Bulk cargo terminal intelligent heavy truck predictive driving control method and system

Abstract

The application discloses a bulk cargo wharf intelligent heavy truck predictive driving control method and system, and belongs to the technical field of driving control. The method specifically comprises the following steps: obtaining environment information and port scheduling data of a heavy truck driving path, calculating energy demand of a future heavy truck driving process based on a pre-trained prediction model, generating a predictive driving strategy according to the energy demand and vehicle power parameters, wherein the predictive driving strategy comprises a speed distribution, a gear shifting timing and an acceleration / deceleration mode, establishing a heavy truck cooperative control model based on the predictive driving strategy, performing feedforward optimization on drive power distribution, and performing driving control on the heavy truck based on the heavy truck cooperative control model after the feedforward optimization. The application realizes real-time coordination and weight self-adaptive adjustment among driving, braking energy recovery and auxiliary electricity, reduces energy fluctuation and invalid consumption under complex port working conditions, and significantly improves the energy saving level, stability and intelligent degree of port transportation operation.

Description

Technical Field

[0001] This invention belongs to the field of driving control technology, specifically the predictive driving control method and system for intelligent heavy-duty trucks at bulk cargo terminals. Background Technology

[0002] With the continuous improvement of port logistics automation, bulk cargo terminal transportation is gradually shifting from traditional manual driving to unmanned and intelligent operations. Heavy-duty trucks, as the main horizontal transport equipment in bulk cargo terminals, directly impact the overall operating costs and carbon emission indicators of the port area due to their energy consumption levels. However, the operating environment of bulk cargo terminals is complex and variable, with uneven road gradients, large variations in ground adhesion coefficients, and dense distribution of loading and unloading machinery with frequent fluctuations in operating rhythms. This leads to problems such as frequent starts and stops, sudden power changes, and wasted braking energy during vehicle operation. Traditional constant power or fixed control strategies can no longer meet the dual demands of energy saving and efficient scheduling.

[0003] Existing intelligent driving control methods are mostly based on highway or fixed-route scenarios, focusing on the smoothness optimization of path planning and speed control, but failing to fully consider the task-driven operational characteristics of port environments. For example, bulk cargo terminal vehicles often need to simultaneously respond to yard scheduling systems, loading and unloading machinery commands, and port safety restrictions, resulting in nonlinear and aperiodic variations in their energy demand and power distribution. Existing control strategies mostly employ post-event feedback adjustment methods, which struggle to respond promptly to future changes in operating conditions, leading to problems such as energy allocation lag, drive system redundancy, and imbalances in coordinated control. Summary of the Invention

[0004] To address the shortcomings of existing technologies, this invention proposes a predictive driving control method and system for intelligent heavy-duty trucks at bulk cargo terminals. By predicting future driving conditions, the system adjusts vehicle power distribution and control logic in advance, and performs multi-domain coordinated control of drive, braking energy recovery, and auxiliary electrical systems to achieve intelligent driving control that optimizes energy utilization and coordinates with work rhythm.

[0005] To achieve the above objectives, the present invention provides the following technical solution:

[0006] A predictive driving control method for intelligent heavy-duty trucks at bulk cargo terminals includes:

[0007] Obtain environmental information and port area scheduling data for the heavy truck's driving route, and calculate the energy demand for the heavy truck's future driving process based on a pre-trained prediction model;

[0008] Based on the energy demand and vehicle power parameters, a predictive driving strategy is generated, which includes speed distribution, shift timing and acceleration / deceleration mode.

[0009] A heavy-duty truck cooperative control model is established based on a predictive driving strategy. The drive power distribution is optimized by feedforward, and the heavy-duty truck is driven based on the feedforward optimized heavy-duty truck cooperative control model.

[0010] Specifically, the step of acquiring environmental information about the heavy-duty truck's travel path and port area scheduling data, and calculating the energy demand for the future heavy-duty truck's travel process based on a pre-trained prediction model, includes:

[0011] The environmental information of the heavy truck's driving path is obtained, and edge preprocessing is performed on abnormal data. The environmental information includes road slope, ground adhesion coefficient, traffic density, dynamic position of loading and unloading machinery, and weather conditions.

[0012] Acquire port area scheduling data and establish port area scheduling data timestamps, and synchronize and integrate them with port area scheduling data. The port area scheduling data includes loading and unloading task priorities, berth congestion status, and operation time windows.

[0013] Spatial semantic modeling is performed on the preprocessed environmental information at the edge and the fused port area scheduling data to obtain a feature vector of the port area's operational status. The feature vector is used to reflect the future driving path of heavy trucks.

[0014] Based on the pre-trained prediction model, the feature vector is input, and a multi-time-scale energy demand curve is generated by simulating a continuous sequence of operating conditions for future driving paths. The simulation is completed before the vehicle is actually driven.

[0015] The energy demand prediction results are weighted and filtered based on model confidence and operating condition similarity. If the confidence is lower than the set threshold, local re-prediction is triggered, and the final energy demand prediction results of the heavy truck driving process are obtained.

[0016] Specifically, the pre-trained prediction model, inputting the feature vector, generates multi-time-scale energy demand curves by simulating continuous operating condition sequences of future driving paths, including:

[0017] The feature vector is divided into continuous time period nodes according to the port area operation rhythm, and a future path prediction sequence is generated in the form of time slices. The future path prediction sequence includes vehicle position assumptions, dynamic state assumptions of the operation area, and road resistance state evolution information.

[0018] The predicted future path sequence is input into a pre-trained prediction model, and the energy demand of the starting section, cruising section, yard interaction section and loading and unloading section are simulated according to the vehicle operation stage.

[0019] Based on the output of the prediction model, energy demand sets at different time scales are constructed according to a preset time window;

[0020] The energy demand set is checked for consistency along the timeline, and nodes that do not conform to the sequential logic are removed and reconstructed to generate energy demand curves at multiple time scales.

[0021] Specifically, based on the energy demand and vehicle power parameters, a predictive driving strategy is generated, including:

[0022] The energy demand curve is mapped and matched with the vehicle power parameters to form an input dataset including power response coefficient, inertia characteristics, transmission ratio range and energy recovery characteristics.

[0023] The strategy hierarchy is divided according to the dual dimensions of time domain and control domain, and the path planning layer, energy allocation layer and execution instruction layer are set.

[0024] Based on the input dataset and policy hierarchy, multiple candidate driving trajectories are generated. Each driving trajectory contains the corresponding speed distribution, gear shift sequence, and acceleration / deceleration node order information.

[0025] The candidate driving trajectories are checked for energy consistency and feasibility. Trajectories that violate the port area operation rhythm or vehicle power boundary are eliminated, and the set of strategies that meet the constraints are retained.

[0026] Within the set of strategies, a target strategy is selected based on task priority and time window index, and a predictive driving strategy is generated.

[0027] Specifically, based on the input dataset and policy hierarchy, multiple candidate driving trajectories are generated, including:

[0028] The parameter projection values ​​in the input dataset are unified into a spatiotemporal coordinate system to generate a multidimensional state matrix.

[0029] Based on the constraint rules of the path planning layer, energy allocation layer, and execution instruction layer in the strategy hierarchy, a trajectory generation constraint set is constructed.

[0030] Based on the multidimensional state matrix and trajectory generation constraint set, a multimodal candidate trajectory group is generated in a time-step manner;

[0031] The candidate trajectory group is calibrated for logical coherence according to the time series. If a logical break or node drift is detected, a local trajectory regeneration process is triggered until multiple candidate driving trajectories are formed.

[0032] Specifically, within the set of strategies, a target strategy is selected based on task priority and time window index, and a predictive driving strategy is generated, including:

[0033] Based on the port area's loading and unloading task scheduling information, operation type, and cargo type attributes, a task priority mapping table is generated;

[0034] Based on the port area's operational sequence and vehicle scheduling cycle, the execution period corresponding to the strategy is divided into multiple time windows, forming a multi-level time index structure.

[0035] Based on the task priority mapping table and time window index structure, the strategy set is preferentially filtered, and the strategy selection order is dynamically adjusted when different strategies overlap in time or task intervals.

[0036] The selected strategies will be used as predictive driving strategies.

[0037] Specifically, the establishment of a heavy-duty truck cooperative control model based on a predictive driving strategy, the feedforward optimization of drive power distribution, and the driving control of the heavy-duty truck based on the feedforward optimized heavy-duty truck cooperative control model include:

[0038] Based on the aforementioned predictive driving strategy, a heavy-duty truck cooperative control model is constructed, which includes a drive domain, a braking energy recovery domain, and an auxiliary electrical domain, and a parameter interaction interface is established between each control domain.

[0039] Based on vehicle power parameters and energy demand curves, a power allocation prediction matrix is ​​formed through a time stepping method. The power allocation prediction matrix defines the power demand ratio of each control domain at different time nodes.

[0040] Based on the power allocation prediction matrix, the driving power allocation is corrected in real time, and an optimized power trajectory is generated through a multi-constraint search algorithm.

[0041] The optimized power trajectory is input into the heavy-duty truck collaborative control model to reconstruct the control weight distribution and execution timing logic, and output control commands.

[0042] The vehicle is driven in different time periods according to the control instructions of the heavy truck collaborative control model.

[0043] Specifically, based on the power allocation prediction matrix, the driving power allocation is corrected in real time, and an optimized power trajectory is generated through a multi-constraint search algorithm, including:

[0044] Extract the set of dynamic constraint parameters for the current time period from the power allocation prediction matrix;

[0045] Based on the dynamic constraint parameter set and the predictive driving strategy, a power correction reference model is established to generate a power benchmark sequence. The power correction reference model takes time step logic as the main line and performs differential mapping between historical energy consumption trajectory and current power prediction value.

[0046] Under the power correction reference model, the power trajectory is explored in multiple dimensions to generate multiple candidate power trajectories;

[0047] The candidate power trajectories are subjected to consistency screening to select the target trajectory that conforms to the dynamic constraint parameter set of the current time period, and an optimized power trajectory is generated.

[0048] Specifically, the optimized power trajectory is input into the heavy-duty truck cooperative control model to reconstruct the control weight distribution and execution timing logic, and output control commands, including:

[0049] The optimized power trajectory is input into the heavy-duty truck cooperative control model, and a corresponding time index mapping table is established inside the heavy-duty truck cooperative control model to identify the target power output node for each time period.

[0050] Based on the node distribution of the optimized power trajectory, the control weight ratios among the drive domain, braking energy recovery domain, and auxiliary electrical domain are dynamically adjusted to form a weight distribution matrix that evolves over time.

[0051] Based on the weight distribution matrix and the time index mapping table, the execution timing logic chain of the control instructions is constructed;

[0052] The timing logic chain is parsed into an executable set of control instructions, which includes torque regulation, energy recovery triggering, and power switching.

[0053] A predictive driving control system for intelligent heavy-duty trucks at bulk cargo terminals, used to implement the predictive driving control method for intelligent heavy-duty trucks at bulk cargo terminals, includes: a demand prediction module, a driving strategy generation module, and a driving control module;

[0054] The demand forecasting module is used to acquire environmental information and port scheduling data of the heavy truck's driving path, and calculate the energy demand of the heavy truck's future driving process based on the pre-trained forecasting model.

[0055] The driving strategy generation module is used to generate a predictive driving strategy based on the energy demand and vehicle power parameters.

[0056] The driving control module establishes a heavy-duty truck cooperative control model based on a predictive driving strategy, performs feedforward optimization of drive power distribution, and performs driving control of the heavy-duty truck based on the feedforward optimized heavy-duty truck cooperative control model.

[0057] Compared with the prior art, the beneficial effects of the present invention are:

[0058] This invention proposes a predictive driving control method and system for intelligent heavy-duty trucks at bulk cargo terminals. It constructs a predictive driving model with time-series perception capabilities to predict energy consumption trends and plan strategies before the vehicle actually drives. Through a hierarchical collaborative control structure, it achieves real-time coordination and adaptive weight adjustment between drive, braking energy recovery, and auxiliary electrical systems, thereby reducing energy fluctuations and ineffective consumption under complex port conditions. It significantly improves the energy-saving level, stability, and intelligence of port transportation operations, and has broad engineering application value and promotion potential. Attached Figure Description

[0059] Figure 1 The flowchart of the intelligent heavy-duty truck predictive driving control method for bulk cargo terminals provided by the present invention is shown below.

[0060] Figure 2 The energy demand curve provided for this invention;

[0061] Figure 3 The driving control diagram provided by this invention;

[0062] Figure 4 This is a diagram illustrating the architecture of the intelligent heavy-duty truck predictive driving control system for bulk cargo terminals provided by the present invention. Detailed Implementation

[0063] The present application will now be described in detail with reference to specific embodiments. These embodiments will help those skilled in the art to further understand the present application, but do not limit the present application in any way. It should be noted that those skilled in the art can make several modifications and improvements without departing from the concept of the present application. These all fall within the protection scope of the present application.

[0064] To make the objectives, technical solutions, and advantages of this application clearer, the following detailed description is provided in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the scope of this application.

[0065] It should be noted that, unless there is a conflict, the various features in the embodiments of this application can be combined with each other, all of which are within the protection scope of this application. Furthermore, although functional modules are divided in the device schematic diagram and a logical order is shown in the flowchart, in some cases, the steps shown or described can be executed in a different order than the module division in the device or the order in the flowchart. In addition, the terms "first," "second," and "third" used in this application do not limit the data or execution order, but only distinguish identical or similar items with essentially the same function and effect.

[0066] Unless otherwise defined, all technical and scientific terms used in this specification have the same meaning as commonly understood by one of ordinary skill in the art to which this application belongs. The terminology used in this specification is for the purpose of describing particular embodiments only and is not intended to limit the scope of this application. The term "and / or" as used in this specification includes any and all combinations of one or more of the associated listed items.

[0067] Example 1

[0068] Please see Figures 1-3 The present invention provides an embodiment of a predictive driving control method for intelligent heavy-duty trucks at bulk cargo terminals, comprising the following specific steps:

[0069] Step S1: Obtain environmental information and port area scheduling data for the heavy truck's driving route, and calculate the energy demand for the heavy truck's future driving process based on the pre-trained prediction model.

[0070] The specific steps of step S1 are as follows:

[0071] Step S101: Obtain environmental information of the heavy truck's driving path and perform edge preprocessing on abnormal data. The environmental information includes road slope, ground adhesion coefficient, traffic density, dynamic position of loading and unloading machinery, and weather conditions.

[0072] In this embodiment, high-precision acquisition of environmental information and real-time correction of abnormal data are achieved by fusing multimodal sensing information with an edge computing model. Specifically, during the operation of the heavy truck, multi-source sensing units distributed throughout the vehicle body and work area, including slope sensors, lidar, ground condition detection modules, traffic monitoring cameras, and meteorological acquisition nodes, continuously collect road and surrounding work environment data. To ensure the real-time performance and reliability of the data, each sensing signal is first synchronized and decoupled in the on-board edge computing unit to form a basic feature sequence. Subsequently, a local consistency analysis of the feature sequence is performed using preset dynamic window filtering and temporal clustering algorithms, such as density-based or similarity-based clustering methods. Anomalies and abrupt changes are identified based on spatial continuity and numerical gradient variation patterns. When abnormal data deviating from historical distribution characteristics is detected, reference samples from adjacent time periods are called for differential interpolation and feature reconstruction to restore missing or abnormal sensing information. Through this process, an environmental information set that meets the real-time control accuracy requirements is generated.

[0073] Step S102: Obtain port area scheduling data and establish a port area scheduling data timestamp, and synchronize and merge it with the port area scheduling data. The port area scheduling data includes loading and unloading task priority, berth congestion status and operation time window.

[0074] In this embodiment, a time-series synchronization mechanism is used to achieve dynamic fusion of port area dispatch data and heavy truck operation information. Specifically, the port area dispatch center generates loading and unloading task data, berth status information, and operation time planning in real time. This data is output in the form of message streams through the port area digital twin platform and is periodically acquired by the dispatch access module on the heavy truck side. To ensure the timeliness consistency of data in transmission and fusion, the dispatch data is marked with a unified timestamp by a clock synchronization protocol before entering the fusion process, and a multi-layer index structure is established based on task type, berth location, and operation number. Subsequently, the heavy truck side uses the timestamp as the primary key to associate and match the dispatch data with the vehicle's own positioning, speed, load, and driving path information, and determines the synchronization node closest to the operation time through a time sliding window alignment algorithm. In this process, the start and stop boundaries of each task window and the berth status change trend are derived by utilizing the temporal continuity of the dispatch information, and a fusion dataset that is updated synchronously with the vehicle's dynamic information is formed, thereby achieving accurate mapping of dispatch data in both spatial and temporal dimensions.

[0075] It should be noted that, based on the time-series characteristics of the scheduling data, the status information of continuously arriving tasks, such as task creation, execution, completion identifiers, and berth occupancy status, is sorted and differentially analyzed by timestamp. By identifying the points of change in task status between adjacent time nodes, such as from "not started" to "in execution" or from "occupied" to "released," the start and end boundaries of the task window are determined. At the same time, the occupancy rate, switching frequency, and duration of berth status within a continuous period are statistically analyzed using a sliding time window. Through trend fitting or state transition probability analysis, the dynamic evolution process of berths from idle, congested to released is depicted, thereby deriving the start and stop intervals of the task execution window and the trend of berth status changes over time.

[0076] Step S103: Perform spatial semantic modeling on the preprocessed environmental information at the edge and the fused port area scheduling data to obtain the feature vector of the port area operation status. The feature vector is used to reflect the future driving path of heavy trucks.

[0077] In this embodiment, a spatial semantic fusion model based on multi-source data is constructed to achieve a characteristic expression of the port area's operational status. Specifically, the environmental information after edge preprocessing and the synchronously fused scheduling data are first mapped to a unified port area geographic reference framework. Within this framework, a multi-dimensional semantic graph structure containing terrain, traffic, operation, and time attributes is established, using road nodes, loading and unloading points, and berth channels as spatial primitives. Subsequently, a feature extraction algorithm based on graph topology association is used to model the dynamic relationships between nodes, extracting a set of semantic features reflecting changes in road slope, frequency of interference from loading and unloading machinery, directionality of traffic flow, and time density of operations. To enhance the model's temporal adaptability, a time-sliding window mechanism is introduced to encode the semantic evolution process within continuous time periods into a quantifiable feature sequence, enabling each node to possess both spatial location attributes and temporal behavior attributes in the semantic graph. Finally, based on the association weights and semantic similarity between nodes, a high-dimensional feature vector is generated to characterize the dynamic characteristics and potential operating condition change trends of future heavy truck driving paths.

[0078] It should be noted that after constructing the semantic graph structure, the edge weights are first calculated based on the spatial connectivity between nodes, the strength of operational associations, and the frequency of traffic interactions. Then, the similarity of semantic features, such as slope, flow rate, and operational density, is combined to construct a node similarity matrix. Subsequently, through graph embedding or weighted aggregation methods, such as feature aggregation based on neighborhood propagation or attention mechanisms, the features of each node and its neighboring nodes are weighted and fused according to their association weights to form a node representation that can reflect the local topology and semantic consistency. Furthermore, the node representations are spliced, pooled, or dimensionality-reduced globally to generate a high-dimensional feature vector that uniformly expresses the port area's operational status.

[0079] Step S104: Based on the pre-trained prediction model, input the feature vector, and generate energy demand curves at multiple time scales by simulating a continuous sequence of operating conditions for future driving paths. The simulation is completed before the vehicle actually drives.

[0080] like Figure 2 As shown, the specific steps of step S104 are as follows:

[0081] Step S1041: Divide the feature vector into continuous time period nodes according to the port area operation rhythm, and generate a future path prediction sequence in the form of time slices. The future path prediction sequence includes vehicle position assumptions, dynamic state assumptions of the operation area, and road resistance state evolution information.

[0082] In this embodiment, a time-series segmentation mechanism based on operational rhythm is introduced to achieve dynamic deconstruction of feature vectors and generation of path prediction sequences. Specifically, the port area operational rhythm is regarded as a time-driven scheduling signal source, which includes features such as loading and unloading cycles, berth switching intervals, and peak traffic flow distribution. First, the port area situation feature vector obtained in step S103 is segmented into time segments according to the periodic change law of the operational rhythm, so that each time segment node corresponds to an independent operational stage. Then, using the time slice structure after node segmentation, the possible driving area of ​​heavy trucks is spatially projected in each segment, and the hypothetical location distribution of vehicles is deduced through path topology constraints and scheduling priority parameters. In order to characterize the dynamics of the port area operating environment, dynamic state variables of the operational area are further embedded in each time slice, including yard loading density, mechanical activity range, and traffic accessibility indicators. Combined with the historical friction coefficient change trend and real-time slope data, an evolution sequence of road resistance state is established. After the above time-series decomposition and spatial reconstruction, a future path prediction sequence containing vehicle position assumptions, operational state evolution, and road resistance changes is formed.

[0083] It should be noted that, within each time slice, the historically collected ground adhesion coefficient sequence is first analyzed for trend, and its fluctuation pattern and rate of change over time are extracted as the basic trend term for resistance change; at the same time, the real-time slope data at the corresponding moment is obtained, and the change of gravity component caused by slope is mapped as a resistance correction term; then, the historical trend term and the real-time correction term are weighted and fused, and combined with the weight difference of the impact of vehicle operation stage on resistance, the comprehensive road resistance value at each time node is recursively calculated, thus forming a resistance state sequence that evolves continuously over time.

[0084] Step S1042: Input the future path prediction sequence into the pre-trained prediction model, and perform energy demand simulation for the starting section, cruising section, yard interaction section and loading and unloading section according to the vehicle operation stage.

[0085] In this embodiment, a phase-based energy demand extrapolation mechanism is used to perform hierarchical energy consumption calculation and dynamic scenario simulation of the future path prediction sequence. Specifically, the pre-trained prediction model pre-establishes the energy consumption mapping relationship of typical operating stages in the port area, with each stage corresponding to a specific power response mode and load characteristics. First, the future path prediction sequence generated in step S1041 is input into the model's time-series analysis unit, and the driving process is divided into stages based on the speed gradient, operation density, and task identifier between path nodes, thereby distinguishing the starting section, cruising section, yard interaction section, and loading / unloading section. Subsequently, the model calls the corresponding energy sub-model within each stage, and constructs the probability distribution curve of stage energy consumption by solving the power demand, load changes, and resistance distribution in parallel. To maintain the continuity and physical rationality of the prediction, the model introduces a transition state compensation factor at the stage boundary to balance the connection error of the energy curves of each stage. Finally, based on the energy distribution results of multiple stages, the model forms a comprehensive energy consumption sequence covering the entire path, realizing a complete extrapolation of energy demand within the future driving cycle.

[0086] It should be noted that by analyzing the speed change rate of adjacent path nodes, a high-speed gradient section from standstill to continuous acceleration is identified as the start-up section; when the speed change tends to stabilize and the operation density is low, it is classified as the cruising section; when there is high operation density or frequent interaction of task markers in the area corresponding to the node, it is classified as the yard interaction section; when the task marker points to loading and unloading operations and the speed is close to low speed or stagnant, it is identified as the loading and unloading section. Based on this, each stage calls the corresponding energy sub-model to jointly calculate the power demand, load change, and road resistance to obtain the energy consumption distribution results of that stage, and form a probability distribution curve through statistical methods. At the same time, a transition state compensation factor is introduced at the boundary between adjacent stages to smoothly correct the power and energy changes and avoid energy abrupt changes caused by stage switching. The transition state compensation factor comes from the need to correct the discontinuity of energy changes at the boundary between adjacent operating stages. It is constructed by analyzing the differences in power output, speed change, and load characteristics before and after stage switching. Specifically, it can be obtained by fitting the energy difference, power gradient change, and historical similar working condition data at the boundary between adjacent stages.

[0087] Step S1043: Based on the output of the prediction model, construct energy demand sets at different time scales according to a preset time window.

[0088] In this embodiment, the output of the prediction model is expressed in a multi-scale energy structure using the principle of time-level aggregation. Specifically, the pre-trained prediction model generates a continuous phase energy consumption sequence in step S1042, which reflects the energy distribution trend within the future travel cycle with time as the main axis. To extract energy features at different time levels, multi-level time windows are first set according to the port operation scheduling cycle and vehicle running sequence, including short-term windows, medium-term windows, and extended windows. Each window has hierarchical differences in time span and sampling density. Subsequently, the prediction sequence is aggregated along the window boundaries, and parameters such as energy fluctuation amplitude, power peak frequency, and load persistence are statistically analyzed within each window. Energy demand subsets are formed through time-domain smoothing and energy interval discretization. To ensure the coherence between windows, the boundary data is further normalized and corrected using a window overlap strategy, so that short-term energy changes and long-term trends have consistent evolutionary logic. After the above-mentioned hierarchical aggregation, an energy demand set containing multiple time scales is generated. This set records the dynamic hierarchical structure of energy distribution over time in a unified data format.

[0089] It should be noted that a certain proportion of overlapping intervals are set between adjacent time windows. The energy data in the overlapping intervals are simultaneously included in the statistical calculations of both windows. The energy values ​​in the intervals are then merged through weighted averaging or normalization to ensure that the values ​​in the two windows remain consistent or transition smoothly. Specifically, the overlapping interval data can be assigned decreasing or increasing weights according to the time position, so that the data closer to the current window has a higher weight, thereby eliminating the abrupt changes or discontinuities caused by segmented processing at the window boundaries.

[0090] Step S1044: Perform consistency checks on the energy demand set according to the time link, and remove and reconstruct nodes that do not conform to the sequential logic to generate energy demand curves at multiple time scales.

[0091] In this embodiment, a time-link consistency verification mechanism is used to logically verify and structurally reconstruct the energy demand set across multiple time scales, ensuring the continuity and computability of the prediction results. Specifically, the energy demand set generated in step S1043 is first connected into a time-series linked list structure according to chronological order, and an energy change link is established using node timestamps as indices. Subsequently, the energy gradient change rate between adjacent nodes is calculated through time-series differential analysis, and this change rate is compared with the energy change model corresponding to the port operation rhythm and vehicle operation phase to identify nodes that do not conform to the time-series rules, such as energy abrupt changes, reverse transitions, or abnormal stability. For nodes determined to be logically abnormal, their energy values ​​are recalculated using energy trend interpolation of adjacent segments and time-domain smoothing reconstruction technology to restore the temporal continuity of the energy curve. During the reconstruction process, multi-level verification thresholds are set to distinguish between local anomalies and global offsets, thereby achieving node-level correction while maintaining the conservation of total energy. After this processing, all energy nodes have a monotonically traceable logical order in the time link and are recombined into a complete energy demand curve with multi-scale correspondence.

[0092] It should be noted that differential calculations are performed on adjacent energy nodes in the time series. The energy gradient change rate is obtained by calculating the energy change per unit time, and this change rate is compared with the standard change range of different operating stages to identify sudden changes or abnormal fluctuations. At the same time, multi-layer verification thresholds are set, including local thresholds for detecting anomalies in individual nodes and global thresholds for evaluating the overall energy trend deviation. When the change of a node exceeds the local threshold, it is judged as a local anomaly, and when multiple consecutive nodes deviate from the reference trend by more than the global threshold, it is judged as a global deviation. This enables hierarchical identification and targeted correction of anomaly types.

[0093] Step S105: The energy demand prediction results are weighted and filtered based on the model confidence and operating condition similarity. If the confidence is lower than the set threshold, local re-prediction is triggered, and the final energy demand prediction results of the heavy truck driving process are obtained.

[0094] In this embodiment, a dual-index screening mechanism of confidence level and operating condition similarity is used to dynamically weight and locally re-evaluate the energy demand prediction results, obtaining a stable and highly reliable energy demand output. Specifically, after generating multi-timescale energy demand curves, the pre-trained prediction model attaches a corresponding model confidence parameter to each prediction node. This parameter is calculated by combining the fit of historical samples and the deviation of real-time data. Simultaneously, considering the current port area operating environment, key features such as road slope changes, traffic density fluctuations, and loading / unloading rhythm frequency are extracted from environmental information and scheduling data to construct an operating condition feature vector, which is then compared with typical operating condition samples from the model training phase. Subsequently, the energy prediction results are sorted according to confidence level and similarity. Similarity is weighted and fused to form an energy consumption assessment matrix that reflects the prediction confidence interval. When the confidence level is found to be lower than the preset threshold in a certain time period (by analyzing the prediction error distribution in the training and validation stages, the combination of the error mean and standard deviation or the percentile range is taken as the threshold interval, for example, the confidence threshold can be set to 80% to 95%) or the operating condition similarity deviates from the training sample distribution, the model automatically triggers a local re-prediction mechanism. By reloading the input features of the corresponding stage through a limited time window and performing local energy recalculation, the accumulated error of the prediction anomaly is eliminated. After this dual screening and re-prediction process, the heavy truck energy demand prediction result that meets the confidence requirements, is time-series stable, and has a high degree of matching with the actual operating conditions is obtained.

[0095] It should be noted that at each time point, the energy prediction value output by the model, the corresponding confidence parameter, and the similarity to the operating condition are used as the weighting basis. First, the confidence and similarity are normalized to obtain weight coefficients that reflect the reliability of the prediction and the degree of matching with the operating condition. Then, the original prediction results are fused by weighted summation or weighted average, so that the prediction values ​​with high confidence and high matching with the current operating condition have a larger proportion, while the weight of the prediction results with low confidence or large similarity deviation is weakened, thereby generating an energy consumption assessment matrix that comprehensively considers the model's credibility and the adaptability to the operating condition.

[0096] Figure 2 The diagram illustrates the dynamic response characteristics of the energy demand curve in the predictive driving control method for intelligent heavy-duty trucks at bulk cargo terminals. The horizontal axis represents time, and the vertical axis represents energy demand intensity. Curves of different colors correspond to the energy demand change trends at different time scales. Specifically, the blue curve represents short-term energy demand, reflecting the energy fluctuation characteristics of vehicles under transient conditions such as starting sections, acceleration on slopes, and frequent braking. The orange curve represents medium-term energy demand, corresponding to the energy distribution changes of vehicles within a single operating cycle (such as a round trip between the yard and berth). The green curve represents long-term energy demand, reflecting the overall trend of energy consumption over multiple operating cycles. The red curve represents ultra-long-term energy demand, used to describe the cumulative changes in energy consumption within the overall scheduling cycle of the port area.

[0097] In the energy prediction modeling process, the input features at different time scales are first extracted through the environmental perception module and the scheduling fusion module, and then input into the multi-time scale prediction model for hierarchical extrapolation. The model fits the energy changes through a time sliding window mechanism, thereby obtaining a continuous and smooth energy demand curve at each time scale. The short-time and medium-time curves are mainly used for real-time driving and energy recovery and distribution control, while the long-time and ultra-long-time curves are used to optimize the port area operation rhythm and charging and discharging strategies. Through this multi-time scale energy demand curve, hierarchical prediction and dynamic control of the energy distribution of heavy trucks are achieved.

[0098] Step S2: Based on the energy demand and vehicle power parameters, generate a predictive driving strategy, which includes speed distribution, shift timing and acceleration / deceleration mode.

[0099] The specific steps of step S2 are as follows:

[0100] Step S201: Map and match the energy demand curve with the vehicle power parameters to form an input dataset including power response coefficient, inertia characteristics, transmission ratio range and energy recovery characteristics.

[0101] In this embodiment, the energy demand curve output in step S105 is first discretized according to the time series, using the energy change rate and power gradient as index variables, and the energy consumption distribution characteristics in each time period are extracted. Then, based on the vehicle's structured dynamics model, basic characteristic data such as engine rated power, drive motor peak torque, transmission ratio range, vehicle rolling resistance, and energy recovery efficiency are retrieved from the power system parameter library. By establishing a multi-dimensional mapping relationship between energy demand characteristics and power system parameters, the energy consumption change trend and power response capability are matched one-to-one. In this process, a time-domain correction coefficient of inertia characteristics is introduced to correct the impact of vehicle mass changes and loading status on energy response lag. At the same time, according to the efficiency curves of different transmission ratio ranges, the energy demand is allocated into an executable power demand matrix in each gear. Finally, the matched parameter results are aggregated to form an input dataset containing power response coefficients, inertia characteristics, transmission ratio ranges, and energy recovery characteristics.

[0102] It should be noted that in the process of matching energy demand with power parameters, a time-domain correction coefficient of inertia characteristics is first introduced based on changes in vehicle mass and load to dynamically adjust the rate of energy change, in order to compensate for the lag in power response caused by the increase or decrease in vehicle inertia, so that the energy demand curve can more accurately reflect the actual power output rhythm. At the same time, combined with the efficiency characteristic curves corresponding to different transmission ratio ranges, the corrected energy demand is decomposed and mapped according to the transmission efficiency and output capacity of each gear. Under the principle of satisfying the power boundary and the optimal efficiency, the total energy demand is allocated to the power range corresponding to each gear, forming an executable power demand matrix.

[0103] Step S202: Divide the strategy hierarchy according to the dual dimensions of time domain and control domain, and set the path planning layer, energy allocation layer and execution instruction layer.

[0104] In this embodiment, a hierarchical predictive driving strategy decision-making architecture is constructed using a dual-dimensional partitioning method of the time domain and control domain. Specifically, firstly, based on the temporal distribution characteristics of the energy demand curve, the entire driving task cycle is divided into multiple time domain intervals, including a short-term response zone, a mid-range adjustment zone, and a long-term planning zone, to correspond to different decision-making time scales. Subsequently, a three-layer strategy structure is established in the control logic: a path planning layer, an energy allocation layer, and an execution command layer. The path planning layer uses the longest prediction window in the time domain as the control benchmark and is responsible for determining the vehicle's spatial route and desired speed trajectory in the future driving cycle. The energy allocation layer optimizes the power flow direction and energy recovery mode of the power system in different time periods based on the output results of the path planning layer, forming an executable energy control sequence. The execution command layer operates within the minimum step size in the time domain, generating specific control signals based on the target power command of the energy allocation layer, and maintaining information consistency between strategy layers through a closed-loop update mechanism. In this process, a mapping index table is established between the time domain and the control domain to realize the context dependency and dynamic constraint transmission between different strategy levels.

[0105] Step S203: Based on the input dataset and policy hierarchy, generate multiple candidate driving trajectories. Each driving trajectory contains the corresponding speed distribution, gear shift sequence, and acceleration / deceleration node order information.

[0106] The specific steps of step S203 are as follows:

[0107] Step S2031: Unify the parameter projection values ​​in the input dataset into a spatiotemporal coordinate system to generate a multidimensional state matrix.

[0108] In this embodiment, the input dataset contains heterogeneous parameters from multiple sources, such as vehicle dynamic response coefficient, inertia characteristics, transmission ratio range, energy recovery characteristics, and energy consumption change trends. These parameters have different temporal resolutions and spatial reference systems during acquisition. To achieve a unified representation of the data, the values ​​of each parameter are first mapped to a spatiotemporal coordinate system with the vehicle's center of mass as the origin, based on the vehicle's spatial trajectory in the port area. The time axis represents the continuous evolution of the vehicle's driving cycle, and the spatial axis represents the pose changes under the path topology. Subsequently, parameter normalization and temporal interpolation methods are used to align data from different sampling frequencies in the time dimension. At the same time, a feature projection algorithm is used to project and superimpose each parameter in the spatial dimension according to the position nodes, forming a multidimensional coordinate point set reflecting the vehicle's operating state changes over time. Based on this, a correlation matrix between parameters is established using spatiotemporal coupling constraints, and a multidimensional state matrix containing time steps, position nodes, and dynamic characteristic variables is generated through matrix expansion.

[0109] Step S2032: Construct a trajectory generation constraint set based on the constraint rules of the path planning layer, energy allocation layer and execution instruction layer in the strategy hierarchy.

[0110] In this embodiment, a constraint set for candidate driving trajectory generation is constructed through abstract modeling of multi-layer constraint logic. Specifically, based on the strategy hierarchy established in step S202, the path planning layer, energy allocation layer, and execution command layer correspond to different control objectives and boundary conditions. First, in the path planning layer, spatial constraint rules are set based on the port area road topology, berth distribution, and operational priorities, including traversable areas, multi-path intersection conflict restrictions, and minimum turning radius conditions. Second, in the energy allocation layer, energy constraint conditions are set by combining the energy demand curve with the power system output characteristics. The system first defines the upper limit of power, the lower limit of energy recovery, the instantaneous load change rate, and the system coupling stability threshold. Then, a control constraint module is established within the execution instruction layer to define the dynamic boundaries of control variables such as acceleration change limits, shift trigger delays, and braking response intervals. To ensure the consistency of the three-layer constraints during trajectory generation, a time-domain constraint is paired with a spatial domain constraint through a time-series mapping algorithm to form a joint constraint index table, and inter-layer priority identifiers are introduced to determine the conflict handling order. Finally, a trajectory generation constraint set is formed based on the above multi-layer rule set, and this set stores the control conditions and boundary parameters in a logical tree structure.

[0111] For example, in a port area transportation task, the path planning layer restricts vehicles to travel only within designated passageways based on the road topology from the yard to the berth, and sets intersection avoidance constraints and minimum turning radius constraints; the energy distribution layer limits the upper limit of driving power to within 80% of the rated power based on uphill conditions and energy demand curves, while setting the minimum energy recovery ratio and power change smoothing threshold during braking; the execution command layer constrains the rate of acceleration change to not exceed the set limit, and limits the shift interval and braking response delay range; during trajectory generation, the above spatial path constraints, energy constraints and control constraints are jointly matched by time steps through temporal mapping. When path conflicts or power over-limits occur, the trajectory is adjusted or reconstructed according to the inter-layer priority to satisfy safety and spatial constraints, thereby forming a candidate driving trajectory that satisfies multi-layer constraint conditions.

[0112] Step S2033: Based on the multidimensional state matrix and trajectory generation constraint set, generate a multimodal candidate trajectory group in a time-step manner.

[0113] In this embodiment, a multimodal trajectory evolution mechanism based on time steps is used to generate candidate trajectories for heavy trucks in complex port environments. Specifically, the multidimensional state matrix formed in step S2031 is first used as the spatiotemporal feature input source. The vehicle's power parameters, spatial pose nodes, and energy demand features are temporally expanded to construct a continuous state evolution chain. Subsequently, the trajectory generation constraint set established in step S2032 is embedded into the trajectory deduction process, simultaneously constraining the vehicle's spatial position changes, energy output range, and control command execution timing in each time step. To achieve both trajectory diversity and stability, a multimodal generation logic is introduced, dividing the vehicle's operating mode into three sub-modes: energy-saving cruise mode, dynamic avoidance mode, and operation priority mode. Each mode corresponds to a different path bias factor and power allocation strategy. During trajectory generation, the time step is progressively advanced through a recursive search algorithm. In each step of the calculation, the spatial position, velocity vector, and energy allocation state are deduced in parallel to form multiple candidate trajectory branches. Through this time-sequence recursion and multimodal combination generation process, a multimodal candidate trajectory group covering different operating modes and energy consumption strategies is finally obtained.

[0114] It should be noted that the vehicle operation modes are divided into three sub-modes: the energy-saving cruise mode aims for optimal energy, with its path bias factor tending to select routes with gentle slopes and less traffic interference, and adopting a distribution strategy of smooth power output and high energy recovery ratio; the dynamic avoidance mode prioritizes safety and traffic efficiency, with its path bias factor dynamically adjusted according to real-time traffic density and obstacle distribution, prioritizing alternative routes with low conflict probability, and adopting a power distribution strategy of fast power response and moderate redundancy to support acceleration and deceleration maneuvers; and the task priority mode focuses on task timeliness, with its path bias factor tending to the shortest operation path or priority route, and increasing the proportion of drive output and reducing the weight of energy recovery in power distribution to ensure that the vehicle can quickly complete the transportation task within the loading and unloading window.

[0115] Step S2034: Perform logical coherence calibration on the candidate trajectory group according to the time sequence. If a logical break or node drift is detected, trigger the local trajectory regeneration process until multiple candidate driving trajectories are formed.

[0116] In this embodiment, a temporal logic consistency calibration mechanism is used to dynamically verify and locally regenerate the multimodal candidate trajectory group, ensuring the continuity and executability of the trajectory in terms of time and control logic. Specifically, the candidate trajectory group generated in step S2033 is first reconstructed into a continuous temporal linked list structure according to the timestamp order, and a logic consistency judgment model is established with the spatial position, velocity vector, and power output state between trajectory nodes as core parameters. This model calculates the trajectory offset rate and control command hysteresis of continuous segments based on the state difference relationship between nodes to identify possible logic breakpoints or node drift areas in the trajectory. When the spatiotemporal difference between the judgment nodes exceeds a set threshold, the set threshold is applied to the trajectory nodes during the training and testing phases. The position difference, velocity difference, and power variation amplitude are statistically analyzed. The combination of the mean and standard deviation or the percentile is set as the threshold interval to trigger a local regeneration process. That is, the trajectory generation algorithm is called again within a local time window centered on the abnormal node. The path, power, and control variables of the segment are resolved under constraints to restore trajectory continuity. After the trajectory regeneration is completed, logical backtracking verification is performed to smoothly adjust the connection between the new trajectory segment and the adjacent time sequence nodes, and the global trajectory index table is updated. After the above cyclic calibration and local reconstruction processing, all trajectories maintain logical coherence in the time series and have resolvability and dynamic consistency at the control level, ultimately forming multiple candidate driving trajectories that meet the input requirements of the predictive driving strategy.

[0117] It should be noted that, using trajectory nodes at adjacent time steps as analysis units, the spatial position difference, velocity vector difference, and power output difference are extracted to construct a state difference vector. The trajectory offset rate is calculated by the continuity deviation of position changes to measure path smoothness and spatial consistency. Simultaneously, the control command lag is calculated by combining the temporal matching relationship between velocity changes and power response to characterize the response delay of power output to changes in motion state. Based on this, the offset rate and lag are compared with a preset threshold range. The preset threshold range is obtained by statistically analyzing the distribution of spatial position difference, velocity difference, and power output difference of trajectory nodes during the training and verification phases, and taking the mean ± k times the standard deviation or percentile range as the threshold range. This threshold range can be fine-tuned according to vehicle type, driving environment, sampling time interval, and control strategy, and is calibrated or dynamically adaptively adjusted using historical data before deployment. When any indicator shows a sudden change or cumulative deviation, it is judged as logical inconsistency, thereby achieving a joint judgment on trajectory continuity and control timing rationality.

[0118] Within adjacent time steps, the actual displacement change is calculated based on the spatial position difference of the trajectory nodes, and compared with the ideal path or the continuous displacement trend of the preceding and following nodes to obtain the position deviation. Then, the deviation is normalized by combining the time interval to form the trajectory offset rate. At the same time, the velocity change is used as the state response input, and it is time-series aligned with the power output change of the corresponding time step. By calculating the time difference or correlation offset required for the two to reach consistent change, the control command hysteresis is obtained.

[0119] Step S204: Perform energy consistency and feasibility checks on the candidate driving trajectories, eliminate trajectories that violate the port area operation rhythm or vehicle power boundary, and retain the set of strategies that meet the constraints.

[0120] In this embodiment, a joint energy consistency and feasibility verification mechanism is used to screen candidate driving trajectories and extract constraints from the strategy set. Specifically, firstly, multiple candidate trajectories output in step S2034 are input into the energy consistency analysis module in chronological order. The energy consumption curves of each trajectory are compared by integration to calculate the energy balance difference between trajectories. The physical rationality of its energy distribution is verified based on the vehicle power boundary model. Subsequently, combined with the operation rhythm data of the port area scheduling system, the speed change nodes, stopping periods, and path occupancy time in the trajectory are compared synchronously to determine whether the trajectory is consistent with the port area operation window. The trajectory is consistent with the berth scheduling plan; for trajectories with sudden changes in energy consumption, power allocation exceeding the limit, or conflict with the operation time, they are marked as infeasible and eliminated according to preset rules; on this basis, the stability of each trajectory in terms of driving torque, braking power, and energy recovery frequency is further evaluated through the control boundary analysis model to identify abnormal trajectories with power overload or control command conflict; after multi-layer verification, the set of trajectories that simultaneously meet the energy consistency, operation rhythm matching, and power boundary constraints is retained, and this is used as the candidate strategy input for subsequent predictive driving strategy selection and command generation to realize the logical closed loop from energy consumption prediction to control execution.

[0121] It should be noted that the preset rules include: power boundary constraints based on the vehicle dynamics model, such as maximum / minimum output power and upper limit of power change rate; energy balance constraints based on energy conservation relationship, such as the need for a closed relationship between driving energy, recovered energy and auxiliary energy consumption; smoothness constraints based on energy curve characteristics, such as the power gradient between adjacent time steps must not exceed a set threshold (by statistically analyzing the energy curves and power change rates during the training and validation phases, taking the combination of the mean and standard deviation or setting a percentile as the threshold); and reasonableness constraints based on operating condition matching, such as the energy distribution corresponding to different operating phases must conform to the characteristics of start-stop, cruise or braking operating conditions.

[0122] The energy consumption curve corresponding to each candidate trajectory is integrated over time to obtain the total energy demand and staged energy distribution of each trajectory within the same time interval. Energy balance difference indicators, such as energy deviation rate and peak power difference, are calculated by comparing them pairwise or with the baseline trajectory. Based on this, a vehicle power boundary model is introduced to constrain and verify the power output at each time node in the trajectory, and to determine whether it meets the physical limitations such as the maximum output power, minimum stable power and power change rate of the engine / motor. At the same time, the energy distribution between drive output, regenerative braking and auxiliary load is verified in conjunction with the energy conservation relationship. When power exceeds the limit, energy is not closed, or does not conform to the dynamic characteristics, it is judged as physically unreasonable.

[0123] Step S205: In the set of strategies, select a target strategy based on task priority and time window index, and generate a predictive driving strategy.

[0124] The specific steps of step S205 are as follows:

[0125] Step S2051: Generate a task priority mapping table based on the port area's loading and unloading task scheduling information, operation type, and cargo type attributes.

[0126] In this embodiment, a port area task priority mapping table is generated by constructing a multi-factor weighted model based on operational characteristics and cargo flow parameters. Specifically, firstly, the loading and unloading task scheduling information for the current period is obtained from the port area scheduling platform, including operation start and end times, berth allocation, yard location, and task dependencies. Subsequently, based on the operation type, such as loading, unloading, transshipment, or short-haul transportation, and cargo attributes, such as particle size, density, humidity, and fragility, key factors affecting operational urgency and resource utilization are extracted. To quantify the relative importance of different tasks in terms of time and resources, a weighted evaluation mechanism is introduced, normalizing the operation timeliness factor, equipment utilization factor, and cargo sensitivity factor, and determining the weight of each factor through the analytic hierarchy process. Next, a task dependency graph structure is established on the time axis, with nodes representing operational units and edges representing resource conflicts or task order. The topological sorting algorithm is used to determine the priority sequence of task execution. Based on this, the comprehensive priority value of each operational unit is calculated, forming a task-priority mapping table. This mapping table uses the task number as an index to record the comprehensive priority of tasks in terms of time, space, and resources.

[0127] It should be noted that for each work unit, factors such as work timeliness (e.g., deadline urgency), equipment occupancy rate (e.g., intensity of use of critical equipment), and cargo sensitivity (e.g., fragility or environmental dependence) are first normalized to obtain dimensionless characteristic values. Then, based on the weight coefficients determined by the analytic hierarchy process (AHP), the factors are weighted and summed to form a basic priority value. Furthermore, priority adjustment terms are introduced for tasks with pre-existing constraints, taking into account task dependencies. For example, the weight of tasks on the critical path is increased, or the weight of tasks on non-critical paths is decreased. Finally, a comprehensive priority value reflecting the combined effect of multiple factors such as time urgency, resource competition, and operational risk is obtained.

[0128] Step S2052: According to the port area operation sequence and vehicle scheduling cycle, divide the execution period corresponding to the strategy into several time windows and form a multi-level time index structure.

[0129] In this embodiment, a multi-level time index structure for predictive driving strategies is established using a time-layered modeling method based on operation sequence and vehicle scheduling cycle. Specifically, firstly, based on the time nodes in the port operation schedule, the entire operation cycle is divided into four core stages: task preparation period, loading and unloading execution period, transfer connection period, and empty return period. The start and end times of each stage are extracted as the boundaries of the first-level time window. Subsequently, at the vehicle scheduling level, based on the task allocation frequency and energy replenishment cycle of heavy trucks, the first-level window is further refined into several second-level time windows, each corresponding to a continuous vehicle running segment. To achieve the time correlation of cross-layer strategies, synchronization identifiers are embedded in each window to mark the trigger points of task events, vehicle status, and energy scheduling. Next, a time-slice sliding algorithm is used to smooth the overlapping areas between different windows to ensure the continuity of strategy execution at window boundaries. Finally, a multi-level time index structure is constructed to associate and map each strategy execution interval with the port scheduling cycle, task priority, and vehicle dynamic status, so that each strategy node can be uniquely located in the time domain and matched with the corresponding operation stage.

[0130] It should be noted that a hierarchical index system is constructed based on multi-layer time windows. The first-level window (operation phase) is used as the global time master index, and the second-level window (vehicle operating interval) is used as the sub-index. A unique time identifier is assigned to each strategy node. Subsequently, the strategy execution interval is matched and bound with the corresponding time window. At the same time, task priority is introduced as a weight label, and vehicle dynamic status such as location, load, and energy level are introduced as status labels. The relationship between time window, task priority, vehicle status, and strategy node is established through a multi-dimensional index mapping table.

[0131] Step S2053: Based on the task priority mapping table and time window index structure, prioritize the selection of the strategy set. When different strategies overlap in time or task intervals, dynamically adjust the strategy selection order.

[0132] In this embodiment, a dynamic filtering and timing coordination mechanism for the strategy set is established by combining a task priority mapping table and a time window index structure, thereby achieving adaptive sorting and conflict resolution among multiple strategies. Specifically, firstly, the task priority mapping table generated in step S2051 is cross-matched with the multi-level time index constructed in step S2052, and a corresponding task weight and time identifier are assigned to each strategy node. Subsequently, the strategy set is initially sorted according to the priority value of the task to which the strategy belongs, forming a priority queue arranged according to task urgency. Based on this, the distribution of each strategy on the time axis is analyzed. The system utilizes a time overlap detection algorithm to identify execution conflicts between strategies within the same or adjacent time windows. When overlapping time intervals of strategies are detected, a dynamic adjustment module is invoked to reconstruct the strategy selection order. Specifically, by comparing the difference in task priorities and the matching degree of strategy energy, the strategy nodes to be retained and delayed are determined. If the priorities of two strategies are close, their resource utilization and execution duration are further calculated to determine the order of strategy execution. The adjusted strategy set maintains logical mutual exclusion and temporal coherence in both the task domain and the time domain, ensuring the coordination and consistency of strategy execution in high-concurrency scenarios in port operations.

[0133] It should be noted that for each strategy node, firstly, a corresponding task weight is assigned based on the comprehensive priority value of its assigned task in the task priority mapping table. Then, in combination with the execution stage corresponding to the strategy, the time window number and time interval range of the node are extracted from the multi-level time index structure to generate a unique time identifier. Subsequently, the task weight is bound to the time identifier to form a composite attribute label that includes priority weight and time location.

[0134] Step S2054: Use the selected strategy as a predictive driving strategy.

[0135] Step S3: Establish a heavy-duty truck cooperative control model based on the predictive driving strategy, perform feedforward optimization on the drive power distribution, and perform driving control on the heavy-duty truck based on the feedforward optimized heavy-duty truck cooperative control model.

[0136] like Figure 3 As shown, the specific steps of step S3 are as follows:

[0137] Step S301: Based on the predictive driving strategy, construct a heavy truck cooperative control model that includes a drive domain, a braking energy recovery domain, and an auxiliary electrical domain, and establish parameter interaction interfaces between each control domain.

[0138] In this embodiment, based on the predictive driving strategy generated in step S205, the heavy-duty truck vehicle control system is modeled in domains to construct a multi-domain collaborative control model that includes a drive domain, a braking energy recovery domain, and an auxiliary electrical domain. The collaborative control model adopts a hierarchical structure design, establishing a unified parameter interaction interface between each control domain, enabling different functional domains to operate collaboratively within the same energy management framework.

[0139] The drive domain, modeled around the vehicle's powertrain system, describes the torque output characteristics, energy conversion behavior, and response to vehicle traction demands of the engine and electric drive unit under different operating conditions. It receives target power demand information from the predictive driving strategy and, combined with current vehicle speed, load status, and transmission system operating status, generates corresponding drive power output parameters, providing basic input for vehicle energy allocation. The braking energy recovery domain describes the capture, transfer, and reuse of energy during vehicle deceleration or braking. Based on vehicle braking demand information, this control domain comprehensively evaluates the electric motor's braking recovery capability and the energy storage unit's status, determines executable energy recovery strategies, and feeds back the recovered energy status to the vehicle's energy management layer to participate in subsequent energy allocation decisions. The auxiliary electrical domain describes the power consumption behavior of onboard auxiliary energy-consuming equipment, including but not limited to non-traction loads such as air compressor systems, hydraulic systems, and thermal management devices. Based on vehicle operating status and current energy availability, the auxiliary electrical domain coordinates and controls the energy consumption levels of each auxiliary device, ensuring that overall energy efficiency is improved in conjunction with the drive and braking energy recovery domains without affecting vehicle safety and basic functions.

[0140] A timing alignment mechanism is introduced into the collaborative control model to coordinate the control cycles of different control domains. By aligning the sampling frequencies and compensating for control response delays, each control domain can complete state updates within the same control reference time frame. Through the above-mentioned domain modeling, parameter interaction, and timing coordination process, a heavy-duty truck collaborative control model with energy flow logic consistency and parameter dynamic coupling characteristics is constructed. Under the guidance of predictive driving strategies, this model can achieve collaborative control between the drive domain, braking energy recovery domain, and auxiliary electrical domain, providing a reliable model foundation for vehicle energy management and operational efficiency optimization.

[0141] Step S302: Based on the vehicle's power parameters and energy demand curve, a power allocation prediction matrix is ​​formed by time stepping. The power allocation prediction matrix defines the power demand ratio of each control domain at different time nodes.

[0142] In this embodiment, a power prediction matrix describing the dynamic energy distribution relationship between the drive domain, braking energy recovery domain, and auxiliary electrical domain is established using a time-stepped power allocation modeling method. Specifically, firstly, the energy flow parameters of the cooperative control model in step S301 are mapped to the vehicle's power parameters, and key variables representing the vehicle's output capacity, transmission efficiency, braking feedback rate, and additional load characteristics are extracted. Subsequently, using the energy demand curve obtained in step S105 as the time series input, the entire driving cycle is discretized into several time steps, each time step corresponding to an independent energy scheduling node for the vehicle. At each time node, through... The energy balance constraint equations are used to jointly solve for the output power of the driving domain, the regenerated power of the braking domain, and the power consumed by the auxiliary electrical domain, forming a local power allocation vector. To eliminate abrupt errors caused by time-domain discretization, a sliding time window method is used to weighted smooth the power ratio of adjacent time nodes to ensure the continuity of the power allocation trend. Then, a two-dimensional power allocation prediction matrix is ​​generated by stacking and standardizing the power vectors of continuous time steps. This matrix records the power demand ratio at different times by using the time step as the row index and the control domain as the column index. Finally, this matrix serves as the feedforward input of the cooperative control model.

[0143] It should be noted that within each time step, with energy demand as the constraint objective, an energy balance relationship is established between the output power of the drive domain, the regenerative power of the braking domain, and the power consumed by the auxiliary electrical domain. The total demand power is decomposed into a combination of drive supply, regenerative compensation, and auxiliary consumption. Constraints are constructed by combining vehicle power boundaries such as maximum drive power, regenerative efficiency, and auxiliary load range. By solving the power allocation equation or optimization model under these constraints, such as the minimum energy consumption or minimum power fluctuation objective, a power allocation result that satisfies energy conservation and conforms to the operating characteristics of each control domain is obtained, thus forming the local power allocation vector at the corresponding time node.

[0144] Step S303: Based on the power allocation prediction matrix, the driving power allocation is corrected in real time, and an optimized power trajectory is generated through a multi-constraint search algorithm.

[0145] The specific steps of step S303 are as follows:

[0146] Step S3031: Extract the set of dynamic constraint parameters for the current time period from the power allocation prediction matrix.

[0147] In this embodiment, a constraint extraction mechanism based on time localization is used to identify the dynamic constraint parameter set for the current driving period from the power allocation prediction matrix. Specifically, the power allocation prediction matrix generated in step S302 is first segmented according to the vehicle's real-time driving time index, and corresponding matrix sub-blocks related to the current time step and its neighboring windows are extracted. Subsequently, the power ratio change rate of each control domain is analyzed in this sub-block, and key dynamic features such as the drive domain output gradient, the braking domain energy recovery response delay, and the auxiliary electrical domain load fluctuation range are calculated. Based on these features, a set of constraint functions for power changes within the time period is established to... This describes the power adjustable boundary and energy distribution sensitivity of each control domain under the current operating condition. To improve the adaptability of the constraints, the stability of the power change trend is judged by the time-series difference analysis method. When a sudden change or fluctuation in energy distribution is detected that exceeds the set threshold (by statistically analyzing the power distribution change rate and response delay during the training and validation phases, taking the combination of the mean and standard deviation or setting the percentile as the threshold), the corresponding upper and lower limits of the constraints are automatically adjusted so that the constraint model can dynamically converge with the real-time changes in the vehicle state. Finally, the extracted dynamic constraint parameter set is structured and encapsulated to form a constraint description object bound to the time step index.

[0148] It should be noted that, within the current time step and its adjacent time windows, time-series difference analysis is performed on the power sequences of each control domain based on the power allocation prediction matrix: the output gradient is obtained by calculating the ratio of the power change in the drive domain to the time interval between adjacent time steps, which is used to characterize the rate of change of power output; the correlation analysis is performed on the time offset between the regenerative braking signal and the vehicle deceleration or braking force demand to extract the lag time of the regenerative response relative to the control command; at the same time, the maximum value, minimum value and variance of the auxiliary electrical domain power sequence are calculated within the sliding window to obtain its load fluctuation range.

[0149] Step S3032: Based on the dynamic constraint parameter set and the predictive driving strategy, establish a power correction reference model and generate a power benchmark sequence. The power correction reference model uses time stepping logic as the main line and performs differential mapping between the historical energy consumption trajectory and the current power prediction value.

[0150] In this embodiment, by introducing the differential mapping principle and time-step control logic, a power correction reference model based on a predictive driving strategy is constructed, and a power benchmark sequence for cooperative control is generated. Specifically, firstly, the dynamic constraint parameter set extracted in step S3031 is used as the boundary condition, and combined with the speed planning and energy allocation rules in the predictive driving strategy, the power adjustable range of each control domain in the current time period is defined. Subsequently, the power prediction value of the corresponding time step in the power allocation prediction matrix is ​​differentially mapped with the historical energy consumption trajectory. By calculating the energy change rate and prediction deviation, a time-series correction function reflecting the power evolution trend is established. During the differential mapping process, a sliding time window recursive logic is used to accumulate the energy difference of consecutive time steps, and a weight attenuation factor is introduced to highlight the dynamic response characteristics of the recent time period, so that the correction model can reflect the vehicle's operating status and operating condition changes in real time. Then, the corrected power value of each time step is normalized and allocated among the drive domain, braking energy recovery domain, and auxiliary electrical domain to generate a power benchmark sequence that meets the constraint conditions.

[0151] It should be noted that within each time step, the energy change rate at adjacent moments is first calculated based on the historical energy consumption trajectory, i.e., the increase or decrease in energy per unit time, and compared with the energy change corresponding to the current power prediction value to obtain the prediction deviation. Subsequently, the energy change rate is used as the trend term and the prediction deviation is used as the error correction term. A time-series correction function is constructed through a linear or weighted combination method to reflect both the direction of energy evolution and the degree of deviation. This function is updated recursively over the time series to ensure its consistency with changes in consecutive time steps, thereby forming a time-series correction function that can dynamically correct the power prediction value and characterize the power evolution trend.

[0152] The weight decay factor is constructed based on time distance or state similarity. The basic idea is that data closer to the current moment has a higher weight, while data from further back has a lower weight. Specifically, it is generated using exponential decay, linear decay, or piecewise decay methods. For example, a decreasing function is constructed with the time step interval as the independent variable, so that the weight gradually decreases as the time window moves forward. In terms of parameter settings, it is calibrated according to the response characteristics and the frequency of changes in operating conditions: if the port environment changes rapidly, a larger decay rate is set to enhance the impact of real-time data; if the operating conditions are relatively stable, the decay rate is reduced to retain historical trend information.

[0153] Step S3033: Under the power correction reference model, explore the power trajectory in multiple dimensions to generate several candidate power trajectories.

[0154] In this embodiment, a power trajectory generation method based on a multi-dimensional exploration mechanism is used to achieve parallel deduction of power allocation modes between different control domains. Specifically, firstly, based on the power correction reference model established in step S3032, a search space for the power trajectory is defined with the time step sequence as the main axis. This space consists of a three-dimensional state vector composed of the drive domain output, the braking energy recovery ratio, and the auxiliary electrical load power. Subsequently, a set of candidate power points is generated in the neighborhood of the power reference sequence using a multi-dimensional sampling algorithm, with each point corresponding to an energy allocation scheme. To ensure that the exploration process has global coverage and local convergence, a hierarchical progressive strategy is adopted to expand the power allocation disturbance step by step according to time, amplitude, and inter-domain coupling degree, thereby generating multiple power evolution paths under low deviation conditions. Next, a trajectory consistency detection method based on energy conservation constraints is used to verify the time continuity and constraint matching degree of each power trajectory, and trajectory samples that do not meet the dynamic boundary conditions are eliminated. The filtered trajectory set is reconstructed into several candidate power trajectories, and each trajectory records the power evolution relationship between different control domains with the time step as the index.

[0155] It should be noted that the dynamic boundary conditions are jointly determined by vehicle dynamics constraints, energy management constraints, and real-time operating status, including: the maximum / minimum output power and power change rate limit of the drive domain, the upper limit of recovery efficiency and response delay constraints of the braking energy recovery domain, the load power fluctuation range of the auxiliary electrical domain, and the energy conservation relationship that must be satisfied among the three. At the same time, these boundaries are not fixed values, but are adaptively adjusted according to the current time step conditions such as slope changes, load status, energy demand level, and system temperature. For example, the upper limit of drive power is increased under high load or uphill conditions, and the upper limit of energy recovery ratio is relaxed during the braking phase.

[0156] Step S3034: Perform consistency screening on the candidate power trajectories, select the target trajectory that conforms to the dynamic constraint parameter set of the current time period, and generate the optimized power trajectory.

[0157] In this embodiment, a power trajectory selection mechanism based on dynamic constraint matching and temporal consistency judgment is used to select a target trajectory that meets the current operating conditions from the candidate power trajectory set and generate an optimized power trajectory. Specifically, firstly, the multiple candidate power trajectories generated in step S3033 are gradually matched with the dynamic constraint parameter set extracted in step S3031, and constraint consistency detection is performed on the power distribution range, rate of change, and inter-domain energy transfer ratio of each trajectory. Subsequently, the deviation of the trajectory from the dynamic constraint boundary is calculated in each time step, and a temporal consistency evaluation index is established, including a power continuity index: reflecting the smoothness of power changes in adjacent time steps, such as power gradient and second-order rate of change, i.e., acceleration term, used to determine whether there are sudden changes or oscillations; a constraint deviation index: measuring the degree of deviation of the power value at each time step from the dynamic constraint boundary, such as the out-of-bounds magnitude or safety margin; and energy. Consistency index: assesses whether the energy distribution among drive, recovery, and auxiliary energy consumption in the time series satisfies the conservation relationship and proportional stability; Response lag index: characterizes the time-series matching degree between control commands and power output, used to identify control delays or asynchrony phenomena; Fluctuation stability index: based on the variance or fluctuation amplitude of the power sequence statistically analyzed through a sliding window, used to characterize the overall operational stability; When a trajectory node experiences energy allocation exceeding limits or a sudden change in power gradient, a local constraint re-detection algorithm is triggered to perform elastic correction or removal operations on the node, restoring trajectory smoothness; To achieve overall coordination of multi-domain power, a weighted aggregation method is used to integrate three indices: drive domain stability, braking and recovery balance, and auxiliary electrical load adjustability, generating a global consistency score matrix; Finally, the optimal trajectory is selected based on the ranking results of the score matrix, and its time-step power sequence is re-interpolated and smoothed to form an optimized power trajectory.

[0158] It should be noted that energy allocation exceeding the limit means that within a certain time step, the power allocation result of each control domain exceeds the corresponding dynamic constraint range or physical boundary condition. For example, the driving power exceeds the maximum output capacity, the braking regeneration power exceeds the upper limit that the regeneration system can withstand, the auxiliary load power exceeds the allowable fluctuation range, or the energy allocation ratio between the three does not meet the energy conservation and system coupling constraints. When any of the above situations occur, it is determined that the energy allocation exceeds the limit.

[0159] Within each time step of each candidate power trajectory, the power value ranges of the drive domain, regenerative braking domain, and auxiliary electrical domain are first extracted and matched with the corresponding upper and lower limits of dynamic constraints. Simultaneously, the rate of change of power difference between adjacent time steps is calculated to determine whether it meets the power change rate constraint. Furthermore, the energy transfer ratio between each control domain is calculated, such as the ratio between drive output and regenerative compensation, and compared with the energy conservation and allocation rules for consistency. Through joint detection of the above three dimensions, trajectory nodes with power exceeding the limit, changing too rapidly, or energy allocation imbalance are identified, thereby determining whether the trajectory meets the overall constraint consistency requirements.

[0160] Step S304: Input the optimized power trajectory into the heavy truck cooperative control model, reconstruct the control weight distribution and execution timing logic, and output control commands.

[0161] The specific steps of step S304 are as follows:

[0162] Step S3041: Input the optimized power trajectory into the heavy truck cooperative control model, and establish a corresponding time index mapping table within the heavy truck cooperative control model to identify the target power output node for each time period.

[0163] In this embodiment, the optimized power trajectory is logically connected with the multi-domain structure of the heavy truck collaborative control model through the time index mapping model, so as to realize the timing calibration of the power target and cross-domain synchronous control. Specifically, firstly, the optimized power trajectory generated in step S3034 is expanded by time step, and the power output sequence containing the drive domain, braking energy recovery domain, and auxiliary electrical domain is extracted and injected as input data stream into the power interface module of the cooperative control model. Subsequently, a time index mapping table is constructed within the model, with the time step as the main index and the power allocation node as the sub-index, recording the target power output parameters and their corresponding control domains for each time period. To achieve temporal consistency between different control domains, a mapping function is established between the time index and the control frame within the model, so that the time nodes of the power trajectory can be precisely aligned with the control cycle. Next, the power target is calibrated by the index mapping table, and the power output target is bound to the corresponding model calculation unit and execution port, thereby achieving global traceability of the energy allocation path in the temporal structure. To prevent index drift, a dynamic synchronization mechanism is introduced during the mapping update process. When changes in vehicle operating status cause sampling offset, the time index offset is automatically adjusted to maintain the consistency of the control mapping.

[0164] It should be noted that by uniformly calibrating the time step index in the optimized power trajectory with the control cycle of the cooperative control model, a time mapping function is constructed to convert discrete trajectory time nodes into corresponding control frame numbers according to the sampling period. Specifically, based on the fixed period of the control system, such as the control period Δt, the position of each trajectory time point in the control frame sequence is calculated, and non-integer period nodes are mapped to the nearest or weighted neighbor control frames through interpolation or alignment rules. At the same time, a time offset compensation parameter is introduced in the mapping process.

[0165] Step S3042: Based on the node distribution of the optimized power trajectory, dynamically adjust the control weight ratio between the drive domain, the braking energy recovery domain, and the auxiliary electrical domain to form a weight distribution matrix that evolves over time.

[0166] In this embodiment, a dynamic weight allocation mechanism based on power trajectory nodes is used to achieve time-adaptive adjustment of control weights among the drive domain, braking energy recovery domain, and auxiliary electrical domain. Specifically, firstly, the power node sequence in the time index mapping table in step S3041 is extracted as a time-driven variable, and the power demand gradient corresponding to each time node is used as the input signal for weight adjustment. Subsequently, based on the trend change of power allocation in the optimized power trajectory, a multi-domain energy interaction function is constructed to describe the coupling relationship between drive domain output, braking feedback, and auxiliary energy consumption. In each time step, the energy proportion of each control domain under the current power node is calculated through a weight adjustment algorithm, and the allocation ratio is dynamically corrected according to the power change rate between nodes, so that the weight distribution can reflect the synchronous evolution of energy flow direction and power demand intensity. To avoid control instability caused by abrupt changes, a moving average and time-domain smoothing technique is used to transitionally constrain the weights of adjacent nodes, thereby maintaining the continuity of weight changes. Finally, the weight proportions calculated in each time step are arranged in chronological order to generate a weight distribution matrix that evolves over time.

[0167] It should be noted that, based on the energy flow relationship between the drive domain, the braking energy recovery domain, and the auxiliary electrical domain, a multi-domain energy interaction function is constructed. The power output, recovery efficiency, and load demand of each control domain are used as input variables. The energy transfer and compensation relationship between different control domains is described by energy conservation constraints and coupling coefficients. Power demand gradient and state adjustment factor are introduced into the function to form a dynamic balance mapping between drive output and recovered energy and auxiliary consumption.

[0168] Within each time step, based on the total power demand of the current power node, and combining the multi-domain energy interaction function with the constraint parameters of each control domain, such as drive efficiency, regeneration efficiency, and auxiliary load demand, the candidate power allocation values ​​of each control domain are first calculated. Subsequently, the power values ​​of each domain are normalized to form a proportional relationship in the total power, and weight adjustment factors such as power demand gradient, energy regeneration priority, and load stability coefficient are introduced to correct the proportion, thereby obtaining the energy proportion of the drive domain, regeneration domain, and auxiliary electrical domain at the current time node, realizing the dynamic weight calculation of multi-domain power allocation.

[0169] Step S3043: Based on the weight distribution matrix and time index mapping table, construct the execution timing logic chain of the control instructions.

[0170] In this embodiment, an execution timing logic chain reflecting the multi-domain collaborative control process is constructed using a control logic modeling method based on timing constraints and weights. Specifically, firstly, the weight distribution matrix generated in step S3042 is matched with the time index mapping table established in step S3041, using the time step as the main index and the control domain weight as the sub-index to determine the execution priority and control triggering order corresponding to each time node. Subsequently, in the timing logic modeling stage, a time constraint function is established using power change rate, energy flow direction, and control response delay as core variables to describe the triggering dependency relationship between control commands in adjacent time steps. When the power of the driving domain suddenly increases or the energy input of the recovery domain changes beyond the balance threshold, the execution order and triggering interval of the commands are automatically adjusted through the constraint function to maintain the dynamic coordination of multi-domain energy flow. Next, a control node linked list structure is introduced, abstracting the control command of each time step into an event node, and establishing a front-to-back dependency mapping between nodes to make the control logic chain traceable and hierarchical. To ensure the time synchronization of the execution logic, synchronization anchors are further embedded in the control chain to identify key power interaction nodes and response delay compensation points.

[0171] It should be noted that the balance threshold refers to the tolerance limit used to determine whether the energy distribution among multiple control domains is in a coordinated state. It is preset or adaptively adjusted according to the dynamic characteristics of the system, and is usually determined based on key parameters such as the rate of change of the drive domain power, the magnitude of the change in the regenerative braking energy input, and the fluctuation range of the auxiliary load.

[0172] Step S3044: Parse the timing logic chain into an executable control instruction set, the instruction set including torque regulation, energy recovery triggering and power switching.

[0173] In this embodiment, the timing logic chain constructed in step S3043 is transformed into an executable control instruction set through instruction parsing and multi-domain mapping mechanisms, realizing coordinated control of the drive, braking energy recovery, and auxiliary electrical domains. Specifically, firstly, semantic parsing is performed on each control event node in the timing logic chain, converting the triggering order and response conditions between nodes into instruction execution templates, and determining the execution time and scope of each instruction based on the time index mapping table. Subsequently, based on the energy proportion of different control domains in the weight distribution matrix, three types of basic instruction sets are defined: torque adjustment instruction, energy recovery trigger instruction, and power switching instruction. Among them, the torque adjustment instruction is used for output power correction and power response scheduling in the drive domain, the energy recovery trigger instruction is used for energy capture and redistribution trigger control in the braking domain, and the power switching instruction is used for load migration and energy rebalancing between the auxiliary electrical domain and the main drive domain. During the instruction generation process, the node variables in the logic chain are bound to the vehicle control parameters through the instruction mapping function, and a synchronous execution flag is introduced to ensure the timing consistency of instructions in different control domains during the execution phase. Finally, after instruction set reorganization and priority arrangement, a complete control instruction set structure is formed.

[0174] It should be noted that, within each time step, the dominance of each control domain is determined based on the energy proportions of the drive domain, braking energy recovery domain, and auxiliary electrical domain in the weight distribution matrix: when the drive domain has the highest weight proportion, a drive control command with torque regulation as its core is generated to adjust the power output; when the braking energy recovery domain weight increases significantly, an energy recovery command is triggered to prioritize regenerative braking and energy recovery; when the load proportion of the auxiliary electrical domain changes or energy redistribution is required, a power switching command is generated to realize power migration between the main drive and auxiliary systems.

[0175] Step S305: Perform time-segmented driving control on the vehicle according to the control instructions of the heavy truck cooperative control model.

[0176] In this embodiment, a time-segmented control execution mechanism is used to apply the control command set generated by the heavy-duty truck collaborative control model to the vehicle operation process, realizing phased scheduling and closed-loop control of multi-domain power output. Specifically, firstly, the control command set formed in step S3044 is used as the input source, and the driving cycle is divided into several control sub-time periods according to the time index mapping table. Each sub-time period corresponds to a specific power allocation and energy flow configuration target. Subsequently, control commands for the drive domain, braking energy recovery domain, and auxiliary electrical domain are loaded in each time period, and cross-domain coordination is performed through the command scheduling module in the model, so that torque adjustment, energy recovery, and power switching commands take effect synchronously under a unified time sequence. To achieve dynamic response and optimal energy control, real-time feedback signals are introduced, including parameters such as vehicle speed, load change, braking torque, and battery status, to compare the state deviation after command execution. When the deviation exceeds a preset threshold, the local control module adjusts the control variables of the corresponding domain in real time according to the power correction model to ensure that the power allocation remains balanced and sequentially continuous in actual operation. Finally, by periodically updating the control command execution state and switching time periods, precise time-segmented control of the vehicle throughout the entire driving cycle is achieved.

[0177] It should be noted that the preset threshold is a threshold used to determine whether the deviation between the actual operating state and the target control command needs to be triggered for adjustment. It is set based on the vehicle dynamics characteristics and the response capability of the control system, and usually includes speed deviation threshold, power output deviation threshold, energy distribution error threshold and battery state fluctuation range, etc.

[0178] Example 2

[0179] Please see Figure 4 Another embodiment of the present invention provides: a predictive driving control system for intelligent heavy-duty trucks at bulk cargo terminals, comprising: a demand prediction module, a driving strategy generation module, and a driving control module;

[0180] The demand forecasting module is used to acquire environmental information and port scheduling data of the heavy truck's driving path, and calculate the energy demand of the heavy truck's future driving process based on the pre-trained forecasting model.

[0181] The driving strategy generation module is used to generate a predictive driving strategy based on the energy demand and vehicle power parameters.

[0182] The driving control module establishes a heavy-duty truck cooperative control model based on a predictive driving strategy, performs feedforward optimization of drive power distribution, and performs driving control of the heavy-duty truck based on the feedforward optimized heavy-duty truck cooperative control model.

[0183] In addition, the parts of the technical solutions provided in the embodiments of this application that are consistent with the implementation principles of the corresponding technical solutions in the prior art have not been described in detail, so as to avoid excessive elaboration.

[0184] The specific embodiments described above further illustrate the purpose, technical solution, and beneficial effects of the present invention. It should be understood that the above descriptions are merely specific embodiments of the present invention and are not intended to limit the invention. Any modifications, equivalent substitutions, or improvements made within the spirit and principles of the present invention should be included within the scope of protection of the present invention.

Claims

1. A predictive driving control method for intelligent heavy-duty trucks at bulk cargo terminals, characterized in that, include: Obtain environmental information and port area scheduling data for the heavy truck's driving route, and calculate the energy demand for the heavy truck's future driving process based on a pre-trained prediction model; Based on the energy demand and vehicle power parameters, a predictive driving strategy is generated, which includes speed distribution, shift timing and acceleration / deceleration mode. A heavy-duty truck cooperative control model is established based on a predictive driving strategy. The drive power distribution is optimized by feedforward, and the heavy-duty truck is driven based on the feedforward optimized heavy-duty truck cooperative control model.

2. The predictive driving control method for intelligent heavy-duty trucks at bulk cargo terminals as described in claim 1, characterized in that, The process of acquiring environmental information about the heavy-duty truck's travel route and port area scheduling data, and calculating the energy demand for the future heavy-duty truck's travel process based on a pre-trained prediction model, includes: The environmental information of the heavy truck's driving path is obtained, and edge preprocessing is performed on abnormal data. The environmental information includes road slope, ground adhesion coefficient, traffic density, dynamic position of loading and unloading machinery, and weather conditions. Acquire port area scheduling data and establish port area scheduling data timestamps, and synchronize and integrate them with port area scheduling data. The port area scheduling data includes loading and unloading task priorities, berth congestion status, and operation time windows. Spatial semantic modeling is performed on the preprocessed environmental information at the edge and the fused port area scheduling data to obtain a feature vector of the port area's operational status. The feature vector is used to reflect the future driving path of heavy trucks. Based on the pre-trained prediction model, the feature vector is input, and a multi-time-scale energy demand curve is generated by simulating a continuous sequence of operating conditions for future driving paths. The simulation is completed before the vehicle is actually driven. The energy demand prediction results are weighted and filtered based on model confidence and operating condition similarity. If the confidence is lower than the set threshold, local re-prediction is triggered, and the final energy demand prediction results of the heavy truck driving process are obtained.

3. The predictive driving control method for intelligent heavy-duty trucks at bulk cargo terminals as described in claim 2, characterized in that, The pre-trained prediction model, inputting the feature vector, generates multi-time-scale energy demand curves by simulating continuous operating condition sequences of future driving paths, including: The feature vector is divided into continuous time period nodes according to the port area operation rhythm, and a future path prediction sequence is generated in the form of time slices. The future path prediction sequence includes vehicle position assumptions, dynamic state assumptions of the operation area, and road resistance state evolution information. The predicted future path sequence is input into a pre-trained prediction model, and the energy demand of the starting section, cruising section, yard interaction section and loading and unloading section are simulated according to the vehicle operation stage. Based on the output of the prediction model, energy demand sets at different time scales are constructed according to a preset time window; The energy demand set is checked for consistency along the timeline, and nodes that do not conform to the sequential logic are removed and reconstructed to generate energy demand curves at multiple time scales.

4. The predictive driving control method for intelligent heavy-duty trucks at bulk cargo terminals as described in claim 3, characterized in that, Based on the energy demand and vehicle power parameters, a predictive driving strategy is generated, including: The energy demand curve is mapped and matched with the vehicle power parameters to form an input dataset including power response coefficient, inertia characteristics, transmission ratio range and energy recovery characteristics. The strategy hierarchy is divided according to the dual dimensions of time domain and control domain, and the path planning layer, energy allocation layer and execution instruction layer are set. Based on the input dataset and policy hierarchy, multiple candidate driving trajectories are generated. Each driving trajectory contains the corresponding speed distribution, gear shift sequence, and acceleration / deceleration node order information. The candidate driving trajectories are checked for energy consistency and feasibility. Trajectories that violate the port area operation rhythm or vehicle power boundary are eliminated, and the set of strategies that meet the constraints are retained. Within the set of strategies, a target strategy is selected based on task priority and time window index, and a predictive driving strategy is generated.

5. The predictive driving control method for intelligent heavy-duty trucks at bulk cargo terminals as described in claim 4, characterized in that, Based on the input dataset and policy hierarchy, multiple candidate driving trajectories are generated, including: The parameter projection values ​​in the input dataset are unified into a spatiotemporal coordinate system to generate a multidimensional state matrix. Based on the constraint rules of the path planning layer, energy allocation layer, and execution instruction layer in the strategy hierarchy, a trajectory generation constraint set is constructed. Based on the multidimensional state matrix and trajectory generation constraint set, a multimodal candidate trajectory group is generated in a time-step manner; The candidate trajectory group is calibrated for logical coherence according to the time series. If a logical break or node drift is detected, a local trajectory regeneration process is triggered until multiple candidate driving trajectories are formed.

6. The predictive driving control method for intelligent heavy-duty trucks at bulk cargo terminals as described in claim 5, characterized in that, Within the set of strategies, a target strategy is selected based on task priority and time window index, and a predictive driving strategy is generated, including: Based on the port area's loading and unloading task scheduling information, operation type, and cargo type attributes, a task priority mapping table is generated; Based on the port area's operational sequence and vehicle scheduling cycle, the execution period corresponding to the strategy is divided into multiple time windows, forming a multi-level time index structure. Based on the task priority mapping table and time window index structure, the strategy set is preferentially filtered, and the strategy selection order is dynamically adjusted when different strategies overlap in time or task intervals. The selected strategies will be used as predictive driving strategies.

7. The predictive driving control method for intelligent heavy-duty trucks at bulk cargo terminals as described in claim 6, characterized in that, The aforementioned heavy-duty truck cooperative control model, established based on a predictive driving strategy, performs feedforward optimization of drive power distribution, and, based on the feedforward optimized heavy-duty truck cooperative control model, performs driving control of the heavy-duty truck, including: Based on the aforementioned predictive driving strategy, a heavy-duty truck cooperative control model is constructed, which includes a drive domain, a braking energy recovery domain, and an auxiliary electrical domain, and a parameter interaction interface is established between each control domain. Based on vehicle power parameters and energy demand curves, a power allocation prediction matrix is ​​formed through a time stepping method. The power allocation prediction matrix defines the power demand ratio of each control domain at different time nodes. Based on the power allocation prediction matrix, the driving power allocation is corrected in real time, and an optimized power trajectory is generated through a multi-constraint search algorithm. The optimized power trajectory is input into the heavy-duty truck collaborative control model to reconstruct the control weight distribution and execution timing logic, and output control commands. The vehicle is driven in different time periods according to the control instructions of the heavy truck collaborative control model.

8. The predictive driving control method for intelligent heavy-duty trucks at bulk cargo terminals as described in claim 7, characterized in that, Based on the power allocation prediction matrix, the drive power allocation is corrected in real time, and an optimized power trajectory is generated through a multi-constraint search algorithm, including: Extract the set of dynamic constraint parameters for the current time period from the power allocation prediction matrix; Based on the dynamic constraint parameter set and the predictive driving strategy, a power correction reference model is established to generate a power benchmark sequence. The power correction reference model takes time step logic as the main line and performs differential mapping between historical energy consumption trajectory and current power prediction value. Under the power correction reference model, the power trajectory is explored in multiple dimensions to generate multiple candidate power trajectories; The candidate power trajectories are subjected to consistency screening to select the target trajectory that conforms to the dynamic constraint parameter set of the current time period, and an optimized power trajectory is generated.

9. The predictive driving control method for intelligent heavy-duty trucks at bulk cargo terminals as described in claim 8, characterized in that, The optimized power trajectory is input into the heavy-duty truck cooperative control model to reconstruct the control weight distribution and execution timing logic, and output control commands, including: The optimized power trajectory is input into the heavy-duty truck cooperative control model, and a corresponding time index mapping table is established inside the heavy-duty truck cooperative control model to identify the target power output node for each time period. Based on the node distribution of the optimized power trajectory, the control weight ratios among the drive domain, braking energy recovery domain, and auxiliary electrical domain are dynamically adjusted to form a weight distribution matrix that evolves over time. Based on the weight distribution matrix and time index mapping table, construct the execution timing logic chain of the control instructions; The timing logic chain is parsed into an executable set of control instructions, which includes torque regulation, energy recovery triggering, and power switching.

10. A predictive driving control system for intelligent heavy-duty trucks at bulk cargo terminals, used to implement the predictive driving control method for intelligent heavy-duty trucks at bulk cargo terminals as described in any one of claims 1-9, characterized in that, include: Demand forecasting module, driving strategy generation module, and driving control module; The demand forecasting module is used to acquire environmental information and port scheduling data of the heavy truck's driving path, and calculate the energy demand of the heavy truck's future driving process based on the pre-trained forecasting model. The driving strategy generation module is used to generate a predictive driving strategy based on the energy demand and vehicle power parameters. The driving control module establishes a heavy-duty truck cooperative control model based on a predictive driving strategy, performs feedforward optimization of drive power distribution, and performs driving control of the heavy-duty truck based on the feedforward optimized heavy-duty truck cooperative control model.

Images