Vehicle energy flow pre-visualization method and device
By integrating multi-source data and using operating condition prediction models, the system enables hierarchical visualization and early warning of energy flow in commercial vehicles. This solves the problems of low prediction accuracy and poor early warning effect in existing technologies, thereby improving drivers' energy management capabilities and vehicle safety.
Patent Information
- Application Number
- CN202611232512.X
- 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 vehicle energy flow display technology cannot predict energy changes in subsequent operating conditions, has low prediction accuracy, cannot meet the needs of long-distance and heavy-load operation of commercial vehicles, and drivers cannot intuitively distinguish between real-time status and predicted status, resulting in poor early warning effect.
By acquiring multi-source vehicle data and combining it with vehicle driving data, navigation traffic conditions, road features, load information, etc., a working condition prediction model is used to make short-cycle working condition predictions. The current energy flow and the predicted energy flow are then displayed in a layered and visualized manner on the vehicle terminal, and an energy flow anomaly warning is executed.
Drivers can adjust their driving behavior in advance, reduce vehicle energy consumption, improve the safety of the three-electric system, adapt to complex operating scenarios of commercial vehicles, and achieve accurate prediction of operating conditions and early warning of energy risks.
Smart Images

Figure CN123104666A_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the field of vehicle technology, and in particular to a method and apparatus for pre-visualizing vehicle energy flow. Background Technology
[0002] Currently, new energy commercial vehicles are generally equipped with energy flow visualization functions, which can display the energy transmission direction and real-time power between the engine, drive motor, power battery and transmission system, helping the driver understand the energy usage status of the whole vehicle.
[0003] However, existing technologies have at least the following drawbacks: 1. It can only display the energy flow at the current moment after the fact, and cannot predict the energy changes corresponding to subsequent working conditions. The driver can only observe passively and it is difficult to take driving optimization actions in advance. 2. Commercial vehicles operate under complex conditions, including long downhill slopes, continuous uphill climbs, alternation between national highways and expressways, traffic light congestion, and heavy-load traction. Under different operating conditions, energy recovery, power output, and power generation states switch frequently, and sudden changes in energy flow can easily cause problems such as battery overcharging, delayed power response, and increased energy consumption. 3. Some solutions rely solely on the vehicle's own sensors to make simple operational condition judgments without combining multi-dimensional data such as navigation road conditions, road slope, traffic flow ahead, and load. As a result, the prediction accuracy is low and cannot meet the needs of long-distance and heavy-load commercial vehicle operations. 4. The existing visualization interface does not have a dedicated display format to distinguish between "current energy flow" and "predicted energy flow", making it impossible for drivers to intuitively distinguish between real-time status and predicted status, resulting in poor warning effect. Summary of the Invention
[0004] The purpose of this invention is to provide a vehicle energy flow pre-visualization method and device, which can at least alleviate the problems of existing energy flow displays that only show in real time, lack predictive ability, have low prediction accuracy, and poor early warning effect. It can combine multi-source vehicle data such as vehicle driving data, navigation road conditions, road features, and load information to achieve short-cycle operating condition prediction. It can also visualize the current energy flow and predicted energy flow in layers on the vehicle terminal. At the same time, it can provide energy risk warnings for extreme operating conditions, guide drivers to optimize driving behavior, reduce vehicle energy consumption, and improve the operational safety of the three-electric system.
[0005] To address the aforementioned technical problems, in a first aspect, the present invention provides a method for pre-visualizing vehicle energy flow, comprising at least: S1. Acquire multi-source vehicle data and input the multi-source vehicle data into the operating condition prediction model to obtain the predicted operating conditions of the vehicles. S2. Determine the energy flow direction, energy range, and operating status between each vehicle component based at least on the predicted vehicle operating conditions, the multi-source vehicle data, and the vehicle energy management strategy, so as to generate a predicted energy flow dataset. S3. At least the predicted energy flow dataset is used as the visualization data source for the vehicle energy flow interface, and the current energy flow and predicted energy flow are rendered in layers, and an energy flow anomaly warning is executed.
[0006] Optionally, the multi-source vehicle data includes at least real-time vehicle operation data, basic vehicle status data, navigation and road data, and historical behavior data; The real-time vehicle operating data includes at least one of the following: vehicle speed, accelerator pedal opening, brake pedal opening, clutch status, gearbox gear, drive motor speed, drive motor torque, drive motor power, engine speed, engine output power, power battery voltage, power battery current, power battery SOC, and braking system status. The vehicle basic status data includes at least one of the following: total vehicle load, traction load, and vehicle operating mode; The navigation and road data includes at least one of the following: road type, road slope, curves, traffic lights, congestion status, distance to any obstacle in the direction of travel, distance to any downhill section in the direction of travel, and distance to any uphill section in the direction of travel within a preset time period. The historical behavior data includes at least the driver's vehicle operation data within a preset time span and / or historical energy flow operation data for road sections under the same working conditions.
[0007] Optionally, the operating condition prediction model includes at least a multi-source data preprocessing layer, a physical demand power calculation layer, an operating condition classification decision tree layer, and a time series prediction output layer; The multi-source data preprocessing layer is used at least to collect the multi-source vehicle data from each vehicle interface and at least to perform time synchronization, outlier removal and dimension normalization to obtain preprocessed data. The physical demand power calculation layer is at least used to calculate the current demand power item by item based on the preprocessed data and the vehicle longitudinal dynamics model. The working condition classification decision tree layer is used at least to determine the vehicle working condition type by taking the slope feature of the current power demand, the pedal intention feature and the navigation road condition label as input, and then outputting a discrete working condition label sequence. The time-series prediction output layer is used at least to perform a sliding window weighted average on the discrete operating condition label sequence to generate continuous prediction outputs for multiple prediction time points, thereby obtaining the predicted operating conditions of the vehicle.
[0008] Optionally, the predicted energy flow dataset is generated through at least a four-level extrapolation chain; The four-level simulation link includes the operating condition to energy mode link, the energy mode to power allocation link, the power allocation to flow direction mapping link, and the data integration and output link. The input of the operating condition to energy mode link includes at least the time series prediction structure output by the time series prediction output layer in the operating condition prediction model, and the output of the operating condition to energy mode link includes at least the energy mode enumeration value. The input of the energy mode to power allocation link includes at least the timing prediction structure and the energy mode enumeration value, and the output of the energy mode to power allocation link includes at least the power allocation structure. The input of the power allocation to flow mapping link includes at least the power allocation structure, and the output of the power allocation to flow mapping link includes at least a list of energy flow tuples. The inputs of the data integration output link include at least the energy flow tuple list, the discrete operating condition label sequence output by the operating condition classification decision tree layer in the operating condition prediction model, and the current demand power calculated by the physical demand power calculation layer in the operating condition prediction model. The output of the data integration output link includes at least the predicted energy flow dataset.
[0009] Optionally, the predicted vehicle operating conditions include at least constant speed driving conditions, acceleration driving conditions, deceleration coasting conditions, long downhill braking conditions, continuous uphill conditions, idling parking conditions, and congestion start-stop conditions.
[0010] Optionally, on the same road segment, the prediction deviation between the predicted operating conditions of the vehicle and the multi-source vehicle data is calculated in stages. When the prediction deviation meets the preset deviation conditions, an online correction process is triggered.
[0011] Optionally, the online correction process includes at least: The real-time corrected power mapping coefficient is calculated iteratively based on at least a preset smoothing coefficient, an initial corrected power mapping coefficient, the multi-source vehicle data, and the predicted operating conditions of the vehicle, until the prediction deviation no longer meets the preset deviation condition.
[0012] Optionally, at least in response to the vehicle returning to the factory, the predicted and measured comparison data of any trip of the vehicle are obtained, and then at least the least squares method is used to solve the combination of physical model parameters that minimizes the preset objective function with the measured power sequence as the objective. The candidate thresholds of each node of the decision tree are traversed, and the threshold combination that maximizes the classification accuracy is re-selected based on the measured working condition label. The updated parameters and thresholds are packaged into an OTA firmware upgrade package so that the vehicle loads the new parameters or threshold combination at least when it is powered on again, thus completing the offline correction closed loop.
[0013] Optionally, the prediction accuracy of each working condition label of the working condition prediction model is statistically analyzed. If there is an abnormal working condition type whose prediction accuracy is consistently lower than a preset accuracy threshold, the decision tree structure is adjusted. The decision tree structure adjustment method includes at least the following: adjusting the integer threshold of the decision node corresponding to the abnormal working condition type; adding new decision branch features; and adjusting the weighted decay coefficient of the sliding window.
[0014] Based on the same concept, in a second aspect, the present invention also provides a vehicle energy flow pre-visualization device for performing the vehicle energy flow pre-visualization method described in any one of the first aspects; The vehicle energy flow pre-visualization device includes at least: The operating condition prediction module is used to acquire multi-source vehicle data and input the multi-source vehicle data into the operating condition prediction model to obtain the predicted operating conditions of the vehicle. The dataset generation module is used to determine the energy flow direction, energy range and operating status between various vehicle components based at least on the vehicle's predicted operating conditions, the multi-source vehicle data and the vehicle energy management strategy, so as to generate a predicted energy flow dataset. The rendering warning module is used to at least use the predicted energy flow dataset as a visualization data source for the vehicle energy flow interface, to render the current energy flow and the predicted energy flow in layers, and to execute an energy flow anomaly warning.
[0015] The technical solution provided by this invention first acquires multi-source vehicle data and inputs it into a condition prediction model to obtain the predicted operating conditions of the vehicle. Further, based on the predicted operating conditions, multi-source vehicle data, and vehicle energy management strategy, the energy flow direction, energy range, and operating status between various vehicle components are determined to generate a predicted energy flow dataset. Finally, the predicted energy flow dataset is used as a visualization data source for the vehicle energy flow interface to render the current energy flow and the predicted energy flow in layers, and an energy flow anomaly warning is executed. Therefore, this invention can at least alleviate the problems of existing energy flow displays, such as only real-time display, lack of predictive capability, low prediction accuracy, and poor warning effect. It can combine multi-source vehicle data, such as vehicle driving data, navigation road conditions, road features, and load information, to achieve short-cycle condition prediction. It can also provide layered visualization of the current energy flow and the predicted energy flow on the vehicle terminal. Simultaneously, it can provide energy risk warnings for extreme operating conditions, guiding drivers to optimize driving behavior, reducing vehicle energy consumption, and improving the operational safety of the three-electric system (battery, motor, and electronic control system).
[0016] The present invention can achieve at least the following beneficial effects: 1. Driving optimization: Drivers can know in advance about subsequent energy changes and adjust gear, throttle, and coasting strategies accordingly, effectively reducing overall vehicle fuel / electricity consumption and improving operational economy; 2. Enhanced safety: Anticipate operating conditions such as high energy recovery and high power output to avoid risks of failure such as battery overcharging, motor overload, and transmission system impact; 3. Strong adaptability: The prediction logic is optimized for typical driving scenarios of commercial vehicles such as heavy trucks, light trucks, and special vehicles, and is compatible with multiple power architectures such as hybrid, pure electric, and fuel cell. 4. User-friendly human-computer interaction: The real-time stream and the predicted stream are clearly distinguishable visually, the warnings are intuitive, and they do not interfere with normal driving observation; 5. A comprehensive solution for predictive energy flow visualization for commercial vehicles is proposed, which constructs a complete technical system including multi-source data acquisition, operating condition prediction, energy flow extrapolation, hierarchical visualization and intelligent early warning, and solves the technical defects of traditional energy flow that only displays in real time and does not have forward-looking prediction. 6. A multi-source data fusion acquisition mechanism for commercial vehicles was designed to unify and integrate vehicle operating status, vehicle load, road gradient, navigation road conditions and driver operation data, so as to realize multi-dimensional input to support operating condition prediction and adapt to complex operating scenarios of commercial vehicles. 7. A lightweight working condition prediction model adapted to low computing power in vehicles was constructed. It was specifically adapted for typical working conditions of commercial vehicles such as heavy-load climbing, long downhill recovery, idling power generation, and congestion start-stop, and can achieve accurate working condition prediction in short cycle. 8. A mechanism for linking operating conditions and energy flow is proposed. Based on the predicted future driving conditions, the mechanism automatically maps and predicts the future energy flow direction, power range and operating status of the power source, transmission system and energy storage components, realizing a closed-loop transformation from "operating condition prediction" to "energy state prediction". 9. An energy risk warning and driving prompt mechanism for typical operating conditions of commercial vehicles has been established, which can provide early warning for scenarios such as high-power energy recovery, heavy-load high-power output, and battery overcharge risk, and match the safety and energy-saving operation needs of commercial vehicles. Attached Figure Description
[0017] Figure 1 This is a flowchart of a vehicle energy flow pre-visualization method provided in an embodiment of the present invention; Figure 2 This is a schematic diagram of the structure of a vehicle energy flow pre-visualization device provided in an embodiment of the present invention. Detailed Implementation
[0018] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to the accompanying drawings. Obviously, the described embodiments are only a part of the embodiments of this invention, and not all of them. Based on the embodiments of this invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this invention.
[0019] It should be noted that any symbols and / or numbers present in the specification that are not marked in the accompanying drawings are not reference numerals.
[0020] Figure 1 This is a flowchart of a vehicle energy flow pre-visualization method provided by an embodiment of the present invention. This embodiment is applicable to vehicle energy flow pre-visualization scenarios of on-board terminals in at least any vehicle (e.g., hybrid, pure electric, or fuel cell commercial vehicles). The vehicle energy flow pre-visualization method can be, but is not limited to, executed by the vehicle energy flow pre-visualization device in this embodiment of the present invention as the execution subject, which can be implemented in software and / or hardware. Figure 1 As shown, this vehicle energy flow pre-visualization method includes at least the following steps: S1. Acquire multi-source vehicle data and input the multi-source vehicle data into the operating condition prediction model to obtain the predicted operating conditions of the vehicles.
[0021] Multi-source vehicle data can be composed of vehicle data from multiple sources, such as various on-board sensors, CAN bus, and vehicle networking.
[0022] In one specific implementation, the multi-source vehicle data may optionally include at least real-time vehicle operation data, basic vehicle status data, navigation and road data, and historical behavior data. Real-time vehicle operating data includes at least one of the following: vehicle speed, accelerator pedal opening, brake pedal opening, clutch status, transmission gear, drive motor speed, drive motor torque, drive motor power, engine speed, engine output power, power battery voltage, power battery current, power battery SOC, and braking system status. The vehicle's basic status data must include at least one of the following: total vehicle load, traction load, and vehicle operating mode (e.g., pure electric, hybrid, power generation, regenerative braking, etc.). The navigation and road data shall include at least one of the following: road type (e.g., highway, national highway, mountain road, etc.), road gradient, curves, traffic lights, congestion status, distance to any obstacle in the direction of travel, distance to any downhill section in the direction of travel, and distance to any uphill section in the direction of travel within a preset time period (e.g., 30 seconds); Historical behavior data includes at least the driver's vehicle operation data and / or historical energy flow operation data of the same road segment within a preset time span (e.g., 2 hours, 1 day, etc.).
[0023] In actual vehicle use, the vehicle terminal can collect real-time vehicle operation data (i.e., the aforementioned real-time vehicle operation data), external road condition data (i.e., the aforementioned navigation and road data), and basic vehicle status data, and transmit them uniformly to the central control display unit and the vehicle controller (VCU).
[0024] In another specific implementation, the operating condition prediction model includes at least a multi-source data preprocessing layer, a physical demand power calculation layer, an operating condition classification decision tree layer, and a time series prediction output layer. The multi-source data preprocessing layer is used at least to collect multi-source vehicle data from each vehicle interface and at least to complete time synchronization, outlier removal and dimension normalization to obtain preprocessed data. The physical demand power calculation layer is used at least to calculate the current demand power item by item based on the preprocessed data and the vehicle longitudinal dynamics model; The working condition classification decision tree layer is used at least to determine the vehicle working condition type by taking the slope feature of the current power demand, pedal intention feature and navigation road condition label as input, and then outputting a discrete working condition label sequence. The time-series prediction output layer is used at least to perform a sliding window weighted average on the discrete operating condition label sequence to generate continuous prediction outputs for multiple prediction time points, thereby obtaining the vehicle's predicted operating conditions.
[0025] For vehicle-mounted lightweight operating condition prediction, multi-source vehicle data can be input into a pre-defined lightweight operating condition prediction model (i.e., the aforementioned operating condition prediction model) to predict the vehicle's continuous operating conditions from 3 to 30 seconds. The operating condition prediction model adopts a hybrid architecture of "rule-based decision tree + physical model + lightweight time-series prediction," organized into four hierarchical layers. Data flows unidirectionally between layers (the output of the upper layer becomes the input of the lower layer). The interface and data interaction with the vehicle's physical system can be as follows: 1. First layer – Multi-source data preprocessing layer: responsible for collecting raw signals from various physical interfaces on the vehicle and completing time synchronization, outlier removal (e.g., 3σ criterion filtering) and dimensional normalization.
[0026] The physical interfaces of each data source are as follows: ① Vehicle speed v, accelerator pedal opening α acc Brake pedal opening α brake Drive motor speed n motor Engine speed n eng —Received via the vehicle's CAN bus, the message ID corresponds to the PGN defined by the vehicle manufacturer, and the refresh cycle is 100ms; ② Road type ahead (expressway / national highway / mountain road / urban road), road gradient θ naviCurve curvature, traffic light distance, and speed limit information are obtained through the RoutePlan interface of the navigation SDK. The data transmission cycle is 1 second, and the road segment ahead of 30 seconds is discretized into 30 sampling points with a step size of 1 second. ③ Total vehicle mass m — obtained from the real-time calibration value of the load sensor, with a calibration accuracy of ±2% and a refresh cycle of 1 second; ④ Current road slope θ imu —The pitch angle signal from the vehicle-mounted IMU is obtained after low-pass filtering and is complemented and fused with the navigation slope to improve accuracy.
[0027] 2. Second Layer – Physical Demand Power Calculation Layer: Based on the vehicle's longitudinal dynamics model, calculates the current demand power item by item. P req Each variable in the formula has a one-to-one correspondence with a specific vehicle-mounted sensor signal: P req =(mgsinθ+mgfcosθ+0.5ρCdAv 2 +δma)v; In the above formula, the physical sources of each variable are: m—total vehicle mass (kg), derived from the load cell calibration value (first-layer interface ③), with an accuracy of ±2%; g—gravitational acceleration constant 9.81 m / s². 2 θ—Current road slope angle (rad), obtained by weighted fusion of navigation map slope data and IMU pitch angle, with the fusion weight dynamically adjusted based on GPS signal quality, sourced from Layer 1 interfaces ②+④; v—Current vehicle speed (m / s), derived from wheel speed sensor CAN messages, processed by the ABS / ESC module, sourced from Layer 1 interface ①; f—Rolling resistance coefficient, typically 0.007~0.015, pre-stored in VCU ROM as an inherent vehicle parameter; ρ—Air density (kg / m³). 3 The standard value is taken as 1.225 kg / m³. 3 , pre-stored in VCU ROM; Cd—drag coefficient, typically 0.5~0.8, pre-stored in VCU ROM from vehicle factory parameters; A—frontal area (m²) 2 A typical value can be taken as 7.0m. 2 ~10.0m 2 The parameters are pre-stored in the VCUROM by the vehicle's factory settings; δ—rotational mass conversion factor, typically 1.04~1.08, determined by looking up a table based on the current gear position of the transmission; a—acceleration (m / s²). 2 The value is obtained by calculating the first-order difference dv / dt of the vehicle speed v, and then smoothed by a 5-point moving average. The source is the v signal.
[0028] Substituting the slope sequence θ(t) from the first layer output for the 30 seconds ahead into the above formula time by time, the power demand curves for each future time can be generated. P req (t). It is understandable that the formula calculations all use fixed-point arithmetic to adapt to the on-board fixed-point MCU and avoid dependence on floating-point arithmetic units.
[0029] 3. Third layer – Working condition classification decision tree layer: based on P req (t) The slope characteristics of the curve, pedal intention characteristics, and navigation road condition labels are used as inputs. The working condition type at each time is determined by the preset rule decision tree (i.e. the aforementioned preset rule decision tree).
[0030] For example, the decision tree depth is ≤5 levels, and all decision nodes are compared using integer thresholds to avoid floating-point operations; the decision tree contains a total of 8 decision nodes, and the complete pseudocode structure can be as follows: DECISION_TREE_PredictDrivingCycle(Input: Δ P req / Δ t , α acc , α brake , v,θ navi RoadType): / / Node 1: Check if the vehicle is in parking mode (Input: v from wheel speed sensor CAN message; threshold 5 km / h is the parking / driving boundary) IF v < 5 km / h THEN / / Node 2: Idle Parking vs. Traffic Start-Stop (Input: Historical frequency of v) IF Number of times v crosses the 5km / h threshold in the past 10 seconds ≥ 3 THEN RETURN Traffic Start-Stop ELSE RETURN Idle Parking END IF / / Node 3: Check if it is a long downhill slope (Input: θ) navi From the navigation SDK; Threshold -3% is the downhill detection threshold) IF θ navi <-3% AND duration ≥ 5s THENRETURN Long downhill braking END IF / / Node 4: Check if it is a continuous uphill climb (Input: θ) navi (Threshold 3% is the climbing threshold) IF θ navi >3% AND Duration ≥ 10s THEN RETURN Continuous Climb END IF / / Node 5: Check Acceleration Intent - Rapid Acceleration Detection / / (Input: Δ Preq / Δ t The slope α from the second-layer power curve acc From CAN message; / / Threshold 50kW / s for emergency acceleration power change rate, 70% corresponds to large throttle opening) IF Δ P req / Δ t >50 kW / s AND α acc >70% THEN RETURN (Accelerate) / / Node 6: Slow-Acceleration Decision ELSE IF Δ P req / Δ t >0 AND α acc >0 THENRETURN Accelerate END IF / / Node 7: Check deceleration intention (Input: α) brake From CAN message, dP req / dt; Threshold 0 distinguishes between acceleration / deceleration) IF α brake >0 OR dP req / dt<-20 kW / s THENRETURN Deceleration coasting END IF / / Node 8: Default constant speed driving (Input: RoadType from navigation SDK, v from CAN message) RETURN constant speed driving.
[0031] The details of the input signal sources and threshold physical meanings for the above 8 decision nodes are as follows: ①Node 1: v<5km / h distinguishes between parking and driving states, and 5km / h is the minimum distinguishable vehicle speed for commercial vehicles in parking / driving states; ②Node 2: Within 10 seconds, v crosses 5km / h ≥ 3 times, distinguishing between congested start-stop and idling parking. 3 crossings correspond to the average start-stop frequency of typical urban congested road sections; ③ Node 3: θ navi A slope of less than -3% and lasting for ≥5 seconds is considered a long downhill slope. The -3% slope corresponds to the slope threshold at which commercial vehicles need to brake continuously, and 5 seconds is the lower limit of the time required to exclude short slopes. ④ Node 4: θ navi A gradient greater than 3% and lasting for ≥10 seconds is considered continuous climbing. 3% is the gradient threshold for switching power modes in commercial vehicles, and 10 seconds is the minimum time limit for excluding short climbs. ⑤ Node 5: Δ P req / Δ t >50kW / s and α acc>70% indicates rapid acceleration; 50kW / s corresponds to the typical power ramp-up rate of the combined output of the engine / motor in a commercial vehicle; 70% pedal opening corresponds to the driver's clear acceleration intention. ⑥ Node 6: Δ P req / Δ t >0 and α acc >0 indicates a gradual acceleration, covering mild acceleration scenarios; ⑦ Node 7: α brake >0 or Δ P req / Δ t <-20kW / s is used to determine deceleration. -20kW / s is the dividing line between the power change of natural deceleration and active braking after a commercial vehicle releases the accelerator. ⑧ Node 8: If any of the above conditions are not met, it is determined to be a constant speed driving.
[0032] Furthermore, the initial values of the thresholds for each node in the decision tree can be determined through the following offline parameter calibration process: ① Collect typical road spectrum data of vehicles on actual operating routes of ≥1000km, and simultaneously record the full signal of CAN bus, GPS trajectory and navigation traffic information; ② The calibration engineer manually labels the operating conditions at each moment based on the vehicle speed curve, power curve, and pedal opening curve; ③ For each decision node, traverse the candidate thresholds within its physical reasonable range, and select the threshold combination that maximizes the classification accuracy based on manually labeled tags. ④ Burn the calibrated threshold parameter table to the non-volatile memory area of the VCU ROM.
[0033] 4. Fourth layer - Time series prediction output layer: Input the discrete working condition label sequence output from the third layer into the sliding window weighted average unit to generate continuous prediction outputs for multiple future time points (i.e., the aforementioned multiple prediction time points).
[0034] For example, the sliding window size N=10, and the weight allocation strategy W=[0.05, 0.05, 0.08, 0.08, 0.10, 0.12, 0.14, 0.16, 0.18, 0.04], where the sampling point closer to the current time has a larger weight.
[0035] It is known that the physical basis for the weight design is that commercial vehicles have large inertia and their operating conditions change continuously, and recent observations are significantly more indicative of future conditions than long-term observations.
[0036] Specifically, the sliding window moves forward one step every 100ms, performs a weighted vote on the N=10 work condition labels within the window, and selects the work condition type with the highest weighted score as the current prediction output. Simultaneously, based on the second layer's generation... P req (t) Based on the curve and navigation road segment data, the system can output predicted data for the following five time points: 3s, 5s, 10s, 20s, and 30s in the future. Each time point output structure (i.e., the aforementioned vehicle predicted operating condition) must include at least: ① Operating condition type; ② Predicted power demand. P pred (kW, float); ③ Predict the target vehicle speed v target (km / h, float). Five time points form a progressive prediction: high accuracy in the short term serves energy flow rendering, in the medium term it serves driving suggestion delivery, and in the long term it serves energy management strategy pre-adjustment.
[0037] 5. Computing power pruning and optimization measures correspond to specific physical hardware indicators: ① The slope-sine / cosine value pre-stored ROM lookup table method can replace real-time trigonometric function calculation. The lookup table resolution is 0.5°, and the ROM space occupies about 360×2×2Byte=1.44KB, which meets the constraint of ROM space ≤2KB. ② Decision tree reasoning: a maximum of 8 integer comparisons, each comparison takes about 3 CPU cycles, and the time taken at a 200MHz main frequency is ≤8×3 / 200MHz=0.12μs, which is much less than the 1ms upper limit; ③ Physical demand power calculation layer: The fixed-point number Q16.16 format can be used. A single calculation involves about 15 multiplications and 8 additions. At a main frequency of 200MHz, the time taken is ≤(15+8) / 200MHz≈0.12μs. After multiplying by 30 future sampling points, the total time is ≤3.6μs, which is much less than the 3ms upper limit. ④ Sliding window weighted average: 30 multiplications and 1 maximum value lookup, time ≤ 1ms; ⑤ The total time for a single inference run across the above four layers is ≤5ms, which meets the real-time requirement of a 100ms refresh cycle for the VCU main loop; ⑥ Total ROM usage ≤ 2KB, RAM usage ≤ 1KB.
[0038] S2. Determine the energy flow direction, energy range, and operating status between each vehicle component based at least on the vehicle's predicted operating conditions, multi-source vehicle data, and the vehicle's energy management strategy, in order to generate a predicted energy flow dataset.
[0039] Among them, the vehicle energy management strategy is updated iteratively according to the actual vehicle use, and its acquisition method may be, but is not limited to, obtaining it through VCU.
[0040] In yet another specific implementation, the predicted energy flow dataset may optionally be generated through at least a four-level extrapolation chain; The four-level simulation link includes the operating condition to energy mode link, the energy mode to power allocation link, the power allocation to flow direction mapping link, and the data integration and output link. The input of the load condition to energy mode link includes at least the time series prediction structure output by the time series prediction output layer in the load condition prediction model, and the output of the load condition to energy mode link includes at least the energy mode enumeration value. The input to the energy mode to power allocation link includes at least a timing prediction structure and energy mode enumeration values, and the output of the energy mode to power allocation link includes at least a power allocation structure. The input to the power allocation to flow mapping link includes at least a power allocation structure, and the output of the power allocation to flow mapping link includes at least a list of energy flow tuples. The inputs to the data integration output link include at least a list of energy flow tuples, a sequence of discrete operating condition labels output by the operating condition classification decision tree layer in the operating condition prediction model, and the current demand power calculated by the physical demand power calculation layer in the operating condition prediction model. The outputs of the data integration output link include at least a predicted energy flow dataset.
[0041] In another specific implementation, optionally, the vehicle's predicted operating conditions include at least constant speed driving conditions, acceleration driving conditions, deceleration coasting conditions, long downhill braking conditions, continuous uphill conditions, idling parking conditions, and congestion start-stop conditions.
[0042] More specifically, based on the future operating condition prediction output obtained in the previous step, combined with the vehicle energy management strategy and multi-source vehicle data, the energy flow direction, power range, and operating status between various components can be deduced to generate a predicted energy flow dataset. The energy flow deduction can adopt a four-level deduction chain: "Operating Condition → Energy Mode → Power Allocation → Flow Direction Mapping." The input / output data format for each level is clearly defined as follows: Level 1 (Operating Condition → Energy Mode): The input is the PredictionOutput structure, which is the output of the time series prediction output layer from the previous step. Its data format is {timestamp t (Unix time, uint32_t), operating condition type CycleType, predicted power demand}. P pre Predict target vehicle speed v target (float, unit: km / h)}
[0043] Based on the operating condition type and the current vehicle status (such as SOC, vehicle speed, etc.), the energy mode enumeration value EnergyMode(uint8_t) is output through the preset operating condition-energy mode mapping rules: 0=pure electric drive, 1=engine direct drive, 2=combined drive, 3=regenerative braking, 4=idle power generation, 5=coasting without regeneration.
[0044] For example, the complete rules of a vehicle energy management strategy can be shown as the following structured conditions: Vehicle Energy Management Strategy Rule Table: |Conditions|Energy Mode|Energy Flow Direction|Priority|; | SOC≥30% AND P req ≤ P motor_max AND v<50km / h AND θ≤3% | Pure electric drive | Power battery → DC / DC → Drive motor → Wheel | 1 |; | SOC<30% OR P req > P motor_max OR v≥50km / h OR θ>3% | Engine direct drive / combined drive | Engine (+Power battery) → Drive motor → Wheel | 2 |; | α brake >0 AND v>5km / h | Regenerative Braking | Wheel → Drive Motor → DC / DC → Power Battery | 3 |; | v≈0km / h AND SOC<30% | Idle power generation | Engine → Generator → Power battery | 4 |; | α brake =0 AND α acc =0 AND v>5km / h AND θ≥0% | Gliding without recovery | No energy flow |5 |; Note: SOC ≥ 30% indicates the lower limit of pure electric drive by the battery; P motor_max This indicates the peak power of the drive motor; v≥50km / h indicates the vehicle speed at which the commercial vehicle switches between pure electric and hybrid modes; θ=3% gradient corresponds to the gradient threshold at which the commercial vehicle needs to switch to combined drive mode.
[0045] Second stage (Energy Mode → Power Allocation): The input is the EnergyMode enumeration value of the first stage output and the PredictionOutput structure, and the output is the PowerAllocation structure.
[0046] The structure of the PowerAllocation structure can be: {engine output power} P eng (float, kW), engine output torque T eng (float, Nm), output power of the drive motor P motor (float, kW), the output torque of the drive motor T motor (float, Nm), power battery charging and discharging power P bat (float, positive value is discharge kW), DC / DC conversion power P DCDC (float, kW), auxiliary load power P aux (float, kW)}.
[0047] The power of each component is limited by the following physical constraints: engine output P eng Limited by external characteristic curve P eng_max ( n eng The torque-speed MAP, transmitted in real-time by the engine ECU via the CAN bus, is determined by looking up a table; the drive motor output... P motor Limited by the peak torque-speed curve P motor_max ( n motor The power battery charging and discharging power is limited by the three-dimensional joint constraint table of SOC-temperature-SOP, which is transmitted in real time by the motor controller MCU via the CAN bus. P bat_max (SOC, T bat The data is transmitted in real time by the Battery Management System (BMS) via the CAN bus.
[0048] Level 3 (Power Allocation → Flow Mapping): The input is the PowerAllocation structure output from Level 2, and the output is a list of energy flow tuples called FlowTupleList. Each tuple can be [source component ID, target component ID, power value]. P flow (float, kW), Energy Type (EnergyType).
[0049] Furthermore, the flow path can be stored using a directed graph adjacency list. The set of component nodes V = {engine (ID=0), generator (ID=1), drive motor (ID=2), gearbox (ID=3), power battery (ID=4), DC / DC converter (ID=5), wheel (ID=6)}, and each directed edge E(i→j,P) represents the power P (kW) transferred from component i to component j.
[0050] Level 4 (Data Integration Output): The list of energy flow tuples, power ranges, durations, etc. at each time point are integrated into a PredictiveEnergyDataset, which serves as the data source for the next step of visualization rendering.
[0051] For example, the dataset contains at least the following four pieces of information: ① Flow direction: Outputted from the adjacency list of the third-level directed graph, and represented by a list of FlowPath structures {from_id(uint8_t), to_id(uint8_t), Pflow(float)}; ② Power range: The power demand under the predicted operating conditions is taken as the range. P req The P10 to P90 quantiles are used as the upper and lower bounds of the interval [Plow, Phigh]. The quantiles can be obtained by looking up a table based on historical actual power statistics for the same operating condition. If there are fewer than 50 samples of historical data for this operating condition, the interval is ±15% of the calculated power demand. ③ Duration Δt: Directly take the start and end time difference of the predicted working condition Δt = t end t start The time series of operating condition labels output from the aforementioned steps is used to determine the condition, with an accuracy of 100ms. ④ Working status: Record the start / stop / run status of each component at the corresponding time, represented by a uint8_t bitmask.
[0052] Based on the above four-level deduction chain, the detailed energy flow deduction examples for seven typical commercial vehicle operating conditions are as follows: (1) Constant speed driving condition: If SOC ≥ 30% and P req ≤ P motor_max If v < 50km / h, it is determined to be in pure electric drive mode, and the energy flow is "power battery → DC / DC → drive motor → wheels". P bat_discharge = P req / (η motor ×ηDCDC ), where η motor The typical efficiency value for the drive motor can be taken as 0.88~0.94, η DCDC This represents a typical DC / DC conversion efficiency value, which can be taken as 0.93~0.97. If SOC < 30% or P req > P motor_max If v ≥ 50km / h, it is determined to be in engine direct drive mode, and the energy flow direction is "engine → transmission → wheels". P eng = P req / η trans , where η trans This is a typical value for the efficiency of the transmission system, which can be taken as 0.85~0.92.
[0053] (2) Acceleration driving condition: If Δ P req / Δ t >50kW / s and α acc >70%, indicating a combined engine and electric motor drive mode, with energy flow following the sequence: "Engine + Power Battery → Drive Motor → Wheels". Engine output. P eng =min[ P req , P eng_max (n eng )],in P eng_max ( n eng () represents the current rotational speed n eng The upper limit of the engine's external characteristic power; the supplementary power of the electric motor. P motor =min[ P req - P eng , P motor_max ( n motor )],in, P motor_max ( n motor () represents the peak power of the motor at the current speed; battery discharge power. P bat = P motor / (η motor ×η DCDC If Δ P req / Δt For speeds ≤50kW / s, pure electric drive is prioritized, with the engine supplementing any shortfall in motor power.
[0054] (3) Deceleration coasting condition: If α brake If the value is >0, it is determined to be in regenerative braking mode, with energy flow following the sequence: "wheels → drive motor → DC / DC → power battery". Braking power requirement. P brake_req The brake pedal opening α brake Obtained via linear mapping and table lookup. Motor recoverable power. P regen =min( P brake_req ×η trans ×η motor_regen , P motor_max_regen ), where η trans For the transmission efficiency, approximately 0.92, η motor_regen The generator efficiency is approximately 0.88 to 0.92. If α... brake =0, no energy recovery, energy flow is in the natural gliding state.
[0055] (4) Braking on a long downhill slope: Based on physics, the component of gravity along the slope surface, Fg = mgsinθ, is the main force driving the vehicle to accelerate when going downhill. To maintain a steady downhill speed v steady Continuous braking force needs to be applied to balance Fg, corresponding to braking power. P brake =Fg×v-mgfcosθv. The drive motor enters continuous energy recovery mode, and the motor recovers power. P regen = min( P brake ×η trans ×η motor_regen , P motor_max_regen ), where η trans With a transmission efficiency of approximately 0.92, η motor_regen The generator's power generation efficiency is approximately 0.88~0.92. Battery charging power... P charge = P regen ×η DCDC - P aux η DCDC Approximately 0.93~0.97, P aux To reduce the power consumption of auxiliary loads.
[0056] At the same time, the maximum charging power is limited by referring to a table based on the battery SOC and temperature. Pcharge_max (SOC, T bat Final battery charging power P bat_charge =min[ P charge , P charge_max (SOC, T bat To prevent battery overcharging, the energy flow is as follows: "wheels → drive motor → DC / DC → power battery". If the battery SOC is close to full charge (SOC≥90%) and the downhill continues, the braking power exceeding the recovery capacity will be dissipated as heat by the air pressure braking system. In this case, the energy flow interface should be marked as "recovery braking limited".
[0057] (5) Continuous climbing condition: Power required for climbing P climb =(mgsinθ+mgfcosθ+0.5ρCdAv 2 )×v / η trans Engine output P eng =min[ P climb , P eng_max ( n eng )], drive motor supplement P motor =min[ P climb - P eng , P motor_max ( n motor Battery discharge power P bat = P motor / (η motor ×η DCDC The combined output power range is bounded by the sum of the rated power of the engine and the motor as the lower bound and the sum of their peak power as the upper bound: [ P eng_rated + P motor_rated , P eng_peak + P motor_peak The energy flow is "engine + power battery → drive motor → wheels". When the SOC drops below 30%, the engine, in addition to driving the wheels, also needs to charge the battery through the generator. P eng_total =P climb + P gen_charge The total engine output should not exceed P eng_peak .
[0058] (6) Idle parking condition: If SOC < 30%, it is determined to be in engine idle power generation mode, and the energy flow is "engine → generator → power battery". Power generation P gen The generator output power is determined by referring to the table for the engine's economic idle range, and a typical output power of 10kW to 20kW can be taken. If the state of charge (SOC) is ≥30%, the engine stops generating electricity, and all components are in standby mode.
[0059] (7) Traffic jam start-stop operation: In low-speed, frequent start-stop scenarios, to avoid increased fuel consumption and emissions caused by frequent engine start-stop, a pure electric drive mode is forcibly adopted, with the energy flow being "power battery → DC / DC → drive motor → wheels". When the SOC drops below 30%, it automatically switches to engine drive mode and starts generating electricity to supplement the load.
[0060] S3. At least use the predicted energy flow dataset as the visualization data source for the vehicle energy flow interface, render the current energy flow and the predicted energy flow in layers, and perform an energy flow anomaly warning.
[0061] The current energy flow and the predicted energy flow can be rendered using different methods.
[0062] Specifically, for layered differentiated visualization rendering, the current energy flow and the predicted energy flow can be overlaid and rendered in layers on the energy flow interface of the vehicle's instrument panel and central control screen, and clearly distinguished by visual features. The rendering rules can be as follows: 1. Current actual energy flow Line type: solid line; Dynamic effect: standard flow light effect; Color scheme: Using the existing system, mechanical transmission energy is orange-yellow, and electrical energy transmission is blue; Display content: Real-time accurate power value and current operating status of components.
[0063] 2. Future energy flow prediction Line type: dashed / dotted line; Dynamic effect: low flow rate, semi-transparent light effect (transparency set to 40%~60%). Color scheme: Same as the current energy flow color scheme, without adding any additional color interference; Display content: Marks the predicted power range and the expected duration of effect.
[0064] 3. Layout rules: The predicted energy flow is superimposed on the current energy flow path for display without changing the original layout of the vehicle components, ensuring interface consistency.
[0065] Furthermore, for condition-based early warnings and driving suggestion pushes, warnings can be issued based on predicted high-risk / high-energy-consumption conditions and the predicted energy flow status. 1. Predicting strong energy recovery on long downhill slopes (predicted recovery power greater than the set threshold): The predicted energy flow path is highlighted and flashes, and a text warning pops up on the interface: "Long downhill slope ahead, high-power energy recovery is about to begin, please pay attention to the battery SOC status"; if the battery SOC is close to full charge, an additional warning is added: "Battery charge is high, it is recommended to reduce braking recovery in advance". 2. Predicting continuous heavy-load climbing (combined output power exceeds threshold): Energy flow is highlighted, and a suggestion is pushed: "Continuous climbing ahead, combined power output increases, it is recommended to maintain a stable gear"; 3. Forecast of prolonged idling power generation: Tip: "Prolonged idling is imminent, which will increase power generation energy consumption. You can shut down the engine in time to save energy." 4. The warning pop-up can adopt a weak reminder style, floating in the corner of the interface without obscuring the core energy flow screen.
[0066] Furthermore, the dynamic refresh and closed-loop correction process of the aforementioned operating condition prediction models will be explained below.
[0067] In another specific implementation, optionally, under the same road segment, the prediction deviation between the predicted vehicle operating conditions and the multi-source vehicle data is calculated in stages, and when the prediction deviation meets the preset deviation conditions, the online correction process is triggered.
[0068] In yet another specific implementation, the online correction process may optionally include at least: The real-time correction power mapping coefficient is calculated iteratively based on at least the preset smoothing coefficient, the initial corrected power mapping coefficient, multi-source vehicle data, and the predicted vehicle operating conditions, until the prediction deviation no longer meets the preset deviation conditions.
[0069] In another specific implementation, optionally, at least in response to the vehicle returning to the factory, the predicted and measured comparison data of any trip of the vehicle are obtained, and then at least the least squares method is used to solve for the combination of physical model parameters that minimizes the preset objective function with the measured power sequence as the objective, and the candidate thresholds of each node of the decision tree are traversed, and the threshold combination that maximizes the classification accuracy is re-selected based on the measured working condition label, and the updated parameters and thresholds are packaged into an OTA firmware upgrade package, so that the vehicle loads the new parameters or threshold combination at least when it is powered on again, thus completing the offline correction closed loop.
[0070] In another specific implementation, optionally, if there is an abnormal working condition type whose prediction accuracy is consistently lower than a preset accuracy threshold, the prediction accuracy of each working condition label in the statistical working condition prediction model is adjusted to trigger the decision tree structure. The decision tree structure adjustment methods include at least: adjusting the integer threshold of the decision node corresponding to the abnormal working condition type; adding new decision branch features; and adjusting the weighted decay coefficient of the sliding window.
[0071] More specifically, the data, operating condition prediction results, and energy flow display can be refreshed every 100ms. Simultaneously, the prediction results are compared with subsequent actual measured data to adaptively adjust model parameters, gradually improving prediction accuracy and achieving closed-loop optimization. The iterative correction mechanism can adopt a division of labor architecture of "online lightweight correction + offline batch recalibration," and the specific method is as follows: (1) Online deviation calculation: The predicted (time t operating condition type CycleType_pred, power value) is calculated. P pred Compare the data with the actual data collected at time t, align them in time, and calculate the prediction deviation ΔP = ( P pred - P actual ) / P actual ×100%.
[0072] Understandably, the cumulative deviation determination must be based on the same road segment ID: each road segment in the RoutePlan data output by the vehicle navigation SDK has a unique identifier, SegmentID. Predicted values are compared with actual values only under the same SegmentID to avoid false triggers caused by sudden changes in operating conditions when entering different road segments. The online correction process is triggered only when the prediction deviation |ΔP| exceeds 15% for five consecutive times under the same SegmentID (i.e., the aforementioned prediction deviation meets the preset deviation condition). This mechanism binds the correction trigger to the road's physical attributes, ensuring that the correction only targets systemic model deviations rather than environmental changes caused by road segment alterations.
[0073] (2) Correction of online power mapping coefficient EWMA: The power mapping coefficient k can be corrected by the exponential weighted moving average method. k new =α k old +(1-α)×( P actual / P pred_base ), where α=0.9 is the smoothing coefficient (i.e. the aforementioned preset smoothing coefficient). k old These are the mapping coefficients from the previous period. P pred_baseThis is the baseline value for the predicted power before correction. The initial value of the k coefficient (i.e., the aforementioned initial corrected power mapping coefficient) can be set to 1.0, gradually converging to a stable value reflecting the actual efficiency of the vehicle's power system and the model bias through multiple iterations. Online correction only adjusts a single scalar coefficient k, without modifying the decision tree threshold or physical model parameters. The computational load is only one multiplication and one addition, taking ≤0.1ms on a 200MHz TC3xx VCU. It can be executed in real time after each power prediction output without increasing the load on the main loop. The corrected k coefficient immediately takes effect in the next round of power prediction: P_pred_corrected= k new × P pred_base .
[0074] (3) Offline batch parameter recalibration: After the vehicle returns to the factory, the VCU can upload all the prediction-measurement comparison data of the current trip to the cloud big data platform via T-Box. The uploaded data includes, but is not limited to: timestamp sequence, SegmentID, prediction operating condition type, and prediction power. P pred Measured power P actual Current value of k coefficient, decision tree path record.
[0075] Furthermore, batch recalibration can be performed on driving data with a cumulative range of ≥500km via a cloud platform: ① The least squares method is used, based on the measured power sequence. P actual Let (t) be the objective, and solve for ∑[ P pred (t) P actual (t)] 2 Minimize the combination of physical model parameters—rolling resistance coefficient f, transmission efficiency η trans Motor efficiency MAPη motor DC / DC efficiency η DCDC ; ② Traverse the candidate thresholds of each node in the decision tree, and re-select the threshold combination that maximizes the classification accuracy based on the actual working condition labels; ③ Package the updated parameters and thresholds into an OTA firmware upgrade package (e.g., differential binary format), send it to the VCU via T-Box, and after verification, write it to the calibration parameter area of non-volatile memory, overwriting the old values. The new parameters will be automatically loaded the next time the vehicle is powered on, completing the offline correction closed loop.
[0076] (4) Quantitative Expectations and Verification Standards for Correction Effects: Based on simulation and test data from typical commercial vehicle operation scenarios, the above-mentioned "online EWMA + offline least squares recalibration" correction mechanism achieves the following quantitative effects: ① After approximately 500km of adaptive correction, the power prediction accuracy converged from ±20% in the initial state to ±8% (using the RMS value of ΔP as the evaluation index, RMS=sqrt[∑(ΔP_i)). 2 / n)]; ② The accuracy of working condition classification has increased from approximately 82% initially to approximately 93%; ③ The time taken for a single online EWMA correction is ≤0.1ms, and the CPU utilization rate of the VCU main loop is ≤0.1%; ④ Offline recalibration is performed on the cloud server. The processing time for a single 500km data run is ≤3 minutes, and the OTA packet size is ≤4KB.
[0077] (5) Adjustment of the working condition classification model structure: The prediction accuracy of each working condition label is statistically analyzed. For working condition types with an accuracy rate consistently below 85% (i.e., the aforementioned preset accuracy threshold can be taken as 85%), the decision tree structure is adjusted during the offline analysis stage in the cloud. The adjustment methods include: adjusting the integer threshold of the decision node corresponding to the working condition; adding new decision branch features; and adjusting the weighted decay coefficient λ of the sliding window. All structural adjustments can be cross-validated in the cloud to ensure that the accuracy is statistically significantly improved after the adjustment.
[0078] (6) Correction data persistence and cross-trip accumulation: The k coefficient after each online correction and the physical model parameters and decision tree threshold after each OTA update are written to the correction parameter area of the vehicle non-volatile memory. It supports power failure memory and cross-trip accumulation optimization. After power failure and restart, the latest correction parameters are automatically loaded from ROM without re-performing cold start calibration.
[0079] The technical solution provided in this embodiment first acquires multi-source vehicle data and inputs it into a working condition prediction model to obtain the predicted operating conditions of the vehicle. Further, based on the predicted operating conditions, multi-source vehicle data, and the vehicle energy management strategy, the energy flow direction, energy range, and operating status between various vehicle components are determined to generate a predicted energy flow dataset. Finally, the predicted energy flow dataset is used as a visualization data source for the vehicle energy flow interface to render the current energy flow and the predicted energy flow in layers, and an energy flow anomaly warning is executed. Therefore, this embodiment can at least alleviate the problems of existing energy flow displays, such as only real-time display, lack of predictive capability, low prediction accuracy, and poor warning effect. It can combine multi-source vehicle data, such as vehicle driving data, navigation road conditions, road features, and load information, to achieve short-cycle working condition prediction. It can also provide layered visualization of the current energy flow and the predicted energy flow on the vehicle terminal. Simultaneously, it can provide energy risk warnings for extreme working conditions, guiding drivers to optimize driving behavior, reduce vehicle energy consumption, and improve the operational safety of the three-electric system.
[0080] It should be noted that this invention constructs a lightweight on-board prediction model by integrating multi-dimensional information such as on-board sensor data, navigation traffic data, road slope data, vehicle load, and driver operating habits. This model predicts the vehicle's operating conditions for the next 3 to 30 seconds. Based on the predicted operating conditions, it infers the energy flow direction, power range, and operating status of the corresponding power source, transmission system, and energy storage components, generating predicted energy flow data. A layered visual rendering strategy is adopted to distinguish the current actual energy flow from the future predicted energy flow using different line types, transparency, and dynamic effects, enabling overlay display on the same interface. For typical commercial vehicle operating conditions such as long downhill slopes with strong energy recovery, continuous uphill climbs with high power output, and high-load power generation at idle, the energy flow interface is highlighted with warnings, and driving optimization suggestions are pushed. The model is adapted to the computing power limitations of commercial vehicles, uses a lightweight algorithm, and can run directly on the on-board VCU and central control unit without the need for an external high-performance computing unit.
[0081] Figure 2 This is a schematic diagram of a vehicle energy flow pre-visualization device provided in an embodiment of the present invention. This embodiment is applicable to at least any vehicle energy flow pre-visualization scenario of an in-vehicle terminal in a vehicle. The vehicle energy flow pre-visualization device can be implemented in software and / or hardware. Figure 2 As shown, the vehicle energy flow pre-visualization device includes at least: The operating condition prediction module 110 is used to acquire multi-source vehicle data and input the multi-source vehicle data into the operating condition prediction model to obtain the predicted operating conditions of the vehicle. The dataset generation module 120 is used to determine the energy flow direction, energy range and operating status between various vehicle components based at least on the vehicle's predicted operating conditions, multi-source vehicle data and the vehicle's energy management strategy, so as to generate a predicted energy flow dataset. The rendering warning module 130 is used to use at least the predicted energy flow dataset as the visualization data source for the vehicle energy flow interface, to render the current energy flow and the predicted energy flow in layers, and to execute an energy flow anomaly warning.
[0082] Optionally, multi-source vehicle data includes at least real-time vehicle operation data, basic vehicle status data, navigation and road data, and historical behavior data; Real-time vehicle operating data includes at least one of the following: vehicle speed, accelerator pedal opening, brake pedal opening, clutch status, transmission gear, drive motor speed, drive motor torque, drive motor power, engine speed, engine output power, power battery voltage, power battery current, power battery SOC, and braking system status. The basic vehicle status data includes at least one of the following: total vehicle load, traction load, and vehicle operating mode; The navigation and road data shall include at least one of the following for a preset time period in the direction in which the vehicle is traveling: road type, road gradient, curves, traffic lights, congestion status, distance to any obstacle in the direction in which the vehicle is traveling, distance to any downhill section in the direction in which the vehicle is traveling, and distance to any uphill section in the direction in which the vehicle is traveling. Historical behavior data includes at least the driver's vehicle operation data within a preset time span and / or historical energy flow operation data for road sections under the same working conditions.
[0083] Optionally, the operating condition prediction model includes at least a multi-source data preprocessing layer, a physical demand power calculation layer, an operating condition classification decision tree layer, and a time series prediction output layer; The multi-source data preprocessing layer is used at least to collect multi-source vehicle data from each vehicle interface and at least to complete time synchronization, outlier removal and dimension normalization to obtain preprocessed data. The physical demand power calculation layer is used at least to calculate the current demand power item by item based on the preprocessed data and the vehicle longitudinal dynamics model; The working condition classification decision tree layer is used at least to determine the vehicle working condition type by taking the slope feature of the current power demand, pedal intention feature and navigation road condition label as input, and then outputting a discrete working condition label sequence. The time-series prediction output layer is used at least to perform a sliding window weighted average on the discrete operating condition label sequence to generate continuous prediction outputs for multiple prediction time points, thereby obtaining the vehicle's predicted operating conditions.
[0084] Optionally, the predicted energy flow dataset is generated through at least a four-level extrapolation chain; The four-level simulation link includes the operating condition to energy mode link, the energy mode to power allocation link, the power allocation to flow direction mapping link, and the data integration and output link. The input of the load condition to energy mode link includes at least the time series prediction structure output by the time series prediction output layer in the load condition prediction model, and the output of the load condition to energy mode link includes at least the energy mode enumeration value. The input to the energy mode to power allocation link includes at least a timing prediction structure and energy mode enumeration values, and the output of the energy mode to power allocation link includes at least a power allocation structure. The input to the power allocation to flow mapping link includes at least a power allocation structure, and the output of the power allocation to flow mapping link includes at least a list of energy flow tuples. The inputs to the data integration output link include at least a list of energy flow tuples, a sequence of discrete operating condition labels output by the operating condition classification decision tree layer in the operating condition prediction model, and the current demand power calculated by the physical demand power calculation layer in the operating condition prediction model. The outputs of the data integration output link include at least a predicted energy flow dataset.
[0085] Optionally, the predicted operating conditions of the vehicle include at least constant speed driving conditions, acceleration driving conditions, deceleration coasting conditions, long downhill braking conditions, continuous uphill conditions, idling parking conditions, and traffic jam start-stop conditions.
[0086] Optionally, on the same road segment, the prediction deviation between the predicted vehicle operating conditions and the multi-source vehicle data is calculated in stages. When the prediction deviation meets the preset deviation conditions, the online correction process is triggered.
[0087] Optionally, the online correction process includes at least: The real-time correction power mapping coefficient is calculated iteratively based on at least the preset smoothing coefficient, the initial corrected power mapping coefficient, multi-source vehicle data, and the predicted vehicle operating conditions, until the prediction deviation no longer meets the preset deviation conditions.
[0088] Optionally, at least in response to the vehicle returning to the factory, the predicted and measured comparison data of any trip of the vehicle are obtained, and then at least the least squares method is used to solve the combination of physical model parameters that minimizes the preset objective function with the measured power sequence as the objective. The candidate thresholds of each node of the decision tree are traversed, and the threshold combination that maximizes the classification accuracy is re-selected based on the measured working condition label. The updated parameters and thresholds are packaged into an OTA firmware upgrade package so that the vehicle loads the new parameters or threshold combination at least when it is powered on again, thus completing the offline correction closed loop.
[0089] Optionally, if there are abnormal working conditions whose prediction accuracy is consistently lower than a preset accuracy threshold, the prediction accuracy of each working condition label in the statistical working condition prediction model will trigger an adjustment of the decision tree structure. The decision tree structure adjustment methods include at least: adjusting the integer threshold of the decision node corresponding to the abnormal working condition type; adding new decision branch features; and adjusting the weighted decay coefficient of the sliding window.
[0090] The technical solution provided in this embodiment first acquires multi-source vehicle data through a condition prediction module and inputs the multi-source vehicle data into a condition prediction model to obtain the predicted operating conditions of the vehicle. Further, a dataset generation module determines the energy flow direction, energy range, and operating status between various vehicle components based on the predicted operating conditions, multi-source vehicle data, and the vehicle energy management strategy to generate a predicted energy flow dataset. Finally, a rendering and warning module uses the predicted energy flow dataset as a visualization data source for the vehicle energy flow interface to render the current energy flow and the predicted energy flow in layers, and executes an energy flow anomaly warnings. Therefore, this embodiment can at least alleviate the problems of existing energy flow displays, such as only real-time display, lack of predictive capability, low prediction accuracy, and poor warning effect. It can combine multi-source vehicle data such as vehicle driving data, navigation road conditions, road features, and load information to achieve short-cycle condition prediction and can provide layered visualization of the current energy flow and predicted energy flow on the vehicle terminal. Simultaneously, it can provide energy risk warnings for extreme operating conditions, guiding drivers to optimize driving behavior, reduce vehicle energy consumption, and improve the operational safety of the three-electric system.
[0091] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, and not to limit them; although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some or all of the technical features; and these modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the scope of the technical solutions of the embodiments of the present invention.
Claims
1. A method for pre-visualizing vehicle energy flow, characterized in that, At least including: S1. Acquire multi-source vehicle data and input the multi-source vehicle data into the operating condition prediction model to obtain the predicted operating conditions of the vehicles. S2. Determine the energy flow direction, energy range, and operating status between each vehicle component based at least on the predicted vehicle operating conditions, the multi-source vehicle data, and the vehicle energy management strategy, so as to generate a predicted energy flow dataset. S3. At least the predicted energy flow dataset is used as the visualization data source for the vehicle energy flow interface, and the current energy flow and predicted energy flow are rendered in layers, and an energy flow anomaly warning is executed.
2. The vehicle energy flow pre-visualization method according to claim 1, characterized in that, The multi-source vehicle data includes at least real-time vehicle operation data, basic vehicle status data, navigation and road data, and historical behavior data; The real-time vehicle operating data includes at least one of the following: vehicle speed, accelerator pedal opening, brake pedal opening, clutch status, gearbox gear, drive motor speed, drive motor torque, drive motor power, engine speed, engine output power, power battery voltage, power battery current, power battery SOC, and braking system status. The vehicle basic status data includes at least one of the following: total vehicle load, traction load, and vehicle operating mode; The navigation and road data includes at least one of the following: road type, road slope, curves, traffic lights, congestion status, distance to any obstacle in the direction of travel, distance to any downhill section in the direction of travel, and distance to any uphill section in the direction of travel within a preset time period. The historical behavior data includes at least the driver's vehicle operation data within a preset time span and / or historical energy flow operation data for road sections under the same working conditions.
3. The vehicle energy flow pre-visualization method according to claim 1, characterized in that, The operating condition prediction model includes at least a multi-source data preprocessing layer, a physical demand power calculation layer, an operating condition classification decision tree layer, and a time series prediction output layer. The multi-source data preprocessing layer is used at least to collect the multi-source vehicle data from each vehicle interface and at least to perform time synchronization, outlier removal and dimension normalization to obtain preprocessed data. The physical demand power calculation layer is at least used to calculate the current demand power item by item based on the preprocessed data and the vehicle longitudinal dynamics model. The working condition classification decision tree layer is used at least to determine the vehicle working condition type by taking the slope feature of the current power demand, the pedal intention feature and the navigation road condition label as input, and then outputting a discrete working condition label sequence. The time-series prediction output layer is used at least to perform a sliding window weighted average on the discrete operating condition label sequence to generate continuous prediction outputs for multiple prediction time points, thereby obtaining the predicted operating conditions of the vehicle.
4. The vehicle energy flow pre-visualization method according to claim 1, characterized in that, The predicted energy flow dataset is generated through at least a four-level inference link; The four-level simulation link includes the operating condition to energy mode link, the energy mode to power allocation link, the power allocation to flow direction mapping link, and the data integration and output link. The input of the operating condition to energy mode link includes at least the time series prediction structure output by the time series prediction output layer in the operating condition prediction model, and the output of the operating condition to energy mode link includes at least the energy mode enumeration value. The input of the energy mode to power allocation link includes at least the timing prediction structure and the energy mode enumeration value, and the output of the energy mode to power allocation link includes at least the power allocation structure. The input of the power allocation to flow mapping link includes at least the power allocation structure, and the output of the power allocation to flow mapping link includes at least a list of energy flow tuples. The inputs of the data integration output link include at least the energy flow tuple list, the discrete operating condition label sequence output by the operating condition classification decision tree layer in the operating condition prediction model, and the current demand power calculated by the physical demand power calculation layer in the operating condition prediction model. The output of the data integration output link includes at least the predicted energy flow dataset.
5. The vehicle energy flow pre-visualization method according to claim 1, characterized in that, The predicted operating conditions of the vehicle include at least constant speed driving, acceleration driving, deceleration coasting, long downhill braking, continuous uphill driving, idling parking, and traffic jam start-stop.
6. The vehicle energy flow pre-visualization method according to claim 1, characterized in that, On the same road segment, the prediction deviation between the predicted operating conditions of the vehicle and the multi-source vehicle data is calculated in stages. When the prediction deviation meets the preset deviation conditions, the online correction process is triggered.
7. The vehicle energy flow pre-visualization method according to claim 6, characterized in that, The online correction process includes at least the following: The real-time corrected power mapping coefficient is calculated iteratively based on at least a preset smoothing coefficient, an initial corrected power mapping coefficient, the multi-source vehicle data, and the predicted operating conditions of the vehicle, until the prediction deviation no longer meets the preset deviation condition.
8. The vehicle energy flow pre-visualization method according to claim 1, characterized in that, At least in response to the vehicle returning to the factory, the system acquires the predicted and measured comparison data for any trip of the vehicle, and then at least uses the least squares method to solve for the combination of physical model parameters that minimizes the preset objective function, with the measured power sequence as the objective. It also traverses the candidate thresholds of each node of the decision tree, re-selects the threshold combination that maximizes the classification accuracy based on the measured operating condition label, and packages the updated parameters and thresholds into an OTA firmware upgrade package, so that the vehicle can load the new parameter or threshold combination at least the next time it is powered on, thus completing the offline correction closed loop.
9. The vehicle energy flow pre-visualization method according to claim 1, characterized in that, The prediction accuracy of each working condition label in the working condition prediction model is statistically analyzed. If there is an abnormal working condition type whose prediction accuracy is consistently lower than a preset accuracy threshold, the decision tree structure is adjusted. The decision tree structure adjustment method includes at least: adjusting the integer threshold of the decision node corresponding to the abnormal operating condition type; adding new decision branch features; In addition, adjust the weighted attenuation coefficient of the sliding window.
10. A vehicle energy flow pre-visualization device, characterized in that, Used to perform the vehicle energy flow pre-visualization method according to any one of claims 1-9; The vehicle energy flow pre-visualization device includes at least: The operating condition prediction module is used to acquire multi-source vehicle data and input the multi-source vehicle data into the operating condition prediction model to obtain the predicted operating conditions of the vehicle. The dataset generation module is used to determine the energy flow direction, energy range and operating status between various vehicle components based at least on the vehicle's predicted operating conditions, the multi-source vehicle data and the vehicle energy management strategy, so as to generate a predicted energy flow dataset. The rendering warning module is used to at least use the predicted energy flow dataset as a visualization data source for the vehicle energy flow interface, to render the current energy flow and the predicted energy flow in layers, and to execute an energy flow anomaly warning.
