Snow-melting agent precise spreading system and method based on digital twinning and swarm intelligence

Abstract

The invention provides a snow-melting agent precise spreading system and method based on digital twinning and swarm intelligence. The method comprises the following steps: in a digital twinborn body, taking a global snow and ice melting effect, resource consumption, traffic influence and environmental influence as comprehensive optimization targets, performing simulation scheduling by utilizing a multi-agent reinforcement learning algorithm, and generating an optimal collaborative spreading strategy for an agent cluster; issuing the optimal cooperative spreading strategy to each spreading vehicle for execution; each spreading vehicle serves as an edge computing node, on the basis of executing the optimal collaborative spreading strategy, the sudden road condition is determined based on the state of the target road obtained by each spreading vehicle, and autonomous collaboration and task dynamic adjustment are carried out on the sudden road condition. According to the method, road snow melting maintenance is changed from single-point automation to global intelligence, from static programming to dynamic self-evolution, and from current snow melting to environment-friendly revolutionary crossing, and a new solution is provided for operation management of intelligent traffic infrastructures.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of traffic and infrastructure maintenance, in particular to the technical field of road snow-melting and ice-melting, and specifically to a snow-melting agent precise spreading system and method based on digital twinning and swarm intelligence. BACKGROUND

[0002] Winter snow and ice weather seriously affects road traffic safety. The traditional snow-melting agent spreading operation mainly relies on manual experience or preset procedures, which has the following significant defects: usually using a "one-size-fits-all" spreading method, which cannot be dynamically adjusted according to the actual needs of road surface temperature, traffic flow, and special structures such as bridges and tunnels, resulting in low utilization rate of snow-melting agent and easy environmental pollution. The operation decision cannot be made for preventive spreading at the critical point of road icing, and often the ice and snow are formed and then disposed, which has high safety risk. Excessive spreading will cause long-term damage to the soil, water and vegetation along the road, and corrode the bridge structure, and the negative impact of chlorides snow-melting agent is particularly significant. SUMMARY

[0003] In view of the above problems, the present application is proposed in order to provide a snow-melting agent precise spreading system and method based on digital twinning and swarm intelligence, which can overcome the above problems or at least partially solve the above problems, and can realize an intelligent snow-melting agent spreading solution with precision, efficiency and environmental protection.

[0004] Specifically, the present application provides a snow-melting agent precise spreading method based on digital twinning and swarm intelligence, which comprises: constructing a digital twin containing a model of a target road; and training to obtain an ice and snow state change trend prediction model, the ice and snow state change trend prediction model being configured to obtain the change trend of the ice and snow layer on the target road in a future preset period according to the state of the target road and meteorological forecast data; the state of the target road includes the parameters of the target road, the weather conditions where the target road is located, the state of the ice and snow layer on the target road, and the state of the vehicles on the target road; the parameters of the target road include ordinary road sections, bridge surfaces, and intersections; when the target road needs to melt ice and snow, collecting the state of the target road to keep the state of the model synchronized with the state of the target road, and making the ice and snow state change trend prediction model output the change trend of the ice and snow layer on the target road in a future preset period according to the state of the target road; The dispatchable spreading vehicles are modeled as a cluster of agents, with the type of snow-melting agent, the state of the target road, and the change trend of the ice and snow layer as inputs, in the digital twin, with the global snow-melting and ice-melting effect, resource consumption, traffic impact, and environmental impact as comprehensive optimization objectives, a multi-agent reinforcement learning algorithm is used to simulate scheduling, and an optimal collaborative spreading strategy for the cluster of agents is generated; the optimal collaborative spreading strategy at least includes the operation path, spreading start and stop time, snow-melting agent spreading dose and travel speed of each spreading vehicle, and the intervention time of mechanical snow removal; The optimal collaborative spreading strategy is issued to each spreading vehicle for execution; each spreading vehicle acts as an edge computing node, based on the state of the target road obtained by each agent, determines the sudden traffic condition, and performs autonomous collaboration and dynamic task adjustment on the sudden traffic condition based on the execution of the optimal collaborative spreading strategy; The snow-melting and ice-melting effect data and environmental impact data after the spreading operation are collected; and the optimal collaborative spreading strategy, the execution process of each spreading vehicle, the snow-melting and ice-melting effect data, the environmental impact data, and the state of the target road are constructed as training samples, and the ice and snow state change trend prediction model and the model involved in the multi-agent reinforcement learning algorithm are updated.

[0005] Optionally, the environmental impact data is obtained by regularly collecting and analyzing the corrosion degree of the target road, and the salt ion concentration of the soil sample along the target road, the water conductivity, and the vegetation spectral characteristics; The sudden traffic condition includes rapid icing caused by local road surface temperature drop and operation path interruption caused by traffic congestion; the dynamic adjustment includes temporary exchange of tasks between the spreading vehicles, adjustment of the spreading dose, adjustment of the travel speed, and activation of the standby path.

[0006] Optionally, the snow-melting agent precise spreading method based on digital twin and swarm intelligence further comprises: A target road state prediction model is trained, which is configured to obtain the predicted state of the target road in a future preset period according to meteorological forecast data and historical state of vehicles on the target road; and the target road state prediction model outputs the predicted state of the target road in a future preset period with the meteorological forecast data as input; The ice and snow state change trend prediction model outputs the predicted change trend of the ice and snow layer on the target road in a future preset period according to the predicted state of the target road; The type of snow-melting agent, the predicted state of the target road, and the predicted change trend of the ice and snow layer are input into the digital twin, global snow-melting and ice-melting effect, resource consumption, traffic influence, and environmental influence are taken as comprehensive optimization targets in the digital twin, a multi-agent reinforcement learning algorithm is used for simulation scheduling to generate a predicted optimal collaborative spreading strategy for the agent cluster; When the target road needs to be deiced and snow-melted, the closeness of the state of the target road and the predicted state of the target road is compared; When the closeness of the state of the target road and the predicted state of the target road is higher than a first degree threshold, the predicted optimal collaborative spreading strategy is taken as the optimal collaborative spreading strategy to be executed.

[0007] Optionally, the snow-melting agent precise spreading method based on digital twin and swarm intelligence further comprises: A target road state prediction model is trained, the target road state prediction model is configured to obtain the predicted state of the target road in a future preset time period according to the meteorological forecast data and the historical state of the vehicle on the target road, and the target road state prediction model outputs the predicted state of the target road in the future preset time period with the meteorological forecast data as input; The ice and snow state change trend prediction model outputs the predicted change trend of the ice and snow layer on the target road in a future preset time period according to the predicted state of the target road; The type of snow-melting agent, the predicted state of the target road, and the predicted change trend of the ice and snow layer are input into the digital twin, global snow-melting and ice-melting effect, resource consumption, traffic influence, and environmental influence are taken as comprehensive optimization targets in the digital twin, a multi-agent reinforcement learning algorithm is used for simulation scheduling to generate a predicted optimal collaborative spreading strategy for the agent cluster; The state of the target road is collected at a preset time interval, and it is determined whether deicing and snow-melting is needed according to the state of the target road; If not, the closeness of the state of the target road and the predicted state of the target road is compared; When the closeness of the state of the target road and the predicted state of the target road is higher than a second degree threshold for a first preset number of times, the predicted optimal collaborative spreading strategy is issued to each spreading vehicle for execution; When the closeness of the state of the target road and the predicted state of the target road is lower than the second degree threshold for a second preset number of times, the target road state prediction model outputs the predicted state of the target road in a future preset time period with the meteorological forecast data as input.

[0008] Optionally, the snow-melting agent precise spreading method based on digital twin and swarm intelligence further comprises: obtaining an ice and snow layer change trend on a reference road near the target road which is not melted by snow-melting agent, denoted as ice and snow layer reference change trend; judging the closeness of the ice and snow layer change trend output by the ice and snow state change trend prediction model and the ice and snow layer reference change trend; when the closeness of the ice and snow layer change trend output by the ice and snow state change trend prediction model and the ice and snow layer reference change trend is lower than the third closeness threshold value for a third preset number of times, causing each spreading vehicle to stop executing the optimal collaborative pre-spreading strategy and re-entering the simulation scheduling in the digital twin with the type of snow-melting agent, the predicted state of the target road and the ice and snow layer predicted change trend as inputs, using a multi-agent reinforcement learning algorithm to simulate scheduling to generate a predicted optimal collaborative spreading strategy for the agent cluster.

[0009] Optionally, each of the spreading vehicles is an agent, and the state vector corresponding to each of the spreading vehicles includes position, load amount, device state, current task progress, and state of the target road at the corresponding position; the action vector corresponding to each of the spreading vehicles includes work path, inter-spreading start and stop time, snow-melting agent spreading dose and travel speed, and mechanical snow removal intervention time; the reward function corresponding to each of the spreading vehicles is composed of snow-melting effect, resource consumption, traffic impact and environmental impact.

[0010] Optionally, the agent cluster further comprises: a robot as an agent for snow-melting agent spreading at pedestrian walkways, bicycle lanes and bridge piers; a drone as an agent for snow-melting agent spreading in emergency areas; a snow removal vehicle as an agent for mechanical snow removal; a road surface state sensing vehicle as an agent for obtaining the state of the target road before or after snow-melting agent spreading.

[0011] Optionally, the snow-melting agent precise spreading method based on digital twin and swarm intelligence further comprises: generating a plurality of optional collaborative spreading strategies for the agent cluster, and outputting the advantages of each of the optional collaborative spreading strategies and the optimal collaborative spreading strategy and the differences between them; judging whether to receive strategy selection information selected by a dispatcher; If so, the optional cooperative spreading strategy or the optimal cooperative spreading strategy corresponding to the policy selection information is issued to each spreading vehicle for execution. If the policy selection information is not received within the preset time, the optimal cooperative spreading strategy is issued to each spreading vehicle for execution.

[0012] Optionally, the snow-melting agent precise spreading method based on digital twinning and swarm intelligence further comprises: Recording the dispatcher's multiple selections to learn his preferences; Highlighting the strategy in the optional cooperative spreading strategy that is closest to the preferences and the optimal cooperative spreading strategy.

[0013] The application also provides a snow-melting agent precise spreading system based on digital twinning and swarm intelligence, which comprises: A swarm of agents, comprising intelligent spreading vehicles; A state acquisition device configured to acquire the state of the target road; A network information acquisition module configured to acquire meteorological forecast data through a network; A digital twin comprising a model of the target road; An ice and snow state change trend prediction module having an ice and snow state change trend prediction model; An effect acquisition device configured to acquire snow-melting and ice-melting effect data and environmental impact data after spreading operation; A multi-agent reinforcement learning module comprising models involved in a multi-agent reinforcement learning algorithm; The snow-melting agent precise spreading system based on digital twinning and swarm intelligence is used to implement the steps of any one of the snow-melting agent precise spreading methods based on digital twinning and swarm intelligence.

[0014] The snow-melting agent precise spreading system and method based on digital twinning and swarm intelligence provided by the present application is an intelligent ice and snow road maintenance decision-making and execution closed-loop snow-melting agent spreading method and system which integrates digital twinning, multi-agent reinforcement learning and edge computing, can significantly improve the winter road safety level, reduce the maintenance cost and reduce the impact on the environment, and has great social and economic value. It can realize the closed loop from physical perception to intelligent decision-making, from global optimization to edge execution, and bring revolutionary technical effects. Specifically, through the deduction function of the digital twinning body, the future consequences of different decisions can be predicted, so as to find the optimal decision, thereby realizing truly scientific decision-making and preventive maintenance. Through the swarm intelligence technology, the multi-vehicle cooperative operation is realized, which is upgraded from "individual optimization" to "global optimization", greatly improving the operation efficiency under complex road network. The system corresponding to the method is a "living body" which can continuously learn, iterate and evolve from massive practical data, and its intelligent level increases infinitely over time. In particular, the long-term environmental protection index monitoring is introduced, which extends the maintenance decision from the short-term "good snow-melting effect" to the long-term "infrastructure health and ecological friendliness", realizing sustainable development. The embodiment of the present application realizes the revolutionary leap of road snow-melting maintenance from single-point automation to global intelligence, from static programming to dynamic self-evolution, from solving current snow-melting to environment-friendly, and provides a new solution for the operation and management of intelligent transportation infrastructure.

[0015] The above and other objects, advantages and features of the present application will become more apparent from the following detailed description of some embodiments thereof, taken in conjunction with the accompanying drawings. BRIEF DESCRIPTION OF DRAWINGS

[0016] Some specific embodiments of the present application will be described in detail below with reference to the accompanying drawings. The same reference numerals in the drawings denote the same or similar components or parts. It should be understood by those skilled in the art that the drawings are not necessarily drawn to scale. In the drawings: Figure 1 is a schematic flow chart of a snow-melting agent precise spreading method based on digital twinning and swarm intelligence according to an embodiment of the present application; Figure 2 is a schematic flow chart of a snow-melting agent precise spreading method based on digital twinning and swarm intelligence according to another embodiment of the present application; Figure 3 is a schematic flow chart of a snow-melting agent precise spreading method based on digital twinning and swarm intelligence according to still another embodiment of the present application. DETAILED DESCRIPTION

[0017] Figure 1 is a schematic flow chart of a snow-melting agent precise spreading method based on digital twinning and swarm intelligence according to an embodiment of the present application, which can generally include: Step S100: Construct a digital twin containing the target road model. Train a snow and ice condition change trend prediction model, configured to predict the snow and ice layer change trend on the target road within a preset future time period based on the target road's condition and weather forecast data. The target road's condition includes its parameters, the weather conditions, the snow and ice layer condition, and the vehicle status. The target road's parameters include ordinary road sections, bridge surfaces, and intersections.

[0018] In this step, a digital twin containing the target road model is constructed. This digital twin includes a three-dimensional geometric model of the road, distinguishing different road parameters such as ordinary road sections, bridge surfaces, and intersections, providing an accurate foundation for subsequent analysis. Simultaneously, an ice and snow condition change trend prediction model is trained. This model, by fusing multi-dimensional data, can predict ice and snow accumulation trends with an accuracy rate exceeding 85%, an improvement of approximately 30 percentage points compared to traditional experience-based judgment.

[0019] Step S200: When the target road requires ice and snow melting, the state of the target road is collected to synchronize the state of the model with that of the target road. The digital twin synchronizes with the real road state, forming a dynamic virtual road environment that maps 1:1 to the real world, providing a reliable basis for decision-making. The state of the target road can be collected in real time through pavement sensors, cameras, weather stations, vehicle GPS, and / or the Internet of Things, forming a beyond-field-of-view, holographic perception of road surface temperature, humidity, snow accumulation, ice layer, and traffic flow intensity. The ice and snow state change trend prediction model outputs the future ice and snow layer change trend of the target road within a preset time period based on the target road's state. Typically, when snowfall begins, it is determined that the target road requires ice and snow melting.

[0020] Step S300 involves modeling the schedulable snow-spreading vehicles as an intelligent agent cluster. Using the type of snow-melting agent, the state of the target road, and the trend of snow and ice layer changes as inputs, and within the digital twin, a multi-agent reinforcement learning algorithm is used to simulate scheduling based on the overall snow and ice melting effect, resource consumption, traffic impact, and environmental impact. This generates an optimal collaborative snow-spreading strategy for the intelligent agent cluster. The optimal collaborative snow-spreading strategy includes at least the operating path of each snow-spreading vehicle, the start and stop times of snow-spreading, the dosage and speed of the snow-melting agent, and the intervention time of mechanical snow removal. Through tens of thousands of simulation experiments within the digital twin, a comprehensive optimal strategy can be generated within 5 minutes, improving resource utilization by 25%, expecting a 35% improvement in de-icing effect, and reducing traffic impact by 20%. With the overall optimization goals of snow and ice melting effect, resource consumption, traffic impact and environmental impact, it can minimize road closure time, maintain road network traffic capacity, effectively reduce the traffic accident rate caused by road ice and snow, and reduce social and economic losses. Moreover, it can reduce the environmental burden, which is in line with the ecological and environmental protection policy orientation, and reduce the erosion of soil, water sources and road facilities by harmful substances in snow melting agents from the source.

[0021] Step S400 involves distributing the optimal collaborative spreading strategy to each spreading vehicle for execution. Each spreading vehicle, acting as an edge computing node, determines unexpected road conditions based on the acquired target road status, in addition to executing the optimal collaborative spreading strategy. It then autonomously coordinates and dynamically adjusts tasks to address these unexpected road conditions. This edge-driven autonomous coordination significantly shortens response time to emergencies, achieving a task adjustment success rate of over 90% between vehicles. When a traffic accident or sudden heavy snowfall occurs on a road segment, nearby spreading vehicles can autonomously negotiate and adjust their work routes and task allocations to ensure priority handling of critical road segments, improving emergency response efficiency by 50%.

[0022] Step S500 involves collecting data on snow and ice melting effects and environmental impacts after the snow and ice spreading operation. The optimal coordinated spreading strategy, the execution process of each spreading vehicle, the snow and ice melting effect data, the environmental impact data, and the target road status are then used as training samples to update the snow and ice state change trend prediction model and the models involved in the multi-agent reinforcement learning algorithm. Through a closed-loop learning mechanism, a continuously self-improving intelligent system is formed. This continuous learning enables the system to adapt to changes in different climatic conditions and road environments.

[0023] In this embodiment of the invention, each model is trained based on corresponding historical data. Utilizing a multi-agent reinforcement learning algorithm, tens of thousands of simulated scheduling operations are performed within the digital twin. With the comprehensive optimization objective of "optimal global snow melting effect, lowest total resource consumption, minimal traffic impact, and minimal environmental impact," the optimal vehicle scheduling path and task allocation scheme are calculated and dynamically assigned to each agent. The spreading vehicles, acting as edge computing nodes, receive macro-level tasks but possess local autonomous decision-making capabilities. Spreading vehicles exchange information with surrounding spreading vehicles. In the event of emergencies (such as severe localized icing or traffic congestion), the spreading vehicle cluster can autonomously negotiate, with the most suitable spreading vehicle making temporary task adjustments, achieving decentralized and agile response. The system continuously optimizes each model through an online learning mechanism, enabling the entire system to self-evolve and become "smarter with use." Through the predictive capabilities of the digital twin, the future consequences of different decisions can be foreseen, thereby finding the optimal decision and achieving truly scientific decision-making and preventative maintenance. By leveraging swarm intelligence technology, collaborative operations among multiple vehicles are achieved, upgrading from "individual optimization" to "global optimization," significantly improving operational efficiency in complex road networks. The corresponding system is a "living entity" capable of continuously learning, iterating, and evolving from massive amounts of real-world data, with its intelligence level increasing infinitely over time. In particular, the introduction of long-term environmental monitoring extends maintenance decisions from short-term "good snow melting effect" to long-term "infrastructure health and eco-friendliness," achieving sustainable development. This invention represents a revolutionary leap in road snow melting maintenance, from single-point automation to global intelligence, from static programming to dynamic self-evolution, and from simply addressing immediate snow melting needs to achieving environmental friendliness, providing a new solution for the operation and management of intelligent transportation infrastructure.

[0024] Traditional methods rely on current road conditions. The method in this invention, however, uses a snow and ice condition change trend prediction model to accurately predict the spatiotemporal evolution of snow and ice accumulation / melting on various road sections several hours in advance, achieving "prevention before snowfall." The decision-making process considers not only the current condition but also future weather impacts and traffic disturbances, making the selection of the "time window" and "spatial focus" for snow-spreading operations more scientific, avoiding spreading too early (when the snow-melting agent is washed away / becomes ineffective) or too late (when safety hazards have already arisen).

[0025] By simulating massive strategies in digital twins through multi-agent reinforcement learning, collaborative solutions approaching the global optimum can be identified. Compared to manual or rule-based scheduling, this achieves faster global road network compliance with the same resources (vehicles, de-icing agents) or minimizes resource consumption (especially expensive environmentally friendly de-icing agents) under the same compliance requirements. The fault-tolerant "center-edge" collaborative architecture ensures macro-level order and efficiency through globally generated strategies at the center, while the autonomous collaborative capabilities of edge nodes enable the system to respond quickly and self-heal to emergencies (local congestion, equipment failure, sudden weather changes), allowing critical areas to still receive emergency treatment and significantly improving the reliability of the entire maintenance system.

[0026] This can significantly improve road safety. More accurate forecasting and more timely operations can minimize the time roads are in high-risk conditions such as "black ice" or "compacted snow," thereby reducing winter traffic accidents at the source. Optimized operation routes and timing can reduce the ineffective driving and exposure time of spreading vehicles in severe weather, while also reducing the interference of their operations on normal traffic flow and minimizing the risk of secondary accidents.

[0027] By precisely controlling dosage, reducing ineffective application, and optimizing type selection, it is estimated that 10% to 30% of de-icing agent usage can be saved. Optimized routes and collaborative operations reduce the total mileage and empty-run rate of the fleet. It reduces reliance on human experience, lowers the workload of dispatching personnel, and enables more efficient "less-staffed" smart operations. The increased road capacity and reduced closure time bring significant economic benefits to social logistics and transportation. Precise control of chloride-based de-icing agents can significantly reduce salinization and corrosion damage to surrounding soil and water bodies.

[0028] It possesses continuous evolution capabilities: the continuous learning loop enables the system to continuously improve itself during use. Each model becomes more accurate with new data, the scheduling strategy becomes more intelligent with new scenarios, and the overall system performance does not decline over time, but rather continues to improve.

[0029] In summary, the precise snow-melting agent application method based on digital twins and swarm intelligence of this invention goes far beyond "automation." Its core lies in achieving "intelligent" and "ecological" maintenance systems. It constructs a decision-making test field parallel to the physical world through digital twins, finds complex optimization solutions that are difficult for human experience to achieve through multi-agent reinforcement learning, injects intelligence into the execution terminal through edge computing, and finally enables the system to evolve through a continuous learning closed loop, achieving self-evolution.

[0030] In some embodiments of the present invention, environmental impact data is obtained by periodically collecting and analyzing the corrosion degree of the target road, as well as the salt ion concentration, water conductivity, and vegetation spectral characteristics of soil samples along the target road. Specifically, the goal is to minimize the corrosion degree of the target road, the impact on the soil, and the impact on plant growth. The environmental impact data corresponding to the lowest corrosion degree, the least impact on the soil, and the least impact on plant growth of the target road is the lowest. The higher the corrosion degree of the target road, the greater the impact on the soil, and the greater the impact on plant growth, the higher the corresponding environmental impact data.

[0031] In some embodiments of the present invention, sudden road conditions include rapid icing caused by a sudden drop in local road surface temperature and interruption of work routes due to traffic congestion. Dynamic adjustments include temporary task switching between spreading vehicles, adjustment of spreading dosage, adjustment of travel speed, and activation of backup routes. Due to the emergency road condition response mechanism, operation failure is avoided due to the obstruction of a single vehicle; dynamic adjustment of spreading dosage increases the effectiveness of handling rapidly icing road sections by 40%, achieving better anti-icing effects with the same resource consumption; intelligent route switching reduces the average delay time of working vehicles by 65%, ensuring maximum utilization of the snow removal operation window. This increases the overall snow melting operation completion rate from 70% in the traditional mode to over 92% when facing uncertainties.

[0032] In some embodiments of the present invention, such as Figure 2 As shown, the method for precise application of de-icing agents based on digital twins and swarm intelligence also includes: Step S110: Train the target road condition prediction model. The target road condition prediction model is configured to obtain the predicted state of the target road within a preset future time period based on meteorological forecast data and the historical state of vehicles on the target road. Using meteorological forecast data as input, the target road condition prediction model outputs the predicted state of the target road within the preset future time period.

[0033] Step S210: The snow and ice condition change trend prediction model outputs the predicted change trend of the snow and ice layer on the target road within a future preset time period based on the predicted state of the target road.

[0034] Step S310: Taking the type of snow melting agent, the predicted state of the target road, and the predicted trend of snow and ice layer changes as inputs, in the digital twin, with the global snow and ice melting effect, resource consumption, traffic impact, and environmental impact as comprehensive optimization objectives, a multi-agent reinforcement learning algorithm is used to simulate scheduling and generate the predicted optimal cooperative distributing strategy for the agent cluster.

[0035] Step S410: When the target road needs to melt ice and snow, compare the degree of similarity between the state of the target road and the predicted state of the target road.

[0036] Step S420: When the similarity between the state of the target road and the predicted state of the target road is higher than a first degree threshold, the predicted optimal cooperative dispensing strategy is used as the optimal cooperative dispensing strategy to be executed. The higher the similarity, the larger the corresponding first degree threshold, with a maximum value of 1.

[0037] In this embodiment of the invention, before snowfall, based on weather forecast data (e.g., snow is predicted within the next 6 hours), a multi-agent reinforcement learning algorithm is used in the digital twin to perform pre-simulation scheduling, generating a predicted optimal cooperative spreading strategy for the agent cluster. When snowfall occurs, if the state of the target road closely matches its predicted state, it indicates that the predicted optimal cooperative spreading strategy is applicable to the current target road, and no further prediction is needed; the predicted optimal cooperative spreading strategy can be directly issued. Otherwise, the process returns to step S200. This embodiment of the invention achieves "zero-latency" startup. When snowfall actually begins, the optimal spreading strategy has already been pre-generated and is in standby mode, allowing the work vehicles to immediately deploy as planned, avoiding the traditional lag time from perception to decision-making. In non-emergency situations, simulation scheduling allows the system to calmly run thousands of iterative optimizations in the digital twin to find the truly globally optimal cooperative strategy, avoiding the limitation of only generating a "satisfactory solution" due to time constraints during snowfall. Completing large-scale reinforcement learning computations before snowfall avoids system congestion caused by multiple computationally intensive tasks (real-time prediction, real-time scheduling) occurring concurrently during snowfall, thus shifting computing resources forward and alleviating peak pressure. It allows for parallel comparison of the predicted optimal strategy and the real-time generated strategy, automatically switching to real-time mode when the deviation exceeds a threshold, forming a double-insurance mechanism. This not only improves operational efficiency but also redefines the safety threshold and cost boundaries of snow and ice management, representing a key breakthrough for intelligent transportation infrastructure towards forward-looking autonomous operation and maintenance.

[0038] In other embodiments of the invention, such as Figure 3 As shown, the method for precise application of snow-melting agents based on digital twins and swarm intelligence also includes: Step S110: Train the target road condition prediction model. The target road condition prediction model is configured to obtain the predicted state of the target road within a preset future time period based on meteorological forecast data and the historical state of vehicles on the target road. Using meteorological forecast data as input, the target road condition prediction model outputs the predicted state of the target road within the preset future time period.

[0039] Step S210: The snow and ice condition change trend prediction model outputs the predicted change trend of the snow and ice layer on the target road within a future preset time period based on the predicted state of the target road.

[0040] Step S310: Taking the type of snow melting agent, the predicted state of the target road, and the predicted trend of snow and ice layer changes as inputs, in the digital twin, with the global snow and ice melting effect, resource consumption, traffic impact, and environmental impact as comprehensive optimization objectives, a multi-agent reinforcement learning algorithm is used to simulate scheduling and generate the predicted optimal cooperative pre-distribution strategy for the agent cluster.

[0041] Step S510: Collect the status of the target road at preset time intervals, and determine whether ice and snow melting is required based on the status of the target road.

[0042] If not, proceed to step S520 to compare the similarity between the state of the target road and its predicted state. That is, when the target road does not require ice or snow melting, compare the similarity between the state of the target road and its predicted state.

[0043] Step S530: When the number of times the proximity between the state of the target road and the predicted state of the target road exceeds a second threshold reaches a first preset number, and when the target road needs ice and snow melting, the predicted optimal coordinated pre-spreading strategy is distributed to each spreading vehicle for execution. The higher the proximity, the larger the corresponding second threshold, with a maximum value of 1.

[0044] Step S540: When the number of times the proximity between the state of the target road and the predicted state of the target road is lower than the second degree threshold reaches the second preset number, return to step S110.

[0045] If ice and snow need to be melted on the target road, proceed to step S200.

[0046] In this embodiment of the invention, the most time-consuming multi-agent scheduling computation is pre-emptively performed during non-urgent periods (before snowfall), avoiding competition for computing resources during snowfall. Only lightweight "verification comparisons" are performed in the real-time phase. Through a mechanism of "continuous multiple verifications," erroneous triggering or cancellation of jobs due to single data fluctuations or brief anomalies is avoided, making decisions more prudent and reliable, and significantly improving decision robustness. It seeks to leverage predictive information to achieve the benefits of proactive planning while managing the risks of predictive uncertainty through rigorous real-world verification, finding an intelligent balance between aggressiveness and conservatism. Instead of blindly trusting any prediction, it dynamically decides whether to adopt a particular pre-decision strategy through continuous, reality-based verification.

[0047] In some embodiments of the present invention, the method for precise application of snow-melting agents based on digital twins and swarm intelligence further includes: Obtain the trend of snow and ice layer changes on a reference road adjacent to the target road that has not been melted, and denot it as the reference trend of snow and ice layer changes.

[0048] Determine how closely the snow and ice layer change trend output by the snow and ice state change trend prediction model is similar to the reference snow and ice layer change trend.

[0049] When the number of times the snow and ice layer change trend output by the snow and ice state change trend prediction model is less than the third-degree threshold reaches a third preset number, the spreading vehicles stop executing the optimal coordinated pre-spreading strategy and re-enter step S300. Using the type of de-icing agent, the predicted state of the target road, and the predicted snow and ice layer change trend as inputs, in the digital twin, with the global snow and ice melting effect, resource consumption, traffic impact, and environmental impact as comprehensive optimization objectives, a multi-agent reinforcement learning algorithm is used for simulation scheduling to generate a predicted optimal coordinated spreading strategy for the agent cluster. The higher the degree of similarity, the larger the corresponding third-degree threshold, with a maximum value of 1.

[0050] In this embodiment of the invention, by introducing an external physical benchmark verification mechanism, the system utilizes the natural snow and ice change trends of nearby unoperated roads as an objective reference to verify the authenticity of the prediction model output, ensuring that the prediction results conform to actual physical laws. When the predicted trend continuously deviates from the natural benchmark, the system can proactively identify the failure state of the prediction model under the current weather conditions, avoiding "model blind confidence." Once a prediction inaccuracy is detected, the system immediately stops executing the currently potentially erroneous spreading strategy, preventing the waste of de-icing agents, fuel, and other resources under the guidance of erroneous predictions, and avoiding safety and environmental risks caused by insufficient or excessive operations. After triggering the safety brake, the system automatically performs global optimization scheduling based on the latest data, quickly generating a new strategy adapted to the real environment to ensure operational effectiveness. A dual protection system of internal consistency verification plus external physical authenticity verification is constructed, significantly reducing the overall decision-making risk of the system in complex scenarios. Furthermore, events where the model prediction continuously deviates from the natural benchmark are automatically marked as high-value anomalous samples, providing a clear direction for subsequent targeted optimization and retraining of the model. This not only significantly improves the safety and reliability of decision-making in dynamic and real-world environments, but also promotes the evolution of the system from static intelligence to dynamic adaptive intelligence, serving as an important technical guarantee for ensuring the safe and robust operation of AI systems on critical infrastructure.

[0051] In some embodiments of the present invention, the method for precise application of snow-melting agents based on digital twins and swarm intelligence further includes: Generate multiple optional cooperative dispensing strategies for a cluster of agents, and output the advantages of each optional cooperative dispensing strategy and the optimal cooperative dispensing strategy, as well as the differences between them.

[0052] Determine whether the policy selection information selected by the scheduler has been received.

[0053] If so, the optional or optimal collaborative spreading strategy corresponding to the strategy selection information will be distributed to each spreading vehicle for execution.

[0054] If no strategy selection information is received within a preset time (e.g., 5-10 minutes), the optimal collaborative spreading strategy will be sent to each spreading vehicle for execution.

[0055] In this embodiment of the invention, a human-machine collaborative decision-making mechanism is introduced based on the original automatic decision-making process. By providing multiple optimization solutions while retaining human decision-making power, the advantages of artificial intelligence and human experts are complemented. The advantage of automatic decision-making is its ability to quickly process massive amounts of data, perform thousands of simulations, and find the globally optimal or near-optimal solution. The advantage of the dispatcher is that it considers non-quantifiable factors such as political event security and special event arrangements, providing multiple high-quality options for human selection. This respects the final decision-making power of human experts while allowing artificial intelligence to handle the heavy computational burden. In complex, boundary-setting, or high-risk scenarios (such as during major event security periods or extreme weather), the final confirmation right of humans is retained, avoiding unpredictable risks that may arise from a fully autonomous system. Human dispatchers can select the most appropriate strategy based on real-time, non-digitized information (such as temporary traffic control and emergencies).

[0056] Furthermore, the expected performance on objectives such as ice melting effect, resource consumption, traffic impact, and environmental impact is presented in a multi-dimensional scoring format. For example, different scenarios may require different preferences: sometimes prioritizing smooth traffic (selecting the fastest snow and ice removal strategy), sometimes prioritizing cost control (selecting the most resource-efficient strategy), and sometimes prioritizing minimizing environmental impact. The multi-strategy generation mechanism can provide solutions with different focuses to adapt to diverse needs. This is particularly suitable for fields with extremely high requirements for security, reliability, and interpretability, such as smart cities and critical infrastructure, and represents an important evolution of artificial intelligence systems from "fully automated" to "trustworthy intelligent collaboration."

[0057] Furthermore, the method for precise application of snow-melting agents based on digital twins and swarm intelligence also includes: Record the dispatcher's multiple choices and learn their preferences.

[0058] The strategy that is closest to the preference among the available cooperative dispensing strategies, as well as the optimal cooperative dispensing strategy, are highlighted.

[0059] Further enhancements include scheduling preference learning and personalized recommendation capabilities, enabling the system to continuously understand and adapt to operators' decision-making habits, thereby improving human-machine collaboration efficiency and decision-making experience. This preference learning mechanism upgrades human-machine collaboration from a "one-time interaction" to a "long-term partnership," allowing the system to: better understand users, proactively learn and adapt to different scheduling styles; be more efficient, reducing decision-making costs through personalized recommendations; and be more intelligent, providing more targeted intelligent support while respecting the ultimate human decision-making power.

[0060] In some embodiments of the present invention, each spreading vehicle is considered an intelligent agent. The state vector for each spreading vehicle includes its position, load dosage, equipment status, current task progress, and the state of the target road at its corresponding location. The action vector for each spreading vehicle includes its work path, actual spreading start and stop times, the amount of de-icing agent applied, its travel speed, and the intervention time of mechanical snow removal. The reward function for each spreading vehicle is a weighted average of snow and ice melting effect, resource consumption, traffic impact, and environmental impact.

[0061] In some embodiments of the present invention, the intelligent agent cluster further includes robots, drones, snowplows, and road condition sensing vehicles. A robot, as an intelligent agent, is used for spreading de-icing agents on sidewalks, bike lanes, and bridge piers. A drone, as an intelligent agent, is used for spreading de-icing agents in emergency areas. A snowplow, as an intelligent agent, is used for mechanical snow removal. A road condition sensing vehicle, as an intelligent agent, is used to acquire the status of the target road before or after the de-icing agent is spread. The optimal cooperative spreading strategy may also include the operating path of each robot / drone, the start and stop times of spreading, the dosage of de-icing agent, and the travel speed.

[0062] In this embodiment, the heterogeneous multi-agent collaborative operation system realizes a three-dimensional operational network from main roads to sidewalks, and from the air to the ground, eliminating blind spots in traditional operations. Various agents perform the most suitable tasks (drone emergency response, robot precision operations, and efficient snow removal by snowplows), improving overall operational efficiency by over 40%. Drones can reach any congested or accident-prone section within 5 minutes, enabling emergency response during the golden window period. Aerial spreading complements ground operations, reducing the handling time for complex terrains such as elevated roads and bridges by 60%. Heavy equipment is deployed only when necessary, while light equipment undertakes appropriate tasks, reducing overall energy consumption by 30%. This heterogeneous multi-agent system upgrades road ice and snow management from a single mechanized operation to intelligent ecological collaboration, achieving breakthrough improvements in efficiency, reliability, adaptability, and economy, providing a scalable framework paradigm for smart city infrastructure management. Furthermore, two hours after snow and ice melting operations, drones equipped with thermal imaging cameras are dispatched for inspection, generating heat maps of the snow and ice melting effects. Further roadside sensors were installed to continuously monitor changes in ice thickness, and the "melting rate / dose" ratio was calculated as an efficiency indicator.

[0063] In some embodiments of the present invention, the method for precise application of snow-melting agents based on digital twins and swarm intelligence further includes: The target road was divided into a test area and a control area; The first cooperative dispensing strategy is executed in the test area, and the second cooperative dispensing strategy is executed in the control area. The first cooperative dispensing strategy is the optimal cooperative dispensing strategy output before the model update of the ice and snow state change trend prediction model and the multi-agent reinforcement learning algorithm. The second cooperative dispensing strategy is the optimal cooperative dispensing strategy output before the model update of the ice and snow state change trend prediction model and the multi-agent reinforcement learning algorithm. Data on snow and ice melting effects, resource consumption, traffic impact, and environmental impact were collected from the test area and the control area. Based on snow and ice melting effect data, resource consumption data, traffic impact data, and environmental impact data, the reward values ​​of the first and second coordinated spreading strategies are determined. The data corresponding to the one with the larger reward value between the first and second cooperative dispensing strategies are used to construct the training samples.

[0064] The system automatically selects policy data with higher reward values ​​to construct training samples, ensuring the model continuously learns towards better solutions. The training set always consists of experience from policies that perform best in the current environment, improving learning efficiency by over 30%. Only complete data chains of high-quality policies are retained, reducing the burden of storing and processing invalid data by up to 40%. Each generation of the model undergoes empirical screening, ensuring a stable upward trend in overall system performance and improving the long-term optimization goal achievement rate by 50%. The overall effectiveness of the policy can be evaluated through comparative testing on limited road partitions, reducing validation costs by 60%.

[0065] The flowcharts provided in this embodiment are not intended to indicate that the operations of the method will be performed in any particular order, or that all operations of the method are included in every case. Furthermore, the method may include additional operations. Within the scope of the technical concept provided by the method in this embodiment, additional variations can be made to the above method.

[0066] It should be understood that in some embodiments, the components may be implemented using hardware, software, firmware, or a combination thereof. In the above embodiments, multiple steps or methods may be implemented using software or firmware stored in memory and executed by a suitable instruction execution system.

[0067] This invention also provides a precision de-icing agent application system based on digital twins and swarm intelligence, comprising: Intelligent agent cluster, which includes intelligent spreader vehicles.

[0068] The status acquisition device is configured to acquire the status of the target road.

[0069] The network information acquisition module is configured to acquire weather forecast data via the network.

[0070] A digital twin, containing a model of the target road.

[0071] The module for predicting trends in ice and snow conditions includes a model for predicting trends in ice and snow conditions.

[0072] The effect acquisition device is configured to collect data on the snow and ice melting effect and environmental impact after the spreading operation.

[0073] The multi-agent reinforcement learning module includes the models involved in multi-agent reinforcement learning algorithms.

[0074] The de-icing agent precision application system based on digital twins and swarm intelligence in this invention embodiment is used to implement the steps of the de-icing agent precision application method based on digital twins and swarm intelligence in any of the above embodiments.

[0075] Therefore, those skilled in the art should recognize that although numerous exemplary embodiments of the present invention have been shown and described in detail herein, many other variations or modifications conforming to the principles of the present invention can be directly determined or derived from the disclosure of the present invention without departing from the spirit and scope of the invention. Thus, the scope of the present invention should be understood and construed as covering all such other variations or modifications.

Claims

1. A method for precise application of de-icing agents based on digital twins and swarm intelligence, characterized in that, include: Construct a digital twin containing the target road model; A snow and ice condition change trend prediction model is trained and configured to obtain the snow and ice condition change trend of the target road within a future preset time period based on the condition of the target road and meteorological forecast data. The condition of the target road includes the parameters of the target road, the weather conditions of the target road, the snow and ice condition of the target road, and the vehicle condition of the target road. The parameters of the target road include ordinary road sections, bridge surfaces, and intersections. When the target road needs to melt ice and snow, the state of the target road is collected to keep the state of the model synchronized with the state of the target road, and the ice and snow state change trend prediction model outputs the ice and snow layer change trend of the target road in the future preset time period based on the state of the target road. The schedulable spreading vehicles are modeled as a cluster of intelligent agents. Taking the type of snow melting agent, the state of the target road, and the trend of the snow and ice layer change as inputs, the digital twin uses the global snow and ice melting effect, resource consumption, traffic impact, and environmental impact as comprehensive optimization objectives. Multi-agent reinforcement learning algorithm is used to simulate scheduling and generate the optimal cooperative spreading strategy for the cluster of intelligent agents. The optimal coordinated spreading strategy includes at least the operating path of each spreading vehicle, the start and stop time of spreading, the spreading dosage and speed of the de-icing agent, and the intervention time of mechanical snow removal. The optimal collaborative spreading strategy is distributed to each spreading vehicle for execution; each spreading vehicle, as an edge computing node, determines the sudden road conditions based on the state of the target road it has obtained, and performs autonomous coordination and dynamic task adjustment for the sudden road conditions, based on the execution of the optimal collaborative spreading strategy. Data on snow and ice melting effects and environmental impacts after the spreading operation are collected; and the optimal collaborative spreading strategy, the execution process of each spreading vehicle, snow and ice melting effect data, environmental impact data, and the state of the target road are constructed as training samples to update the snow and ice state change trend prediction model and the models involved in the multi-agent reinforcement learning algorithm.

2. The method for precise application of de-icing agents based on digital twins and swarm intelligence according to claim 1, characterized in that, The environmental impact data is obtained by periodically collecting and analyzing the degree of corrosion of the target road, as well as the salt ion concentration, water conductivity and vegetation spectral characteristics of soil samples along the target road. The sudden road conditions mentioned include rapid icing caused by a sudden drop in local road surface temperature and interruption of the work route due to traffic congestion; the dynamic adjustments include the temporary exchange of tasks between the spreading vehicles, the adjustment of the spreading dosage, the adjustment of the travel speed, and the activation of backup routes.

3. The method for precise application of de-icing agents based on digital twins and swarm intelligence according to claim 1, characterized in that, Also includes: A target road condition prediction model is trained and configured to obtain the predicted state of the target road within a future preset time period based on meteorological forecast data and the historical state of vehicles on the target road; and to output the predicted state of the target road within the future preset time period using the meteorological forecast data as input. The ice and snow state change trend prediction model outputs the predicted change trend of the ice and snow layer on the target road within a future preset time period based on the predicted state of the target road. Using the type of snow melting agent, the predicted state of the target road, and the predicted trend of the snow and ice layer as inputs, the digital twin uses the global snow and ice melting effect, resource consumption, traffic impact, and environmental impact as comprehensive optimization objectives. It employs a multi-agent reinforcement learning algorithm to simulate scheduling and generate a predicted optimal cooperative spreading strategy for the agent cluster. When the target road needs to be melted for ice and snow, compare the degree of similarity between the current state of the target road and the predicted state of the target road. When the degree of similarity between the state of the target road and the predicted state of the target road is higher than a first degree threshold, the predicted optimal cooperative spreading strategy is used as the optimal cooperative spreading strategy to be executed.

4. The method for precise application of de-icing agents based on digital twins and swarm intelligence according to claim 1, characterized in that, Also includes: A target road condition prediction model is trained and configured to obtain the predicted state of the target road within a future preset time period based on meteorological forecast data and the historical state of vehicles on the target road; and to output the predicted state of the target road within the future preset time period using the meteorological forecast data as input. The ice and snow state change trend prediction model outputs the predicted change trend of the ice and snow layer on the target road within a future preset time period based on the predicted state of the target road. Using the type of snow melting agent, the predicted state of the target road, and the predicted trend of the snow and ice layer as inputs, the digital twin uses the global snow and ice melting effect, resource consumption, traffic impact, and environmental impact as comprehensive optimization objectives. It employs a multi-agent reinforcement learning algorithm for simulation scheduling to generate a predicted optimal cooperative pre-distribution strategy for the agent cluster. The status of the target road is collected at preset time intervals, and it is determined whether ice and snow melting is needed based on the status of the target road. If not, compare the degree of similarity between the state of the target road and the predicted state of the target road; When the number of times the proximity between the state of the target road and the predicted state of the target road exceeds the second threshold reaches the first preset number, the predicted optimal cooperative pre-spreading strategy is sent to each spreading vehicle for execution. When the number of times the proximity between the state of the target road and the predicted state of the target road is lower than the second threshold reaches a second preset number, the system returns to use the meteorological forecast data as input, so that the target road state prediction model outputs the predicted state of the target road within a preset future time period.

5. The method for precise application of de-icing agents based on digital twins and swarm intelligence according to claim 1, characterized in that, Also includes: Obtain the snow and ice layer change trend on a reference road adjacent to the target road that has not been melted, and record it as the snow and ice layer reference change trend; Determine the degree of similarity between the snow and ice layer change trend output by the snow and ice state change trend prediction model and the reference snow and ice layer change trend; When the number of times the snow and ice layer change trend output by the snow and ice state change trend prediction model is less than the third degree threshold reaches a third preset number, each spreading vehicle stops executing the optimal cooperative pre-spreading strategy and re-enters the process of using the type of snow melting agent, the predicted state of the target road, and the predicted snow and ice layer change trend as inputs. In the digital twin, with the global snow and ice melting effect, resource consumption, traffic impact, and environmental impact as comprehensive optimization objectives, a multi-agent reinforcement learning algorithm is used for simulation scheduling to generate a predicted optimal cooperative spreading strategy for the agent cluster.

6. The method for precise application of de-icing agents based on digital twins and swarm intelligence according to claim 1, characterized in that, Each of the spreading vehicles is an intelligent agent, and the state vector corresponding to each spreading vehicle includes position, load dose, equipment status, current task progress, and the state of the target road at the corresponding position; The motion vectors corresponding to each of the spreading vehicles include the work path, actual spreading start and stop time, snow melting agent spreading dosage and travel speed, and mechanical snow removal intervention time; The reward function for each of the aforementioned spreading vehicles is composed of a weighted average of snow and ice melting effects, resource consumption, traffic impact, and environmental impact.

7. The method for precise application of de-icing agents based on digital twins and swarm intelligence according to claim 1, characterized in that, The intelligent agent cluster also includes: The robot, as an intelligent agent, is used to spread de-icing agents on sidewalks, bike lanes, and bridge piers; Drones, as intelligent agents, are used for spreading de-icing agents in emergency areas; Snowplows, as intelligent agents, are used for mechanical snow removal; The road condition sensing vehicle, as an intelligent agent, is used to acquire the state of the target road before or after the application of de-icing agent.

8. The method for precise application of de-icing agents based on digital twins and swarm intelligence according to claim 1, characterized in that, Also includes: Generate multiple optional cooperative dispensing strategies for the agent cluster, and output the advantages of each optional cooperative dispensing strategy and the optimal cooperative dispensing strategy, as well as the differences between them; Determine whether the policy selection information selected by the scheduler has been received; If so, the optional collaborative spreading strategy or the optimal collaborative spreading strategy corresponding to the strategy selection information is sent to each spreading vehicle for execution; If the strategy selection information is not received within the preset time, the optimal coordinated spreading strategy will be sent to each spreading vehicle for execution. Record the dispatcher's multiple choices and learn their preferences; The strategy that is closest to the preference among the optional cooperative dispensing strategies, as well as the optimal cooperative dispensing strategy, are highlighted.

9. The method for precise application of de-icing agents based on digital twins and swarm intelligence according to claim 1, characterized in that, Also includes: The target road is divided into a test area and a control area; A first cooperative dispensing strategy is executed in the test area, and a second cooperative dispensing strategy is executed in the control area; The first collaborative distributing strategy is the optimal collaborative distributing strategy output before the model update of the ice and snow state change trend prediction model and the multi-agent reinforcement learning algorithm; the second collaborative distributing strategy is the optimal collaborative distributing strategy output before the model update of the ice and snow state change trend prediction model and the multi-agent reinforcement learning algorithm. Collect snow and ice melting effect data, resource consumption data, traffic impact data, and environmental impact data for the test area and the control area; Based on the snow and ice melting effect data, resource consumption data, traffic impact data, and environmental impact data, the reward values ​​of the first coordinated spreading strategy and the second coordinated spreading strategy are determined. The data corresponding to the larger reward value of the first cooperative dispensing strategy and the second cooperative dispensing strategy are used to construct training samples.

10. A precision de-icing agent application system based on digital twins and swarm intelligence, characterized in that, include: A cluster of intelligent agents, including an intelligent spreader vehicle; A status acquisition device is configured to acquire the status of the target road; The network information acquisition module is configured to acquire weather forecast data via the network. A digital twin, containing a model of the target road; The module for predicting trends in ice and snow conditions includes a model for predicting trends in ice and snow conditions. The effect acquisition device is configured to collect data on the snow and ice melting effect and environmental impact after the spreading operation. The multi-agent reinforcement learning module includes the models involved in the multi-agent reinforcement learning algorithm; The de-icing agent precision application system based on digital twins and swarm intelligence is used to implement the steps of the de-icing agent precision application method based on digital twins and swarm intelligence as described in any one of claims 1 to 9.

Images