Method and device for assisted driving control of sanitation vehicles based on historical driving data

Abstract

This invention relates to the field of vehicle control technology and discloses a method and device for assisted driving control of sanitation vehicles based on historical driving data. The method involves collecting historical driving and operational status data to construct a spatiotemporal operational knowledge graph where nodes include historical expected dust levels and node knowledge confidence. The method matches the current node in the graph with the vehicle's current spatial position and longitudinal speed, and searches for a target node ahead. It verifies the spatiotemporal dual feedforward triggering conditions based on the historical expected dust levels and node knowledge confidence within the target node ahead. When the conditions are met, it generates the target deceleration and target pre-pressurization power. It then arbitrates the comprehensive intervention intent intensity to coordinate the control of the drive motor, braking system, and water pump driver. This invention achieves spatiotemporal feedforward to shorten operational response time, avoids human-machine conflict through intent arbitration, and optimizes overall vehicle energy consumption through energy coordination.

Description

Technical Field

[0001] This invention relates to the field of vehicle control technology, specifically to a method and device for assisted driving control of sanitation vehicles based on historical driving data. Background Technology

[0002] During road cleaning operations, sanitation vehicles need to adjust their driving speed and water spray pressure of the water pump based on the dust conditions on the road surface.

[0003] The operational status adjustment of conventional sanitation vehicles mainly relies on manual judgment by the driver or real-time feedback data from onboard dust sensors. This control method has a physical response delay. When the vehicle enters an area with high dust concentration and triggers the sensor action command, the water spraying system is limited by the signal transmission cycle and the physical time required for the water pump pipeline to establish the target water pressure. As a result, the water spraying system is unable to meet the required operational requirements in the initial stage of entering the area, leading to poor local dust suppression.

[0004] On the other hand, the chassis control system and the superstructure operating system of existing sanitation vehicles mostly adopt independent control architectures. When passing through sections of road with sudden changes in dust, the chassis braking system and the superstructure water pump system cannot achieve advance planning of control commands. When the vehicle is decelerating and the water pump is running at high power, the regenerative braking energy generated by the electric chassis drive system usually needs to be recharged back to the power battery before the water pump drive draws power from the power battery to operate independently. This energy flow method increases the charging and discharging cycle steps of the power battery, resulting in energy conversion losses and hindering the optimization of overall vehicle energy consumption. Summary of the Invention

[0005] To address the shortcomings of existing technologies, this invention provides a method and device for assisted driving control of sanitation vehicles based on historical driving data. This addresses the problem that existing sanitation vehicles rely on subjective judgment or delayed feedback for water spraying and braking operations, resulting in poor dust suppression and unreasonable energy consumption. Furthermore, the chassis and superstructure lack a feedforward collaborative control mechanism, making it impossible to establish water pump pressure and plan deceleration in advance before entering high-dust areas.

[0006] To address the above problems, the present invention provides the following technical solution: The first aspect of this invention provides a method for assisted driving control of sanitation vehicles based on historical driving data. The sanitation vehicle includes an electric chassis, a braking system, a water pump driver, and a drive motor, and includes the following steps: Historical driving and operational status data are collected to construct a spatiotemporal operational knowledge graph where nodes include historical expected dust levels and node knowledge confidence. The current node in the spatiotemporal operational knowledge graph is matched based on the vehicle's current spatial position and current longitudinal driving speed, and a target node ahead is searched. The spatiotemporal dual feedforward triggering condition is verified based on the historical expected dust levels and node knowledge confidence of the target node ahead. When the spatiotemporal dual feedforward triggering condition is met, a target deceleration and a target pre-pressurization power are generated. After generating the target deceleration and target pre-pressurization power, the comprehensive intervention intent intensity is obtained to arbitrate the target deceleration and target pre-pressurization power, thereby coordinating the control of the drive motor, braking system, and water pump driver.

[0007] Furthermore, the collection of historical driving and operational status data, and the construction of a spatiotemporal operational knowledge graph where nodes include historical expected dust levels and node knowledge confidence, specifically includes: Historical driving and operation status data, including historical spatial location, historical longitudinal driving speed, measured values ​​of environmental dust concentration, and actual water spray pressure, are collected, and valid sampling point data are selected from them; the target operation area is spatially discretized to generate multiple location nodes, which are matched with the valid sampling point data; feature statistics within each location node are extracted, and the historical expected dust level and node knowledge confidence of each location node are calculated by fusion.

[0008] Furthermore, the step of filtering out valid sampling point data specifically includes: The positioning quality factor of each sampling point in the historical driving and operation status data is extracted. When the positioning quality factor is lower than the positioning validity threshold, the corresponding data is removed. The retained data is then subjected to boundary verification based on the historical longitudinal driving speed and the actual water spray pressure to obtain the effective sampling point data. The matching with the effective sampling point data specifically includes: defining a search domain for the effective sampling point data to extract candidate location nodes; calculating the matching cost value based on the spatial position deviation and heading angle deviation between the effective sampling point data and the candidate location nodes, and extracting the candidate location node that produces the minimum matching cost value as the matching target.

[0009] Furthermore, the step of matching the current node in the spatiotemporal operation knowledge graph based on the vehicle's current spatial position and current longitudinal speed, and searching for the target node ahead, specifically includes: The current coordinates of the vehicle are obtained, and the node that is closest to the vehicle and meets the spatial matching threshold is selected as the current node. The dynamic forward search distance is calculated by combining the current longitudinal driving speed and the system response delay, so as to search for the target node ahead from the current node along the road topology.

[0010] Furthermore, the verification of the spatiotemporal dual feedforward triggering condition specifically includes: calculating the dynamic target arrival time based on the path travel characteristics from the current node to the target node ahead; and verifying the spatiotemporal dual feedforward triggering condition based on the dynamic target arrival time, the minimum safe braking distance, the historical expected dust level, and the node knowledge confidence.

[0011] Furthermore, the spatiotemporal dual feedforward triggering condition is verified based on the dynamic target arrival time, minimum safe braking distance, historical expected dust levels, and node knowledge confidence, specifically including: The node knowledge confidence level and the historical expected dust level are determined to be no less than the confidence level judgment threshold and the preset minimum action level, respectively; the minimum safe braking distance is calculated by combining the current longitudinal driving speed and the maximum mechanical deceleration of the electric chassis; when the arrival time of the dynamic target is determined to match the preset system feedforward preparation time, and the actual path distance to the target node ahead is greater than the minimum safe braking distance, the spatiotemporal dual feedforward triggering condition is determined to be satisfied.

[0012] Furthermore, the sanitation vehicle also includes a monitoring module, a sensing module, and a centrifugal water pump; the generation of the target deceleration and target pre-pressurization power specifically includes: The monitoring module collects the driver's line-of-sight deviation time percentage and driver abnormal load index; obtains the current actual water spray pressure to calculate the normalized water pump pressure, and constructs a line-of-sight deviation compensation coefficient based on the normalized water pump pressure to weight and compensate for the line-of-sight deviation time percentage, and calculates and corrects the driving load index in combination with the driver abnormal load index; integrates the data from the sensing module to obtain the environmental target state, and constructs a control state vector in combination with the corrected driving load index, thereby generating the target deceleration and the target pre-pressurization power.

[0013] Furthermore, the sensing module includes a camera and a millimeter-wave radar; fusing the data from the sensing module to obtain the environmental target state specifically includes: The visual quality factor of the camera and the radar health factor of the millimeter-wave radar are extracted; the visual quality factor is negatively compensated using the normalized water pump pressure to calculate the initial visual weight, and the radar health factor is used as the initial radar weight; the initial visual weight and the initial radar weight are subjected to amplitude limiting and normalization processing to obtain the visual fusion weight and the radar fusion weight, respectively; the target state vector output by the camera and the target state vector output by the millimeter-wave radar are weighted and updated using the visual fusion weight and the radar fusion weight to obtain the environmental target state.

[0014] Furthermore, a control state vector is constructed by combining the modified driving load index, thereby generating the target deceleration and the target pre-pressurization power, specifically including: The control state vector is substituted into a preset control strategy model to generate candidate deceleration and candidate pre-pressurization power; combined with the current longitudinal driving speed and water pump operating status, the absolute upper limit of the maximum deceleration and water pump pre-pressurization power is determined; the candidate deceleration and the candidate pre-pressurization power are respectively subjected to boundary limiting processing to output the target deceleration and the target pre-pressurization power.

[0015] A second aspect of the present invention provides a sanitation vehicle assisted driving control device based on historical driving data, for communicating with the braking system, water pump driver, and drive motor of the sanitation vehicle, and for communicating with the corresponding power battery, sensing module, and monitoring module according to execution logic, including a communication interface, a processor, and a memory, wherein the memory is used to store a computer program, and the processor is used to execute the computer program to implement the method described in any of the first aspects above.

[0016] Furthermore, the sanitation vehicle also includes a monitoring module and a power battery; the arbitration of the target deceleration and the target pre-pressurization power based on the acquisition of the comprehensive intervention intent intensity specifically includes: The intensity of the overall intervention intent is calculated based on the accelerator pedal opening, brake pedal travel, and steering wheel torque obtained by the monitoring module. When the intensity of the overall intervention intent is between the slight intervention threshold and the clear intervention threshold and is determined to be an uncertain intervention state, the current control arbitration coefficient is calculated to perform proportional attenuation on the target deceleration. When the intensity of the overall intervention intent is detected to drop below the slight intervention threshold, the current control arbitration coefficient is controlled to smoothly increase to restore automatic control authority. During the process of coordinating the control of the drive motor, braking system, and water pump driver according to the arbitration result, the regenerative braking power generated by the drive motor is guided to be preferentially supplied to the water pump driver, and the power supply and demand difference is dynamically compensated by the power battery.

[0017] Furthermore, the fusion calculation to obtain the historical expected dust level includes: calculating the theoretical average dust level by combining the disturbance parameters corresponding to the road surface material code and the average longitudinal driving speed; allocating weights according to the sensor calibration accuracy, weighting and fusing the measured average environmental dust concentration with the theoretical average, and using a stepped threshold comparison method to map and generate a discrete historical expected dust level.

[0018] Furthermore, the process of obtaining the node knowledge confidence level through fusion calculation is based on a calculation model that integrates the sample size adjustment coefficient, the optimal positioning status code, and the positioning quality factor of each sample. It also introduces the sample variance of environmental dust concentration and longitudinal driving speed as an exponential penalty term for weighted normalization and summation, so as to output a quantitative representation of the comprehensive confidence level after multi-dimensional data fusion.

[0019] Furthermore, the calculation of the dynamic target arrival time includes: dividing the actual path distance from the current point to the target node ahead into multiple discrete distance segments, calculating and limiting the predicted speed of each segment based on vehicle speed and acceleration; calculating the travel time of a single segment segment by segment and summing them up to output the dynamic target arrival time.

[0020] Furthermore, when coordinating the braking system, the current water load is calculated by analyzing the water tank level sensor signal, and the vehicle's curb weight is superimposed to obtain the vehicle's dynamic mass; the road slope angle extracted by the inertial measurement unit is combined to calculate the slope and rolling resistance; then the target deceleration is dynamically calculated and the above resistance is compensated, and the target total braking force is output.

[0021] Furthermore, the target total braking force is allocated to the drive motor to generate regenerative braking power, and the upper limit of power generation is limited when the charging power of the power battery is exceeded. After judging the vehicle speed conditions, the limited power is converted into actual regenerative braking force and compared with the target total braking force. When there is a difference, it is converted into hydraulic braking pressure control braking system to establish friction braking force to fill the gap.

[0022] Furthermore, the system also includes a safety monitoring and degradation mechanism: when the confidence level of the dynamic node calculated based on the real-time spatial deviation and effective positioning error is lower than the safety threshold, it switches to closed-loop water circuit control; when the high-voltage bus voltage or battery charging and discharging capacity is limited, or when the chassis active safety anti-lock braking intervention is triggered, the feedforward collaborative control process is suspended and the water pump pre-pressurization is canceled, and the hydraulic system undertakes the chassis braking requirements.

[0023] The assisted driving control method and device for sanitation vehicles based on historical driving data provided by this invention have the following beneficial effects: This invention constructs a spatiotemporal operational knowledge graph that includes historical expected dust levels and node knowledge confidence. It establishes a spatiotemporal dual feedforward triggering mechanism by combining spatial location and vehicle speed, enabling the vehicle to generate target deceleration and target pre-pressurization power in advance before entering high-dust road sections. At the same time, it introduces comprehensive intervention intent intensity for arbitration before executing control commands, shortening the response time difference between the superstructure water pump pressurization and chassis deceleration, and improving the timeliness of dust suppression. Furthermore, through the arbitration mechanism, it coordinates the control of the drive motor, braking system, and water pump driver, avoiding direct conflicts between system feedforward control commands and manual operation, and ensuring the smoothness and stability of driving during operation.

[0024] This invention establishes an energy coordination control mechanism between the electric chassis drive system and the superstructure operation system; during the vehicle's feedforward deceleration phase, the regenerative braking power generated by the guide drive motor is preferentially supplied to the water pump driver to maintain water pressure, and the power supply and demand difference is dynamically compensated by the power battery; this reduces the energy loss generated by the regenerative braking energy through the battery charge and discharge cycle, and improves the vehicle's overall energy consumption.

[0025] This invention optimizes the authority transition strategy between automatic control and manual operation. The system uses the accelerator pedal opening, brake pedal travel and steering wheel torque to quantify the driver intervention state. When the intervention state is determined to be uncertain, the target deceleration is proportionally reduced. After the intervention intensity decreases, the arbitration coefficient is smoothly increased to restore the automatic control authority. This reduces vehicle jerking and water pressure fluctuations caused by the switch of control authority, thereby improving the safety and stability of vehicle operation. Attached Figure Description

[0026] Figure 1 This is a diagram of the chassis and superstructure collaborative system architecture according to an embodiment of the present invention; Figure 2 This is a flowchart of the chassis-superstructure collaborative control method based on historical driving data according to an embodiment of the present invention. Figure 3 This is a flowchart illustrating the calculation of expected dust levels and confidence levels in an embodiment of the present invention. Figure 4 This is a flowchart of target node spatial matching and arrival time prediction according to an embodiment of the present invention; Figure 5 This is a logic diagram of spatiotemporal dual feedforward triggering boundary verification according to an embodiment of the present invention; Figure 6 This is a logic diagram for dynamic adjustment of environmental perception fusion weights in an embodiment of the present invention. Figure 7 This is a low-pass decoupling control logic diagram for energy source coupling and control flow according to an embodiment of the present invention. Figure 8 The diagram shows the dynamic response timing curves of the chassis and superstructure coordinated control in an embodiment of the present invention; wherein, (a) is the chassis motion state coordinated feedforward response curve, (b) is the spatiotemporal knowledge graph dust prediction effect curve, and (c) is the superstructure water pressure and chassis perception weight coupling mechanism curve.

[0027] Among them, 10. Electric chassis; 11. Power battery; 12. Drive motor; 13. Braking system; 14. Traction inverter; 20. Superstructure system; 21. Water tank; 22. Centrifugal water pump; 23. Water pump driver; 30. Sensing module; 31. Camera; 32. Millimeter-wave radar; 33. Positioning module; 34. Dust sensor; 40. Monitoring module; 51. Vehicle controller; 52. Edge computing platform; 60. High-voltage energy distribution module. Detailed Implementation

[0028] The technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0029] See Figure 1 The present invention provides a chassis-superstructure collaborative assisted driving system for low-speed operation scenarios of electric sanitation vehicles based on historical driving data, which may include: electric chassis 10, superstructure system 20, sensing module 30, monitoring module 40, vehicle controller 51, edge computing platform 52 and high-voltage energy distribution module 60.

[0030] The electric chassis 10 includes a power battery 11, a drive motor 12, a braking system 13, and a traction inverter 14. The braking system 13 is a basic friction braking mechanism and includes an electro-hydraulic brake actuator or a brake-by-wire actuator capable of receiving braking commands from the vehicle controller 51.

[0031] The superstructure system 20 includes a water tank 21, a centrifugal water pump 22, and a water pump driver 23.

[0032] The high-voltage energy distribution module 60 is connected to the power battery 11, traction inverter 14 and water pump driver 23 to form a shared high-voltage bus architecture. It coordinates and manages the connection status, power boundary and bus voltage of each high-voltage branch according to the control signal of the vehicle controller 51, so as to realize the energy distribution between the electric chassis drive system and the superstructure water pump system.

[0033] The sensing module 30 is located on the exterior of the vehicle body. The sensing module 30 includes a camera 31, a millimeter-wave radar 32, a positioning module 33, and a dust sensor 34 for collecting ambient dust concentration. The positioning module 33 integrates a global positioning receiver and an inertial measurement unit.

[0034] The monitoring module 40 is installed in the cockpit to collect the driver's facial feature parameters as well as the steering wheel angle and pedal displacement operation feature data.

[0035] The edge computing platform 52 receives image data, radar point clouds and position coordinates, dust concentration, positioning quality factor and heading angle information collected by the sensing module 30. The edge computing platform 52 performs target state fusion calculation and historical driving data comparison processing.

[0036] The edge computing platform 52 communicates with the vehicle controller 51 and receives driver status data uploaded by the monitoring module 40 via the communication bus, or obtains the driver status data by forwarding it through the vehicle controller 51. The vehicle controller 51 communicates with the electric chassis 10, the superstructure system 20, the monitoring module 40, and the high-voltage energy distribution module 60. Data transmission paths are established between the various hardware components within the system via the communication bus. The communication bus adopts a structure in which the controller area network CAN bus or CAN FD bus and the vehicle Ethernet bus are configured in parallel.

[0037] During operation, the edge computing platform 52 generates deceleration requests and water pump power requests based on real-time perception results and historical operation knowledge graphs; the vehicle controller 51, in conjunction with the vehicle's underlying state, sends the corresponding instructions to the traction inverter 14, braking system 13, and water pump driver 23, respectively.

[0038] When the vehicle decelerates, the drive motor 12 operates in generator mode to produce regenerative braking energy. The vehicle controller 51 obtains the voltage status of the high-voltage bus and the real-time power request of the centrifugal water pump 22, and dispatches the regenerative braking energy to the water pump driver 23 and the power battery 11 through the high-voltage energy distribution module 60.

[0039] The edge computing platform 52 and vehicle controller 51, and other control devices in this embodiment of the invention, may specifically include a communication interface, a processor, and a memory in their internal hardware structure. The memory stores computer programs, and the processor executes the computer programs to implement the assisted driving control method for sanitation vehicles based on historical driving data according to this invention.

[0040] See Figure 2 This invention provides a chassis-superstructure collaborative assisted driving control method for low-speed operation scenarios of electric sanitation vehicles based on historical driving data, comprising the following steps: S100 collects historical driving and operation status data including vehicle spatial location, driving speed, dust concentration and water spraying parameters. Based on this historical data, a spatiotemporal operation knowledge graph containing historical expected dust levels and knowledge confidence is constructed through spatial discretization and feature extraction. The spatiotemporal operation knowledge graph is then stored on the edge computing platform 52.

[0041] S200 obtains the vehicle's current driving speed, spatial position, and heading angle information through the perception module 30 and the chassis communication bus to match the vehicle's current node in the spatiotemporal operation knowledge graph; searches for target nodes ahead based on the preset look-ahead distance and predicts the time to reach the target node; calculates the minimum safe braking distance in combination with the current vehicle speed; and performs dual feed-forward trigger condition verification based on time and space dimensions.

[0042] S300, when the feedforward triggering condition is met, evaluates the driver's load status through the monitoring module 40; dynamically adjusts the fusion weight of the sensing module 30 according to the water spray pressure of the superstructure system 20; generates candidate chassis target deceleration and superstructure water pump pre-pressurization power based on the adjusted system status, and performs boundary limiting and rate of change constraints by the safety supervisor software module deployed in the edge computing platform 52 or vehicle controller 51 to output the target control request.

[0043] S400 determines the intensity of the driver's intervention intention based on the driver's operation of the pedal and steering wheel, arbitrates the automatic control request and the driver input, and outputs the final deceleration request and water pump power request. The vehicle controller 51 calculates the target total braking force taking into account the road slope and rolling resistance based on the final deceleration request, and schedules the high-voltage energy distribution module 60 to make the centrifugal water pump 22 prioritize the absorption of the regenerative braking power generated by the drive motor 12.

[0044] The S500 continuously monitors the knowledge graph confidence level, positioning quality, high-voltage bus status, upper equipment status, and chassis stability during the control cycle. When any status parameter falls below the preset safety threshold, it executes the exit or degradation of the collaborative feedforward control.

[0045] like Figure 2 As shown, the edge computing platform 52, through interaction with various communication nodes in the vehicle, transforms the signals captured by the underlying sensors into a dataset containing spatiotemporal features to construct a multi-dimensional historical operation trajectory sample set. The specific implementation process includes the following steps: S101, perform synchronous acquisition and structured encapsulation of multi-source heterogeneous operation status data. In this embodiment, when the vehicle performs routine driving or water spraying operations within the target operation area, the edge computing platform 52 extracts the corresponding underlying hardware data according to the set sampling frequency, and simultaneously acquires the spatial position coordinates and positioning quality feedback output by the perception module 30, the driving speed feedback output by the electric chassis 10, and the operation parameters fed back by the upper structure system 20. The edge computing platform 52 defines a single operation process as a historical trajectory and aggregates multiple historical trajectories to form a complete set of effective historical trajectories. Each historical trajectory contains a series of discrete sampling point data arranged in a time sequence. For any given moment in the trajectory, a single sampling point data is defined as a dataset containing multi-dimensional state features, specifically including: The system includes the absolute spatial coordinates of the vehicle obtained by the positioning module 33, the longitudinal speed of the vehicle analyzed by the bottom wheel speed sensor or drive motor 12, the measured value of the ambient dust concentration collected by the dust sensor 34, the actual water spray pressure of the centrifugal water pump 22 in the upper system 20, the current road surface material code obtained by high-precision map or visual recognition, and the positioning quality factor that matches the positioning signal at that moment.

[0046] The positioning quality factor is obtained by converting the positioning status code output by the GPS receiver through a preset mapping table, rather than directly judging the positioning quality based on the value of the status code. For example, RTK fixed solutions can be mapped to higher quality values, RTK floating-point solutions, differential solutions, and single-point solutions can be mapped to lower quality values ​​in sequence, and invalid positioning can be mapped to the lowest quality value. The mapped positioning quality factor is used for subsequent trajectory selection, node confidence calculation, and feedforward control safety verification.

[0047] S102, perform preliminary feature extraction based on vehicle driving disturbance mechanism. Preferably, to reduce the system response lag caused by conventional passive control logic relying solely on measured dust concentration results, the edge computing platform 52 uses the vehicle's historical longitudinal driving speed and absolute spatial position coordinates as preliminary features that cause environmental dust. In the feature extraction stage, the edge computing platform 52 not only stores the measured dust concentration values ​​but also establishes a mapping relationship between driving characteristics and the degree of dust disturbance. Specifically, the edge computing platform 52 constructs longitudinal driving speed as a feature representing the intensity of airflow wake and tire friction disturbance, and establishes a theoretical disturbance prediction model in conjunction with road surface material encoding. Since there is a positive correlation between airflow disturbance force and the square of driving speed, this theoretical disturbance prediction model can be expressed in the form of a nonlinear empirical function as follows: ; In the formula, Indicates based on the current longitudinal driving speed The theoretically predicted dust disturbance value; This is the excitation scaling factor; It is the basic environmental concentration constant. and The values ​​are all matched and retrieved from a pre-defined material property lookup table based on the corresponding road surface material code. Taking asphalt road surface characterized by high dust adhesion rate as an example, A higher parameter calibration value will be assigned. By extracting the driving speed as an antecedent factor, this invention can predict the trend of dust concentration change in the area based on the specific spatial location the vehicle is about to reach and its current driving speed in subsequent control, thereby establishing a priori reference for cooperative control based on historical driving trajectory data.

[0048] S103, perform boundary verification and anomaly removal of trajectory samples. The edge computing platform 52 performs validity screening on the historical trajectory set. Specifically, the edge computing platform 52 extracts the positioning quality factor from the data of each sampling point and compares it with a set positioning validity threshold. This positioning validity threshold is a normalized positioning quality threshold after status code mapping, not the satellite positioning status code itself. If the positioning quality factor of multiple consecutive sampling periods is lower than this threshold, it is determined that the absolute spatial position coordinates of the corresponding interval have a risk of coordinate drift, thereby removing the sampling point data of the corresponding time period. On this basis, the edge computing platform 52 further performs boundary verification on the status parameters and removes the failed sampling point data where the longitudinal travel speed is negative or the actual water spray pressure exceeds the rated operating range of the centrifugal water pump 22.

[0049] To ensure the completeness and temporal consistency of the algorithm logic, the edge computing platform 52 performs statistical calculations on the trajectories after removing abnormal data. If the proportion of the number of removed sampling points in a single historical trajectory exceeds a set removal ratio threshold (e.g., set to 30%), the entire trajectory is discarded to prevent data gaps from causing errors in the extraction of map node features. Only data retained after processing by the above verification mechanism can be used as valid samples input into the map construction stage.

[0050] The edge computing platform 52 integrates previously collected continuous time-series data into a fixed geographic reference frame by executing a spatial mapping algorithm. The specific implementation process includes the following steps: S104, Spatial Discretization and Node Generation of the Target Operation Area. In this embodiment, the edge computing platform 52 discretizes the space of the target operation area into several location nodes, thereby constructing the spatial topological skeleton of the spatiotemporal operation knowledge graph. During the above process, the edge computing platform 52 performs discretization sampling along the centerline of the operation road at a set spatial step size, based on high-precision map data or pre-collected benchmark road networks. Each generated location node is assigned a unique node number, and the edge computing platform 52 records its inherent static planar coordinates and the standard road heading angle of its location. The spatially connected location nodes constitute the reference targets for subsequent aggregation of multiple batches of historical trajectory data.

[0051] S105, Calculation of the spatial feature matching cost between valid sampling point data and location nodes. In actual operation, vehicles are often affected by lane keeping deviations or road obstacle avoidance maneuvers, making it difficult for valid sampling point data in a single historical trajectory to completely overlap with standard location nodes in spatial coordinates. Based on this, the edge computing platform 52 determines the matching index from valid sampling point data to each location node by calculating the weighted sum of position Euclidean deviation and heading angle deviation. For each valid sampling point data retained in the previous stage, the edge computing platform 52 delineates a search domain with a set radius around its spatial coordinate system, extracts several location nodes within this domain as candidate objects, and calculates the corresponding matching cost: ; In the formula, Indicates the data of the currently valid sampling point and the number of sampling points. The matching cost between candidate location nodes; and These represent the horizontal and vertical planar coordinate components obtained by transforming the current valid sample point data from absolute spatial position coordinates, respectively; and Indicates the first The inherent horizontal and vertical planar coordinate components of each candidate location node; This represents the actual heading angle of the vehicle, calculated from the absolute spatial coordinates of adjacent valid sampling points. Indicates the first Standard road heading angles for each candidate location node; This represents the distance deviation weighting coefficient; This represents the heading angle deviation weighting coefficient. Due to the dimensional difference between the Euclidean distance and the cosine penalty term, the distance term can be pre-normalized or assigned a weight based on the road width tolerance. and Different orders of magnitude of calibration values ​​are used to reasonably allocate the contribution of distance and orientation deviations to the matching results. The cost function considers both spatial position deviation and heading angle deviation.

[0052] S106, Matching index confirmation and outlier removal based on minimum cost value. After calculating the cost of all candidate objects within a single search domain, the edge computing platform 52 extracts the minimum matching cost. The location node is used as the matching target for the valid sampled point data. Preferably, to avoid invalid data being forcibly matched to regular road nodes, the edge computing platform 52 sets a maximum allowable matching threshold. The value of this maximum allowable matching threshold is usually set with reference to the half-width range of the lane boundary of the working road and the maximum drift error of the underlying positioning hardware. If the minimum matching cost calculated above exceeds the maximum allowable matching threshold, the edge computing platform 52 determines that the valid sampled point data is in an abnormal branch road or a non-working isolation zone, and terminates the mapping operation to remove it. If the minimum matching cost is within the acceptable range, the edge computing platform 52 updates the matching index of the valid sampled point data to the node number of the optimal location node.

[0053] With the establishment of the mapping index, the edge computing platform 52 aggregates the feature parameters such as longitudinal driving speed, measured value of environmental dust concentration and actual water spray pressure contained in the effective sampling point data into the corresponding location node, completing the transfer of discrete trajectory features to fixed location nodes.

[0054] See Figure 3 The edge computing platform 52 performs quantitative evaluation of the data mapped to each location node by executing statistical algorithms, thereby determining the operational characteristics of the corresponding node and its data reliability. The specific implementation process includes the following steps: S107, aggregate historical feature data of location nodes and extract statistics. In this embodiment, given the execution of previous spatial matching processing, valid sample point data from multiple historical job batches are often aggregated within a single location node. The edge computing platform 52 aggregates data to the... Statistical feature extraction is performed on the valid sampling point data within each location node. Specifically, the edge computing platform 52 counts the total number of valid sampling point data samples contained within that location node. If extracted If the value is zero, the edge computing platform 52 marks it as an empty node and skips subsequent computation logic. Under normal operating conditions where the concentration is greater than zero, the average measured value of environmental dust concentration is further calculated by the edge computing platform 52. Theoretical average value of dust disturbance Average longitudinal driving speed and the sample variance of measured environmental dust concentration Sample variance of longitudinal driving speed The extracted statistics collectively constitute the data foundation characterizing the historical operational status of this location node.

[0055] S108, Calculating the expected dust level based on multi-source data fusion. Addressing the issue that measured environmental dust concentrations are often delayed due to sensor response time limitations, and that theoretical dust disturbance values ​​are easily influenced by external wind direction or humidity, the edge computing platform 52 employs a weighted fusion mechanism to calculate the historical expected dust concentration for this location node. In specific implementation, the system averages the aforementioned measured environmental dust concentration values. =The average value of theoretically disturbed dust Assign corresponding weight coefficients to the measured data and theoretical predictions, respectively, and calculate the weighted sum of the two to determine the weighted sum of the predicted values. Historical expected dust concentration at each location node The sum of the measured data weighting coefficient and the theoretical prediction weighting coefficient is constrained to be 1.

[0056] Preferably, the specific values ​​of these two weighting coefficients are proportionally allocated based on the calibration accuracy and response delay characteristics of the measured sensor. For example, when selecting sensor hardware with a large response delay, the weighting coefficient of the measured data can be appropriately lowered and the weighting coefficient of the theoretical prediction can be increased accordingly to improve the weighting proportion of the theoretical prediction value. After completing the data fusion calculation, the edge computing platform 52, based on the system's preset environmental dust concentration classification standard, calculates the continuous historical expected dust concentration... The expected dust levels are mapped to discrete values ​​using a stepped threshold comparison method and encoded as comparable values ​​(e.g., 1 for light dust, 2 for moderate dust, and 3 for heavy dust). This classification of expected dust levels provides a quantitative benchmark for triggering subsequent feedforward control strategies.

[0057] S109, calculate the knowledge confidence of the nodes and perform validity screening. To ensure the reliability of the feedforward control logic, the edge computing platform 52 needs to further evaluate the data availability of the expected dust levels. In this step, the edge computing platform 52 integrates the total number of samples extracted from the aforementioned statistics, the sample variance of the measured environmental dust concentration, the sample variance of the longitudinal driving speed, and the positioning signal quality of the underlying hardware to construct a multi-dimensional confidence evaluation function. This function uses the natural index model to nonlinearly map the discrete variance values ​​and the continuous total number of samples to a normalized interval. For the first... Each location node has a confidence level in its basic node knowledge. The specific calculation formula for (i.e., the confidence level of historical knowledge stored in the spatiotemporal operation knowledge graph later) is constructed as follows: ; In the formula, This represents the sample variance of the measured values ​​of environmental dust concentration; The sample variance representing the longitudinal driving speed; This represents the total number of valid sample points; the output node knowledge confidence score is calculated. The range of values ​​for is normalized to be between 0 and 1; , , and These represent the evaluation weights for the corresponding evaluation dimensions, and the sum of each evaluation weight is always 1. This is the sample size adjustment coefficient, used to constrain the non-linear growth rate caused by an increase in the total number of samples. and These are the dust variance penalty coefficient and the vehicle speed variance penalty coefficient, respectively, so that the smaller the input sample variance, the closer the corresponding stability score is to 1; This indicates the position of the node within that location. Location quality factor for each sample; This represents the maximum value of the positioning quality factor. Among them, the parameter... , , The specific value is often determined offline based on the statistical distribution characteristics of historical data samples.

[0058] The node knowledge confidence is obtained through the aforementioned custom evaluation function. Then, the edge computing platform 52 compares the result with a preset confidence threshold. This confidence threshold is set according to the control precision requirements of the specific project, and its value is usually configured between 0.6 and 0.8. When the confidence level is greater than or equal to the threshold, it indicates that the historical feature data of the location node has sufficient statistical consistency. The edge computing platform 52 determines it as a valid node and allows the valid node and its internal parameters to participate in subsequent spatiotemporal feedforward control operations.

[0059] Conversely, when When the data level falls below the confidence threshold, the edge computing platform 52 determines that the data within the node at that location has abnormal fluctuations or insufficient sample size, marks it as an untrustworthy node, and removes or downgrades it.

[0060] like Figure 4 As shown, in this embodiment, the edge computing platform 52 acquires the vehicle's driving status in real time and uses a pre-built historical driving and operation map as data to match the current location and locate the target operation ahead. The specific implementation process includes the following steps: S201, Real-time Status Acquisition and Positioning Quality Verification. During operation, the edge computing platform 52 periodically acquires the vehicle's current real-time positioning coordinates and heading angle via the vehicle-mounted satellite positioning receiver and chassis bus. Longitudinal driving speed And the positioning quality factor. The edge computing platform 52 verifies the positioning quality factor in real time. If the factor is lower than a preset safety threshold (such as the normalized value of 0.9 for RTK fixed demapping), it indicates a risk of coordinate drift. The system will then suspend feedforward control and degrade to a control mode based on real-time sensors until the signal is restored.

[0061] S202, Current location node spatial alignment. When the positioning quality factor meets the requirements, the edge computing platform 52 converts the real-time positioning coordinates into the current coordinates in a local plane coordinate system with the same reference as the historical driving and operation map. After coordinate transformation, the edge computing platform 52 calls a spatial neighborhood search algorithm to find coordinates in the graph. The nearest location node is identified, and the Euclidean distance between them is calculated. If this distance exceeds a preset spatial matching threshold (e.g., 1.5 times the grid side length), the vehicle is determined not to be within the historical operation route, and feedforward control is not triggered; otherwise, it is locked as the current node index.

[0062] S203, Dynamic forward distance calculation and target coordinate deduction. To match the dust suppression control action with the disturbance area in front of the vehicle, the edge computing platform 52 deduces the target area ahead based on the real-time vehicle status. In this embodiment, the system configures the forward search distance as the longitudinal driving speed. Dynamically adaptive variables. Edge computing platform 52 integrated longitudinal driving speed. Calculate the dynamic forward search distance based on system response delay and baseline lead time. Its calculation model is as follows: ; In the formula, Indicates the dynamic forward search distance. The longitudinal travel speed, This indicates the overall response delay time of the feedforward control system, covering the time from the issuance of the command to the pump reaching the rated pressure value; This represents the basic forward search distance, used to ensure the minimum lead time for operations at extremely low vehicle speeds. Parameter Distance from base forward search Offline empirical calibration is performed based on the actuator response test results of a specific vehicle's water system. The dynamic forward search distance is then calculated. Then, the edge computing platform 52, starting from the current node index, accumulates the path length between nodes segment by segment along the road topology edges in the spatiotemporal operation knowledge graph that are consistent with the current heading. When the accumulated path length reaches... At that time, the corresponding position is determined as the coordinates of the target ahead. (or the target node ahead.)

[0063] S204, Target Node Addressing and Feature Information Extraction. Obtain the coordinates of the target ahead ( After that, the edge computing platform 52 invokes the graph search mechanism to match the coordinates of the target ahead ( The edge computing platform 52 identifies the nearest spatially located node and reuses the aforementioned logic to verify whether this distance meets the spatial matching threshold. If the match is valid, it is established as the index of the target node ahead. Subsequently, the edge computing platform 52 extracts the stored expected dust level and node knowledge confidence from this target node. If the match fails, no feedforward action is triggered for that forward location. After completing feature information extraction, the edge computing platform 52 calculates the current coordinates. To the target coordinates ahead ( The actual path distance between the current node and the target node is used. In this embodiment, the actual path distance is preferably the path distance accumulated along the topology of the graph from the current node to the target node ahead; in cases where the road curvature is small and there are no forks, the Euclidean straight-line distance between the two points can also be used for approximate calculation. This actual path distance constitutes the basic parameter for subsequent calculation of the feedforward command issuance time. The relevant coordinate transformation and spatial search can be implemented using conventional algorithms in this field.

[0064] In this embodiment, after obtaining the actual path distance matched in the preceding sequence, the edge computing platform 52 divides the continuous forward path into multiple discrete distance segments and calculates the estimated time for the vehicle to reach the target location based on the speed baseline constraint. The specific implementation process includes the following steps: S205, Discretization and segmentation of the forward path. The edge computing platform 52 divides the topological path from the current node to the forward target node into segments. There are discrete distance segments, and the length of each segment is denoted as . Resolution division It can be preset according to the feedforward control cycle and the vehicle operating speed.

[0065] S206, Discrete Segment Traffic Speed ​​Prediction and Bottom-Line Constraints. This is for the segmented and generated... For each discrete distance segment, the edge computing platform 52 calculates the reference traffic speed within that segment based on the current longitudinal travel speed and vehicle acceleration state using a uniformly accelerated linear motion model. Specifically, the formula for calculating the reference traffic speed within that segment is: ; ; In the formula, Indicates that the vehicle has left the area. Predicted end velocity for each discrete distance segment; This represents the function that takes the maximum value, used to prevent negative numbers from appearing within the square root when the vehicle decelerates to a stop; Indicates that the vehicle has entered the [number]. The initial velocity of each discrete distance segment; This indicates the actual acceleration of the vehicle at present; Indicates the first The length of each discrete distance segment; The average speed within this section is defined as the reference speed. To avoid computational anomalies caused by low speeds or stops, the edge computing platform 52 uses a preset minimum speed limit to protect the reference speed. In the computational logic, the system extracts the larger value between the reference speed and the minimum speed limit, and establishes it as the reference speed for the current section. The predicted segment speed for each discrete distance segment. The value of the minimum speed limit mentioned above is taken with reference to the idle crawl speed configuration of the electric chassis 10, for example, set to 0.5 km / h, so as to avoid the denominator of the calculation being zero.

[0066] S207, Target arrival time accumulation calculation. The edge computing platform 52 calculates the travel time based on the length of each segment and the predicted vehicle speed of the segment, and accumulates them to obtain the dynamic target arrival time. In this embodiment, the arrival time of the dynamic target... The specific calculation model is constructed as follows: ; In the formula, This indicates the estimated dynamic target arrival time. This represents the total number of discrete distance segments; Indicates the first The actual length of each discrete distance segment, for a regular segment For the tail section It equals the difference between the actual path distance and the total length of the regular segment; Indicates the first The reference traffic speed is obtained by extrapolating vehicle acceleration from discrete distance segments; This indicates the preset minimum speed limit. This represents the maximum value extraction function. The calculated dynamic target arrival time forms the data basis for subsequent calculations of the feedforward instruction issuance time.

[0067] See Figure 5 In this embodiment, the edge computing platform 52 constructs a joint constraint mechanism based on time and space dimensions for the low-speed operation of sanitation vehicles to prevent overshoot caused by premature issuance of control commands or insufficient spatial distance. The specific implementation process includes the following steps: S208, Verification of the Validity of the Forward Target Node Features. Before the final determination of the feedforward instruction, the edge computing platform 52 performs an availability screening on the internal data of the matched forward target node. The system compares the node knowledge confidence level stored in the node with the preset confidence level judgment threshold and determines whether its expected dust level reaches the minimum action level to trigger waterway dust suppression operations. When the node knowledge confidence level is lower than the confidence level judgment threshold, or the expected dust level is lower than the minimum action level, the feedforward control of the current cycle is terminated.

[0068] S209, Trigger Boundary Parameter Calculation. After completing the feature validity verification, the edge computing platform 52 calculates the critical parameters for triggering the operation in both the time and spatial dimensions. The system feedforward preparation time is determined by the water pump pressure build-up delay, braking response delay, high-pressure distribution delay, network transmission delay, and safety redundancy time. This time parameter characterizes the system time from when the chassis actuator receives the command to when it reaches the rated operating state. In the spatial dimension, considering that time estimation usually corresponds to a shorter spatial distance under low-speed conditions, relying solely on time judgment can easily lead to spray position deviation. The edge computing platform 52 combines the current longitudinal driving speed and the maximum mechanical deceleration of the electric chassis 10, and uses kinematic braking calculation logic based on mathematical formulas (its underlying principle is the same as the aforementioned kinematic model, both based on classical Newton's laws of motion) to calculate the minimum safe braking distance. In specific execution, the minimum safe braking distance is composed of the sum of the vehicle's constant speed travel distance during the system command response period and the deceleration and sliding distance under braking intervention. The specific calculation formula is as follows: ; In the formula, To achieve the minimum safe braking distance, The current longitudinal travel speed, This refers to the system command response period (i.e., the system feedforward preparation time). This is the maximum mechanical deceleration of the electric chassis 10 (take the absolute value).

[0069] S210, Spatiotemporal Dual-Trigger Logic Judgment. After determining the above boundary parameters, the edge computing platform 52 synchronously incorporates the previously calculated dynamic target arrival time and actual path distance into the evaluation conditions. When the dynamic target arrival time meets the feedforward preparation time window and the actual path distance is greater than the minimum safe braking distance, the feedforward control command is released. For the current decision cycle, its spatiotemporal dual-feedforward trigger judgment model is constructed as follows: ; In the formula, This indicates the spatiotemporal dual feedforward trigger flag for the output. A value of 1 indicates that the trigger condition is met, and a value of 0 indicates that the trigger condition is not met. This represents the confidence level of the basic node knowledge corresponding to the aforementioned target node, extracted from the spatiotemporal operation knowledge graph (i.e., the confidence level calculated by historical data fusion). ); This indicates the set confidence threshold. Indicates the expected dust level extracted; Indicates the preset minimum action level; Indicates the estimated arrival time of the dynamic target; This indicates the system feedforward preparation time calculated in this step; This indicates the set trigger time tolerance, which is related to... Together they constitute an effective time forecast window; This represents the calculated actual path distance to the target node ahead; This represents the calculated minimum safe braking distance.

[0070] Preferably, the above-mentioned trigger time tolerance The communication heartbeat cycle of the reference controller is set, for example, between 100 and 200 milliseconds, to accommodate network transmission fluctuations; confidence threshold. The value range is typically configured between 0.6 and 0.8. When the spatiotemporal dual feedforward trigger flag is determined based on the above judgment conditions... When the value is 1, the edge computing platform 52 determines that the triggering condition is met, and then sends the corresponding control command and the expected dust level to the chassis actuator.

[0071] In this embodiment, the edge computing platform 52 combines the vehicle's waterway operation status with the driver's operational behavior to construct an evaluation mechanism to quantify the current driver's workload level. The specific implementation process includes the following steps: S301, Multi-source driving behavior feature extraction. During sanitation operations, the edge computing platform 52 monitors the driver's operational status and physiological characteristics. The system acquires monitoring data of the driver's eyes and head through a facial perception camera in the cockpit. Within a set time window, it calculates the proportion of time the driver's gaze leaves the center of the road ahead as the gaze deviation time percentage; it calculates the proportion of time the eyelid closure exceeds a set closure threshold as a blinking anomaly indicator; and it counts the frequency of head deflection angles exceeding a set safe angle threshold as a head posture anomaly indicator. The blinking anomaly indicator and the head posture anomaly indicator together constitute the driver's abnormal load indicator in this embodiment. Combined with physiological characteristic monitoring, the system reads the steering column angle data through the electric chassis control bus, counts the number of times the steering wheel crosses zero in a preset small angle range within the above time window, and uses this as the normalized steering wheel fine-tuning frequency.

[0072] S302, Calculation of water mist interference compensation coefficient. Since the water mist generated by high-pressure spraying can obstruct the vehicle's side and rearview mirror visibility, it is normal for the driver to adjust their head and shift their gaze to observe the work boundary. To address this scenario, the edge computing platform 52 reads the current real-time water pump injection pressure and divides it by the pump's maximum rated pressure to calculate the normalized pump pressure. Based on the obtained pressure status, the edge computing platform 52 constructs a line-of-sight deviation compensation coefficient to reduce the weight of visual deviation features in the overall evaluation under water mist conditions.

[0073] S303, Construction and Calculation of the Corrected Driving Load Index. After extracting basic behavioral features and compensation coefficients, the edge computing platform 52, combined with abnormal driver load indicators, calculates the corrected driving load index using a preset mathematical formula. The calculation model for this corrected driving load index is constructed as follows: ; In the formula, This represents the corrected driving load index obtained from the solution; This indicates the percentage of time the line of sight was deviated from the target. and Together, they are used to reflect abnormal driver load indicators, among which, This indicates the extracted head posture anomaly indicators; This indicates the extracted abnormal blinking indicators; This represents the calculated normalized steering wheel tuning frequency. This represents the constructed line-of-sight deviation compensation coefficient, whose internal calculation relationship is as follows: ,in To obtain the normalized pump pressure, The set compensation adjustment constant and satisfying Preferably, The value is configured between 0.2 and 0.5; , , , The weight coefficients corresponding to each feature parameter are obtained offline through annotation of historically labeled payload datasets, and satisfy the following conditions: .

[0074] S304, Load Status Determination and Control Level Switching. After completing the index calculation, the edge computing platform 52 compares the corrected driving load index with the set load safety threshold. When the corrected driving load index exceeds the load safety threshold, the system determines that the driver is in a high-load state. In response to this state, the edge computing platform 52 issues a permission downgrade command to the underlying execution module via the control bus to adjust the safety boundary parameters of the automatic control. Preferably, this adjustment includes reducing the vehicle's maximum operating speed or increasing the feedforward warning distance threshold for obstacles ahead by the safety monitor module within the edge computing platform.

[0075] See Figure 6 In this embodiment, the edge computing platform 52 constructs a sensor degradation fusion mechanism based on feedback from the vehicle's actuators to address the water mist interference characteristics during sanitation vehicle operations. This mechanism dynamically adjusts the ratio of visual data to radar data to improve the stability of environmental perception results. The specific implementation process includes the following steps: S305, Sensor State Factor Extraction. During vehicle operation and operation, the edge computing platform 52 acquires basic operating indicators of the environmental perception sensors in real time. The system extracts the visual quality factor from the camera 31, which can be obtained by calculating the variance of the gray-level gradient of the current frame image or by calculating the image information entropy, and normalized to the range of 0 to 1, to characterize the clarity and contrast status of the current image; simultaneously, it extracts the radar health factor from the millimeter-wave radar 32, which can be obtained by analyzing the internal hardware self-test status bit of the radar controller and the average echo signal-to-noise ratio of the current period, and is also normalized to the range of 0 to 1, to reflect the radar echo quality and internal hardware status. The above two factors provide a basic data confidence reference for multi-sensor fusion.

[0076] S306, Initial weight calculation for perception fusion. Given that the water mist generated by high-pressure water jets during sanitation vehicle water spraying operations can obstruct optical images and scatter light, leading to a decrease in the detection confidence of camera 31 for environmental obstacles, while millimeter-wave radar 32 has good penetration characteristics through water mist, the edge computing platform 52 introduces the water system operation parameters of the electric chassis 10 as an external adjustment excitation for the perception weights. Since the water pump output pressure is positively correlated with the concentration of water mist during operation, the system obtains the normalized water pump pressure under the current operating state and combines it with the previously extracted sensor state factors. During the calculation process, the system introduces water pump pressure to negatively compensate for the weights of the visual sensors, while allowing the radar's initial weights to directly use the extracted radar health factors. The initial weight calculation model for the visual data is constructed as follows: ; In the formula, This represents the calculated initial visual weights; This represents the extracted visual quality factor; The value represents the set water mist attenuation coefficient. Preferably, its value ranges from 0.4 to 0.8. The specific value can be obtained by offline fitting and calibration of water mist occlusion image datasets under different water pump pressures, which is used to control the slope of the influence of water mist on visual confidence. This indicates the normalized pump pressure obtained.

[0077] S307, Weight Normalization and Limiting Protection. After obtaining the two initial weights mentioned above, the edge computing platform 52 performs baseline limiting on the initial visual and radar weights before normalization. Specifically, the system compares the initial visual and radar weights with preset zero-baseline weights, extracts the maximum value as the effective weight after limiting, and then performs normalization processing through conventional proportional conversion, outputting visual fusion weights and radar fusion weights that sum to 1 respectively. The zero-baseline weights can be calibrated according to sensor performance.

[0078] S308, Environmental Target State Weighted Fusion. After normalization, the edge computing platform 52 incorporates the visual fusion weights and radar fusion weights into the multi-sensor target-level fusion framework. For the same detected environmental obstacle, the system uses the aforementioned weights to perform a weighted summation update of the target state vector output by the visual sensor and the target state vector (such as relative distance and relative velocity) output by the millimeter-wave radar 32. This reduces the weight of visual data under water mist conditions.

[0079] In this embodiment, after completing the multi-source environmental perception fusion and driving load assessment, the edge computing platform 52 needs to convert the decision results into action commands for the underlying actuators. To prevent overshoot of actuator commands due to policy calculation deviations, the system implements dual constraints of rate of change and absolute boundary before issuing commands. The specific implementation process includes the following steps: S309, Control State Vector Construction. The edge computing platform 52 aligns the environmental target state, feedforward trigger flag, and corrected driving load index by timestamp to construct the control state vector.

[0080] S310, Candidate Control Variable Generation. The edge computing platform 52 receives the control state vector and substitutes it into the preset control strategy model for mapping and calculation. The control strategy model can internally set mapping relationships based on multi-dimensional lookup tables or rule logic. For the generation of candidate pre-pressurization power, the model maps it positively to the expected dust level and negatively to the actual path distance. That is, the more severe the expected dust and the closer to the target node ahead, the greater the output candidate pre-pressurization power, to ensure that the water pump builds pressure quickly in advance. For the generation of candidate deceleration, the model comprehensively evaluates the environmental target state (such as the relative distance to obstacles) and the driving load index. When the environmental target distance is lower than the safety threshold, or when the driving load index is in a high state (i.e., the driver's attention is distracted), the model outputs a larger candidate deceleration, thereby actively reducing speed in advance to ensure operational and driving safety.

[0081] S311, Dynamic Determination of Safety Limit Boundaries. While acquiring initial commands, the system dynamically defines the upper limit of control output based on operating conditions. When the vehicle performs water-based dust suppression operations, the sprayed water reduces the road surface adhesion coefficient, and excessive braking commands can easily cause wheel lock-up and slippage; similarly, if the water pump receives excessive pressure commands instantaneously, it can easily cause water hammer effects inside the pipeline. The edge computing platform 52, combined with the current chassis speed and water pump operating status, determines the maximum allowable deceleration and the absolute upper limit of the water pump pre-pressurization power under the current operating conditions. Preferably, the system introduces a derating factor to the braking upper limit based on the water-based operation activation status, for example, multiplying the chassis's nominal maximum deceleration by a derating ratio of 0.7 to 0.8 to obtain the maximum deceleration; the absolute upper limit of the water pump pre-pressurization power is configured to be 80% to 90% of the water pump motor's rated power. Simultaneously, based on the braking system 13 response delay and high-pressure water pump mechanical inertia fed back from the chassis bus, the system pre-calibrates the maximum allowable deceleration change and the maximum allowable power change within a single control cycle.

[0082] S312, dual-constraint execution and target command output. The edge computing platform 52 performs rate-of-change limiting and absolute boundary limiting on the candidate deceleration and candidate pre-pressurization power respectively to obtain the target deceleration and target pre-pressurization power, and sends them to the chassis drive-by-wire actuator and the superstructure controller through the vehicle control network.

[0083] In this embodiment, before issuing control commands, the edge computing platform 52 needs to coordinate the driver's manual operation and the system's automatic control request. When the system detects the driver's intention to intervene, it introduces a smooth arbitration and hysteresis recovery mechanism to achieve a smooth transition of human-machine control. The specific implementation process includes the following steps: S401, Multi-dimensional Intervention Intent Intensity Extraction. During vehicle operation, the edge computing platform 52 reads real-time interaction interface data from the chassis drive-by-wire system, including accelerator pedal opening, brake pedal travel, and steering wheel torque. The edge computing platform 52 normalizes the accelerator pedal opening, brake pedal travel, and steering wheel torque, and calculates the comprehensive intervention intent intensity. This indicator is used to quantify the driver's current level of willingness to intervene in the underlying actuators.

[0084] S402, Defining the Uncertainty of Intervention Status. After determining the level of intent, the edge computing platform 52 compares it with a preset intent range. The system has built-in explicit intervention thresholds and slight intervention thresholds. Preferably, the slight intervention threshold is set to 0.1 to 0.2, and the explicit intervention threshold is set to 0.7 to 0.8. When the overall intensity of the intervention intent is lower than or equal to the slight intervention threshold, it is determined that there is no effective intervention, and the automatic control state is maintained; if the intensity is greater than or equal to the explicit intervention threshold, control is returned to the driver, and the automatic deceleration request is reduced to zero at a preset descent slope, starting from the current control arbitration coefficient. The water pump pre-pressurization power request is switched to manual operation mode or maintains the current safe power before exiting at a preset slope.

[0085] When the intensity of the overall intervention intention is greater than the slight intervention threshold but less than the explicit intervention threshold, the system determines that it is currently in an uncertain intervention state. At this time, the driver has slight contact with the operating parts, but the force has not reached the condition for complete takeover.

[0086] S403, Smoothing Arbitration and Proportional Attenuation Calculation. For the aforementioned uncertain intervention state, the edge computing platform 52 does not directly cancel the driver input, but instead performs proportional attenuation on the automatic control request based on the intervention intensity. As the driver intervention intensity increases, the system linearly reduces the weight of the automatic control command to adapt to the driver's takeover tendency and reduce the human-machine interaction torque generated by the chassis. The edge computing platform 52 constructs an attenuation arbitration model to calculate the current control arbitration coefficient. The calculation model is as follows: ; In the formula, Indicates the first The current control arbitration coefficient calculated for each control cycle; Indicates the first The intensity of comprehensive intervention intention calculated over each control cycle; This indicates the preset threshold for minor intervention; This indicates a pre-defined, explicit intervention threshold; This represents the calibrated maximum attenuation ratio constant, preferably set between 0.5 and 0.8. After obtaining this coefficient, the edge computing platform 52 first limits the current control arbitration coefficient to between 0 and 1; under uncertain intervention conditions, it can be further limited to... Between 1 and 1, the target deceleration calculated in the previous process is multiplied by the current control arbitration coefficient to output the attenuated actual braking command, while the target pre-pressurization power remains the same as the original automatic control command (unless active intervention against the upper structure operation panel is detected). This achieves a smooth transition of control in human-machine co-driving mode.

[0087] S404, Hysteresis Recovery Mechanism Execution. When the intensity of the overall intervention intention is detected to decrease below the minor intervention threshold, it indicates that the driver's intervention behavior has ended. To avoid chassis jerking or pipeline water hammer effect caused by command jumps when the automatic system restores full-authority control, the edge computing platform 52 initiates hysteresis recovery logic. After entering the recovery transition period, the system controls the current control arbitration coefficient to smoothly increase in each control cycle according to a pre-calibrated controlled recovery change rate. Preferably, this controlled recovery change rate is configured to increase by 0.02 to 0.05 in each control cycle. When the current control arbitration coefficient increases to 1, the transition period ends, and the system restores full automatic control authority.

[0088] In this embodiment, after receiving the control commands output from the preceding arbitration process, the vehicle controller 51 needs to convert them into the actual braking force required by the underlying actuators. Based on the continuous water consumption of the water tank 21 during sanitation vehicle operation, the system introduces a dynamic calculation mechanism that considers longitudinal resistance to output braking requirements matching the current motion state. The specific implementation process includes the following steps: S405, Vehicle dynamic parameters and environmental status acquisition. Before braking, the vehicle controller 51 collects the vehicle's longitudinal dynamic parameters in real time. In practice, the system reads the load sensor data of the chassis suspension or analyzes the liquid level sensor signal of the superstructure water tank 21. The vehicle controller 51 calculates the current water load based on the liquid level sensor signal of the water tank 21 and obtains the vehicle's dynamic mass by combining it with the vehicle's curb weight.

[0089] Simultaneously, the system analyzes data from the onboard inertial measurement unit to extract the current road slope angle, and combines this with the average value of the anti-lock braking system or the wheel speed sensors of all four wheels to obtain the current vehicle speed. Real-time updates to the vehicle's dynamic mass improve the accuracy of braking force calculations.

[0090] S406, Calculation of longitudinal driving resistance. The vehicle controller 51 calculates the current slope resistance and tire rolling resistance of the vehicle based on the vehicle's dynamic mass, road slope angle, and preset rolling resistance coefficient.

[0091] S407, Target Total Braking Force Calculation. After clarifying the environmental resistance state, the vehicle controller 51 extracts the target deceleration determined through arbitration in the aforementioned process and constructs a dynamic model for mapping calculation. The calculation model used to calculate the braking force required to maintain the target's motion state is constructed as follows: ; In the formula, This represents the calculated target total braking force; This indicates the acquired dynamic quality of the entire vehicle; Indicates the target deceleration input; This represents the acceleration due to gravity, typically taken as 9.8 m / s². 2 ; This indicates the preset rolling resistance coefficient, which is preferably configured between 0.015 and 0.025 for heavy-duty commercial tires. This indicates the obtained road slope angle, which is defined as a positive value when going uphill and a negative value when going downhill; Represents the cosine function; This represents the sine function.

[0092] After calculation, the system applies a non-negative limit to the total braking force of the target. If the calculated value is less than zero (i.e., the environmental resistance of the steep slope already meets the deceleration requirements), it is limited to zero to avoid issuing a reverse command.

[0093] See Figure 7 In this embodiment, after determining the chassis braking requirements and the superstructure water pump power requirements, the vehicle controller 51 needs to coordinate energy scheduling within the vehicle's high-voltage network. To avoid high-frequency fluctuations on the chassis braking side interfering with the superstructure equipment, the system introduces an energy source coupling and control flow low-pass decoupling mechanism. The specific implementation process includes the following steps: S408, Chassis Regenerative Braking Power Analysis. During vehicle deceleration, the vehicle controller 51 performs electromechanical composite braking distribution based on the previously calculated target total braking force, extracting the electric braking torque allocated to the drive motor 12. By combining this electric braking torque with the wheel rolling radius and the mechanical speed ratio of the transmission system, the system converts it into the vehicle's longitudinal electric braking force. Based on this, the longitudinal electric braking force is multiplied by the current vehicle speed and the overall power generation efficiency of the electric drive system to obtain the current recoverable real-time regenerative braking power.

[0094] S409, control flow command decoupling execution. The vehicle controller 51 sends the target pre-pressurization power to the water pump driver 23, which independently controls the operation of the centrifugal water pump 22 to avoid fluctuations in water pump speed with braking requests.

[0095] S410, Energy Source Coupling Distribution and Buffering. After clarifying the control commands for each device, the system establishes an energy distribution mechanism at the vehicle's high-voltage electrical topology level based on the principle of power conservation. The vehicle controller 51 sets energy priority absorption logic based on the shared high-voltage bus, prioritizing the supply of regenerative braking power to the water pump driver 23. For the difference between power supply and demand, the system utilizes the charging and discharging characteristics of the power battery 11 to absorb fluctuating energy on the bus. In a specific embodiment, the dynamic compensation power of the power battery 11 is set as the difference between the real-time electrical power required for the water pump driver 23 to execute the target pre-pressurization power and the real-time regenerative braking power.

[0096] Preferably, during dynamic compensation, the vehicle controller 51 combines the current state of charge and cell temperature fed back by the battery management system to obtain the current allowable charge / discharge power boundary of the power battery 11. Through this allocation architecture, when the regenerative braking energy is insufficient to maintain the water pump operation but does not exceed the allowable discharge boundary, the power shortfall is filled by the discharge of the power battery 11; when there is a surplus of regenerative energy but it does not exceed the allowable charging boundary, the excess power is charged into the power battery 11 for storage. For abnormal states where the power exceeds the limit due to full charge or low temperature, subsequent power limiting and mechanical compensation control are triggered. In this process, the power battery 11 plays a role in power smoothing in the vehicle's energy network, realizing energy coupling at the source end and command decoupling at the device control level.

[0097] In this embodiment, after coordinating the energy distribution of the high-voltage network, the vehicle controller 51 needs to further ensure that the deceleration request of the chassis braking system 13 is effectively responded to. During electromechanical hybrid braking, due to limitations in the state of the energy storage devices, the electric braking force often cannot fully meet the braking demand. Therefore, the system establishes an excess power handling and basic braking compensation mechanism. The specific implementation process includes the following steps: S411, Excess Power Limitation and Executable Electric Power Determination. During deceleration, if the calculated real-time regenerative braking power exceeds the electric power required to maintain the target pre-pressurization power, the system determines that there is excess regenerative power and allocates it to the power battery 11 for recharge storage. During this stage, the vehicle controller 51 continuously monitors the real-time allowable charging power limit reported by the battery management system. When the excess regenerative power exceeds the allowable charging power limit of the power battery 11, the vehicle controller 51 limits the power generation of the electric drive system and determines the restricted regenerative braking power.

[0098] S412, Actual Regenerative Braking Force Mapping. After obtaining the power boundary, the vehicle controller 51 converts the power command into a low-level mechanical control quantity. Specifically, the system extracts the limited regenerative braking power. When the current vehicle speed is higher than the preset minimum regenerative braking speed threshold, the limited regenerative braking power is divided by the current vehicle speed and multiplied by the comprehensive mechanical efficiency of the electromechanical transmission system to calculate the actual regenerative braking force that the drive motor 12 can currently output to the wheels. When the current vehicle speed is not higher than the minimum speed threshold, the system sets the actual regenerative braking force to zero or limits it to a low-speed executable value, and the braking system 13 undertakes the corresponding braking demand. Subsequently, the system sends a negative torque command corresponding to the actual regenerative braking force to the drive motor 12 controller, and the drive motor 12 performs the electric braking action.

[0099] S413, Basic Braking Compensation Calculation and Control. Preferably, the electromechanical composite braking control follows the principle of prioritizing regenerative braking by the motor and using friction braking as a supplement. After determining the braking share shared by the electric drive system, the system performs hydraulic compensation allocation for the difference in vehicle deceleration demand. The vehicle controller 51 extracts the previously determined target total braking force and compares it with the actual regenerative braking force. Before pressure conversion, the system performs a non-negative amplitude limit judgment. If the difference braking force is less than or equal to zero, it indicates that the electric braking has met the vehicle deceleration demand, and the pressure command sent to the braking system 13 is set to zero; if the difference braking force is greater than zero, the system converts it into wheel braking torque and calculates the target hydraulic braking pressure by comprehensively considering mechanical parameters such as the number of wheels participating in braking, friction efficiency coefficient, piston area, and brake disc radius.

[0100] When the actual regenerative braking force is insufficient to meet the target total braking force, the braking system 13 establishes friction braking force based on the differential braking force to compensate for the insufficient electric braking.

[0101] In this embodiment, the vehicle controller 51, while performing routine feedforward coordination and energy distribution, must ensure the vehicle's driving safety under abnormal operating conditions. For conditions such as positioning loss, high voltage limitation, or chassis instability, the system introduces a multi-dimensional status monitoring and safety degradation mechanism. The specific implementation process includes the following steps: S501, Navigation Positioning and Confidence Degradation Monitoring. During the control cycle, the vehicle controller 51 continuously reads the positioning quality factor and spatial coordinates output by the onboard positioning module 33. Based on the spatial deviation between the vehicle's real-time coordinates and the target map node coordinates, the vehicle controller 51 constructs a confidence degradation model. ; In the formula, This represents the dynamic node confidence score obtained after applying spatial bias penalty for the current node. This represents the initial node knowledge confidence level extracted from the spatiotemporal task knowledge graph corresponding to the current node (its value is equivalent to the basic node knowledge confidence level calculated above). ); This indicates the actual spatial deviation distance between the vehicle's current real-time coordinates and the target map node coordinates; The standard deviation of the effective positioning error is represented by the following formula: As the standard deviation of the effective positioning error, This represents the standard deviation of the positioning error currently reported by the vehicle positioning module 33. This is a preset minimum standard deviation lower limit, used to prevent numerical overflow of the exponential penalty term caused by abnormal reporting of positioning errors.

[0102] After obtaining the dynamic node confidence level, the system compares it with a preset safety confidence threshold. Preferably, this safety confidence threshold can be dynamically determined by looking up a table based on the vehicle's current speed, and can be configured between 0.6 and 0.8 at normal operating speeds. If the dynamic node confidence level is lower than the safety confidence threshold, the graph-based feedforward matching is paused, and the water pump driver 23 is switched to closed-loop control based on real-time water pressure feedback.

[0103] S502, High-voltage electrical boundary limitation degradation. In addition to monitoring at the information dimension, the boundary state at the energy dimension is also included in the monitoring system. The vehicle controller 51 continuously collects the high-voltage DC bus voltage and the real-time charge and discharge limit values ​​of the power battery 11. When the high-voltage bus voltage exceeds the safe operating range, or the charge and discharge capacity of the power battery 11 is limited, the system triggers energy flow degradation. Under this condition, the vehicle controller 51 actively cuts off the electric braking request in the electromechanical composite braking system, so that the chassis braking demand is fully borne by the braking system 13, while limiting or suspending the power output of the water pump driver 23 to prevent the vehicle's high-voltage network from triggering hardware protection due to power over-limit.

[0104] S503, chassis dynamic instability intervention degradation. The vehicle controller 51 monitors the activation status of the anti-lock braking system and the electronic stability control system in real time; when active intervention is triggered, feedforward control is suspended and the target pre-pressurization power is canceled. After the active intervention ends and there is no re-triggering for a preset duration, feedforward cooperative control is allowed to re-intervene. The underlying execution logic involved in this embodiment, such as conventional coordinate system transformation, data fusion filtering (e.g., Kalman filtering), communication protocol parsing, spatial search algorithms (e.g., KD tree), motor vector control, and conventional hydraulic brake regulation, are all well-known technologies in the field. Those skilled in the art can implement them using existing mature algorithms or hardware, and will not be elaborated further in this document.

[0105] Taking a pure electric sweeper as an example, it performs patrol operations at 15km / h on a straight asphalt road. The construction intersection ahead is marked as a high dust area in the historical map.

[0106] Step 1: Preliminary Feature Prediction and Forward Search. The edge computing platform 52 obtains the current vehicle speed of 4.17 m / s and matches it with the current asphalt road surface disturbance ratio coefficient (set to 0.5) and the basic concentration constant (set to 20). Substituting these values ​​into the dust disturbance theoretical prediction model, the theoretical dust disturbance value based on the current vehicle speed is calculated to be 28.69 μg / m³. 3 Based on the calibration value of 1.5s for the comprehensive delay time of the water pump pressurization and the foundation lead of 3m, the dynamic forward search distance calculation model is substituted into the model, resulting in a dynamic forward search distance of 4.17×1.5+3=9.25m. The system matches the target node 9.25m ahead of the map and extracts the historical expected dust level of that node as Level 3 (severe).

[0107] Step 2: Dual Feedforward Trigger and Degradation Monitoring. The actual path distance between the current vehicle and the target node is 9m. Based on the current vehicle speed, the estimated arrival time of the dynamic target is 9 / 4.17≈2.16s. This time enters the set feedforward preparation time window, and the spatial distance meets the minimum safe braking distance at the current vehicle speed, triggering the feedforward control conditions. Perform monitoring and positioning status verification: Current vehicle GPS positioning error standard deviation. The actual spatial deviation between the current coordinates and the target map node coordinates is 0.5m. The initial knowledge confidence level of the node is 0.85, with a value of 0.3m.

[0108] Substitute into the spatial bias penalty model: ; The calculated dynamic node confidence level of 0.71 is greater than the preset security threshold (0.6), so the confidence level downgrade was not triggered and the system issued a control command.

[0109] Step 3: The water mist interference compensation and perception weight adjustment system issues a water pump pre-pressurization command. When the actual water pump pressure rises to 80% of the rated pressure (normalized water pump pressure is 0.8), the following adjustments are performed: In the driving load calculation, the compensation adjustment constant is set to 0.4, and the line-of-sight deviation compensation coefficient is calculated to be 1. 0.4 × 0.8 = 0.68. This coefficient is used to reduce the load score weight given by the system due to the driver's head deflection caused by observing the water mist boundary.

[0110] In the sensor weight adjustment, the current camera 31 visual quality factor is evaluated as 0.9, and the water mist attenuation coefficient is set to 0.6. Substituting these values ​​into the initial weight calculation model, the initial visual weight is calculated as 0.9 × (1 / 2)^2. (0.6 × 0.8) = 0.468. After normalization, the fusion weight ratio of the millimeter-wave radar 32 increases, reducing the interference of water mist on visual detection.

[0111] Step 4: The underlying target braking force and energy scheduling calculation system feeds forward a target deceleration request of 0.8 m / s².

[0112] During the vehicle dynamic weight assessment, the initial total weight was 18,000 kg. The water level sensor data was read to calculate the current water consumption of the water tank as 2,000 kg, and the vehicle dynamic weight was updated to 16,000 kg.

[0113] In the calculation of the total braking force of the target, the gravitational acceleration is taken as 9.8 m / s². 2 The rolling resistance coefficient of the heavy-duty tire is 0.02, and the slope angle is 0°. Substituting these values ​​into the calculation model, the target total braking force after overcoming the rolling resistance is... for: ; During energy dispatching, drive motor 12 enters power generation mode to generate regenerative braking energy. Vehicle controller 51, through high-voltage energy distribution module 60, preferentially distributes this regenerative braking power to water pump driver 23. When the motor's power generation is limited due to battery charging constraints, the difference in total braking force is compensated by braking system 13.

[0114] To verify the actual operating effect of the system, a 2km test section containing simulated dust sources was set up in a closed area. Two identical pure electric sanitation vehicles were used, one equipped with the basic feedback control system (control group) and the other with the collaborative feedforward control system of this invention (experimental group), to perform 20 low-speed operation cycles. The test data are as follows: Test results show that the collaborative feedforward control system eliminates the response lag of dust suppression operations by building up pressure in advance. At the same time, it reduces the false detection rate of targets and the energy consumption of the whole vehicle in the water mist environment by using dynamic adjustment of sensor weights and coupling with regenerative braking energy.

[0115] The attached chart uses a timeline of 0 to 10 seconds to record the control process of the vehicle entering the dusty area: Figure 8 (a) The right ordinate represents the target deceleration, and the left ordinate represents the real-time vehicle speed. At 2.8 seconds, the system issues the target deceleration command (dashed line), and the deceleration smoothly rises to 0.8 m / s² in a sinusoidal half-wave pattern. 2 Under this command, the real-time vehicle speed (solid line) smoothly decreased from 15 km / h and stabilized at 5 km / h, demonstrating the effect of limiting the rate of change when the command was issued.

[0116] Figure 8 (b) The vertical axis represents dust concentration. Solid squares in the graph represent the actual dust concentration measured by the sensor, while dashed lines represent the expected dust concentration. The expected dust concentration abruptly changes to 90 μg / m³ at 2.8 seconds. 3 (This triggers feedforward control), while the measured dust concentration only begins to rise after about 5 seconds. The time difference between the two verifies the lead time provided by the feedforward prediction mechanism.

[0117] Figure 8 (c) The left ordinate represents the normalized water pressure, and the right ordinate represents the perception fusion weight. Starting from 3.2 seconds, the water pump pre-pressurizes, causing the normalized water pressure (solid line) to gradually rise to 0.85. During this process, due to the introduction of a water mist interference compensation algorithm, the visual perception weight (dashed line) decreases synchronously with the increase in water pressure, while the radar perception weight (dotted line) increases accordingly.

Claims

1. A method for assisted driving control of sanitation vehicles based on historical driving data, wherein the sanitation vehicle includes an electric chassis (10), a braking system (13), a water pump driver (23), and a drive motor (12), characterized in that, Includes the following steps: Collect historical driving and operation status data to construct a spatiotemporal operation knowledge graph where nodes include historical expected dust levels and node knowledge confidence. Based on the vehicle's current spatial location and current longitudinal speed, the current node in the spatiotemporal operation knowledge graph is matched, and the target node ahead is searched. Based on the historical expected dust levels and the node knowledge confidence in the target nodes ahead, verify the spatiotemporal dual feedforward triggering conditions. When the spatiotemporal dual feedforward triggering condition is met, the target deceleration and the target pre-pressurization power are generated; After generating the target deceleration and the target pre-pressurization power, the intensity of the comprehensive intervention intention is obtained to arbitrate the target deceleration and the target pre-pressurization power, so as to coordinate the control of the drive motor (12), the braking system (13) and the water pump driver (23).

2. The method according to claim 1, characterized in that, The collection of historical driving and operational status data is used to construct a spatiotemporal operational knowledge graph where nodes include historical expected dust levels and node knowledge confidence levels. Specifically, this includes: Collect historical driving and operation status data including historical spatial location, historical longitudinal driving speed, measured values ​​of environmental dust concentration and actual water spray pressure, and filter out valid sampling point data from it; The target work area is spatially discretized to generate multiple location nodes, which are then matched with the effective sampling point data. Feature statistics are extracted from each of the location nodes to calculate the historical expected dust level and the node knowledge confidence level of each location node.

3. The method according to claim 2, characterized in that, The process of filtering out valid sampling point data specifically includes: The positioning quality factor of each sampling point data in the historical driving and operation status data is extracted. When the positioning quality factor is lower than the positioning validity threshold, the corresponding data is removed. The boundary of the retained data is checked based on the historical longitudinal driving speed and the actual water spray pressure to obtain the effective sampling point data. The matching with the valid sampling point data specifically includes: A search domain is defined for the effective sampling point data to extract candidate location nodes; The matching cost is calculated based on the spatial position deviation and heading angle deviation between the effective sampling point data and the candidate location node, and the candidate location node that produces the minimum matching cost value is extracted as the matching target.

4. The method according to claim 1, characterized in that, The step of matching the current node in the spatiotemporal operation knowledge graph based on the vehicle's current spatial position and current longitudinal speed, and searching for the target node ahead, specifically includes: Obtain the current coordinates of the vehicle, and match the node that is closest to it and meets the spatial matching threshold as the current node; The dynamic forward search distance is calculated by combining the current longitudinal driving speed and the system response delay, so as to search for the target node ahead from the current node along the road topology.

5. The method according to claim 1, characterized in that, The verification spatiotemporal dual feedforward triggering condition specifically includes: Calculate the dynamic target arrival time based on the path travel characteristics from the current node to the target node ahead; The spatiotemporal dual feedforward triggering conditions are verified based on the dynamic target arrival time, minimum safe braking distance, historical expected dust level, and node knowledge confidence.

6. The method according to claim 5, characterized in that, The spatiotemporal dual feedforward triggering conditions are verified based on the dynamic target arrival time, minimum safe braking distance, historical expected dust levels, and node knowledge confidence, specifically including: The node knowledge confidence level and the historical expected dust level are determined to be no less than the confidence level determination threshold and the preset minimum action level, respectively. The minimum safe braking distance is calculated by combining the current longitudinal driving speed with the maximum mechanical deceleration of the electric chassis (10); When the arrival time of the dynamic target is determined to match the preset system feedforward preparation time, and the actual path distance to the target node ahead is greater than the minimum safe braking distance, the spatiotemporal dual feedforward triggering condition is determined to be satisfied.

7. The method according to claim 1, characterized in that, The sanitation vehicle also includes a monitoring module (40), a sensing module (30), and a centrifugal water pump (22). The generation of target deceleration and target pre-pressurization power specifically includes: The monitoring module (40) collects the driver's line-of-sight deviation time ratio and the driver's abnormal load index. The current actual water spray pressure is obtained to calculate the normalized water pump pressure, and a line-of-sight deviation compensation coefficient is constructed based on the normalized water pump pressure to weight the line-of-sight deviation time ratio, and the driving load index is calculated and corrected in combination with the driver abnormal load index. The environmental target state is obtained by integrating the data from the sensing module (30) and combined with the modified driving load index to construct a control state vector, thereby generating the target deceleration and the target pre-pressurization power.

8. The method according to claim 7, characterized in that, The sensing module (30) includes a camera (31) and a millimeter-wave radar (32); The acquisition of the environmental target state by integrating the data from the perception module (30) specifically includes: Extract the visual quality factor of the camera (31) and the radar health factor of the millimeter-wave radar (32); The normalized water pump pressure is used to negatively compensate the visual quality factor to calculate the initial visual weight, and the radar health factor is used as the initial radar weight. The initial visual weights and the initial radar weights are subjected to amplitude limiting and normalization processing to obtain the visual fusion weights and radar fusion weights, respectively. Using the visual fusion weight and the radar fusion weight, the target state vector output by the camera (31) and the target state vector output by the millimeter-wave radar (32) are weighted and summed to update the environmental target state.

9. The method according to claim 7, characterized in that, By combining the modified driving load index to construct a control state vector, the target deceleration and the target pre-pressurization power are generated, specifically including: The control state vector is substituted into a preset control strategy model to generate candidate deceleration and candidate pre-pressurization power. Based on the current longitudinal travel speed and water pump operating status, determine the maximum deceleration and the absolute upper limit of the water pump pre-pressurization power; The candidate deceleration and the candidate pre-pressurization power are respectively subjected to boundary limiting processing to output the target deceleration and the target pre-pressurization power.

10. A sanitation vehicle auxiliary driving control device based on historical driving data, used for communication connection with the braking system (13), water pump driver (23), and drive motor (12) of the sanitation vehicle, and for communication connection with the corresponding power battery (11), sensing module (30), and monitoring module (40) according to the execution logic, characterized in that, It includes a communication interface, a processor, and a memory, wherein the memory is used to store a computer program, and the processor is used to execute the computer program to implement the method as described in any one of claims 1 to 9.

Images