various embodiments in this specification are described in a progressive manner. For directly identical or similar parts among the embodiments, refer to each other. Each embodiment focuses on its differences from other embodiments. In particular, the device embodiments are basically similar to the method embodiments, so the description is relatively simple; refer to the description of the method embodiments for relevant details. It should be noted that the technical features of the above embodiments can be combined arbitrarily. For the sake of brevity, not all possible combinations of the technical features in the above embodiments are described. However, as long as the combination of these technical features does not contradict each other, it should be considered within the scope of this specification.
[0085] The above-described embodiments are merely preferred embodiments of the present invention, and while the descriptions are specific and detailed, they should not be construed as limiting the scope of the invention. It should be noted that those skilled in the art can make various improvements and substitutions without departing from the principles of the present invention, and these improvements and substitutions should also be considered within the scope of protection of the present invention. Therefore, the scope of protection of this invention should be determined by the scope of the claims.
[0086] It should be noted that the terms "first," "second," etc., in the specification, claims, and accompanying drawings of this invention are used to distinguish similar objects and are not necessarily used to describe a specific order or sequence. It should be understood that such data can be interchanged where appropriate so that the embodiments of the invention described herein can be implemented in orders other than those illustrated or described herein. Furthermore, the terms "comprising" and "having," and any variations thereof, are intended to cover a non-exclusive inclusion; for example, a process, method, system, product, or apparatus that comprises a series of steps or units is not necessarily limited to those steps or units explicitly listed, but may include other steps or units not explicitly listed or inherent to such processes, methods, products, or apparatus.
Claims
1. A method for predicting the energy of electric vehicles, characterized in that, Includes the following steps: The proportion of trip progress cutoff in the predicted trip place, with Corresponding window index The journey is divided into observed segments and untraveled segments, and the initial statistical conditions for the cut-off point are obtained based on the observed segments. Based on whether travel information was obtained and I3 level fusion weight Whether calibration is completed on the target platform verification set and obtaining calibration results, the calibration results determine the target level from information availability levels I0-I4 according to the deployment grading strategy; Based on the target level and the initial conditions of the observed statistics, the cut-off location is constructed. The future window features of each future sliding window thereafter; Using the pre-trained window-level regression map and the future window features as input, instantaneous energy consumption rate is predicted sequentially for each future sliding window, resulting in the predicted instantaneous energy consumption rate for each future sliding window. ; The instantaneous energy consumption rate prediction values of each future sliding window are rolled over. and the driving distance inside the window Converted to the corresponding window consumes energy. ; Cut off the location All subsequent sliding windows will consume energy. The remaining energy after the cut-off point is obtained by rolling and accumulating. , will the This is a forecast of the remaining energy for the current journey.
2. The electric vehicle energy forecasting method according to claim 1, characterized in that: The future window feature is the B1 hysteresis enhancement feature set; The B1 hysteresis enhancement feature set has a dimension of 44. The B1 hysteresis enhancement feature set includes the exogenous feature set B0 and the instantaneous energy consumption rate hysteresis term within the stroke; The exogenous feature set B0 includes kinematic features, road gradient, specific power VSP, ambient temperature, and accessory power features.
3. The electric vehicle energy forecasting method according to claim 1, characterized in that: The window-level regression mapping is a LightGBM regressor f; The hyperparameters of the LightGBM regressor f include: maximum number of leaves per tree, maximum tree depth, learning rate, number of weak learners, and random seed; The maximum number of leaves in a single tree is 50, the maximum tree depth is 10, the learning rate is 0.05, the number of weak learners is 500, and the random seed is 42.
4. The electric vehicle energy forecasting method according to claim 1, characterized in that: The percentage of trip progress cutoff It ranges from 30% to 70%; The percentage of trip progress cutoff It can be any one of 30%, 50%, or 70%; The window index for: in, The percentage of time the trip is cut off. This represents the total number of sliding windows within the travel period. This is the floor symbol.
5. The electric vehicle energy forecasting method according to claim 1, characterized in that: In the information availability levels I0-I4, I0 is the Trip mean baseline level; In the information availability levels I0-I4, I1 is the Markov total blindness level; Among the information availability levels I0-I4, I2 is the navigation slope semi-blindness level; Among the information availability levels I0-I4, I3 is the GPS-Trip mean integration level; Among the information availability levels I0-I4, I4 is the upper limit level for all information.
6. The electric vehicle energy forecasting method according to claim 5, characterized in that: The trip information includes navigation route information or GPS gradient information; The deployment tiered strategy is as follows: Obtain navigation path information or GPS slope information, and simultaneously fuse weights at the I3 level. If calibration is completed on the target platform validation set, then I3 is the target level; otherwise, I0 is the target level.
7. The electric vehicle energy forecasting method according to claim 6, characterized in that: When blindly pushing future sliding windows at I1-I3 levels, the actual instantaneous energy consumption rate of the last two sliding windows before the cutoff position is used. Perform initialization; then use the predicted values from this window to backfill the instantaneous energy consumption rate lag term; The instantaneous energy consumption rate lag term is: in, The sliding window number. This is the lag order.
8. The electric vehicle energy forecasting method according to claim 5, characterized in that: The state transition probability of I1 is: in, Given the starting speed state, To reach the speed state; Remaining energy after the cutoff point of level I3 for: in, To integrate weights, The remaining energy after the cutoff point at the I2 level. The remaining energy after the cutoff point of level I0.
9. The electric vehicle energy forecasting method according to claim 1, characterized in that: The cutting position Previously, this also included preprocessing of real vehicle data; The preprocessing includes: The original non-uniformly spaced timestamps are resampled according to a 1-second equal-interval time grid to obtain the battery output energy within the sliding window. and regenerative braking energy ; The battery outputs energy for: The braking energy recovery for: in, For battery power, This represents the window duration.
10. An electric vehicle energy forecasting system, characterized in that, include: Segmentation module: Used to cut off the percentage of travel progress in the predicted travel route. place, with Corresponding window index The journey is divided into observed segments and untraveled segments, and the initial statistical conditions for the cut-off point are obtained based on the observed segments. Determining Module: Used to determine whether trip information has been obtained and the I3 level fusion weight. Whether calibration is completed on the target platform verification set and obtaining calibration results, the calibration results determine the target level from information availability levels I0-I4 according to the deployment grading strategy; Construction module: Used to construct cut-off locations based on target level and observed statistical initial conditions. The future window features of each future sliding window thereafter; Prediction module: This module takes the pre-trained window-level regression map, uses the future window features as input, and performs rolling predictions of the instantaneous energy consumption rate for each future sliding window, obtaining the rolling predicted value of the instantaneous energy consumption rate for each future sliding window. ; Conversion module: used to convert the instantaneous energy consumption rate prediction values of each future sliding window into a rolling prediction value. and the driving distance inside the window Converted to the corresponding window consumes energy. ; Accumulation module: used to calculate the cut-off position All subsequent sliding windows will consume energy. The remaining energy after the cut-off point is obtained by rolling and accumulating. The This is the forecast result for the remaining energy of the current journey.
