Mountain tourism route planning method based on weather and natural disasters

Abstract

The invention discloses a mountain tourism route planning method based on weather and natural disasters, and the method comprises the steps: 1, building a Markov game model based on POMDP, combining Mama multi-modal feature compression and RAG semantic enhancement, designing a dynamic risk matrix integrating geological disaster probability, weather trend and ecological sensitivity, and providing partial observable environment state perception; 2, establishing a Nash equilibrium constraint optimization model, sharing an extreme weather strategy through federal learning, introducing an NSGA-II-Actor-Critic architecture to optimize unmanned aerial vehicle inspection and path adjustment, and triggering dynamic communication when rainfall exceeds a threshold value; 3, fusing a security and experience income design reward function, processing continuous action planning by using improved DDPG, and accelerating multi-target Pareto frontier convergence in combination with NSGA-II; 4, constructing a user weather sensitivity portrait, and dynamically adjusting weight balance income and AR simulation preview preference; and 5, based on a GIS closed-loop decision engine, performing agent and environment feature complementation, recording an emergency log by a block chain, and constructing a disaster case library by meta-learning.

Description

TECHNICAL FIELD

[0001] The application relates to the technical field of route planning, in particular to a mountain tourism route planning method based on weather and natural disasters. BACKGROUND

[0002] In recent years, mountain tourism has rapidly risen as an important growth pole in the global tourism market due to its unique natural landscapes and adventure experiences. However, the complexity of mountain environments, such as steep terrain, variable weather, and ecological fragility, poses serious challenges to tourist safety.

[0003] Over time, the problem of insufficient dynamic risk perception in traditional route planning methods has become increasingly prominent. These methods rely heavily on static geological maps and historical weather data, lacking real-time perception capabilities for geological disaster probabilities, sudden weather changes, and ecological sensitivities, resulting in delayed decision-making. For example, in 2023, three tourists were stranded in a dangerous area for six hours after a landslide blocked their path in a scenic area of a mountainous region in Sichuan during continuous rainfall, as the traditional system failed to timely monitor the expansion of local mountain cracks. Therefore, a mountain tourism route planning method based on weather and natural disasters is proposed, which combines a dynamic risk matrix with multi-agent game decision-making in a closed-loop framework, and introduces a dynamic weight adjustment mechanism driven by user portraits to improve dynamic risk perception and accurately balance the safety and experience of mountain tourism routes in real time. SUMMARY

[0004] The purpose of the application is to solve the problems in the prior art and provide a mountain tourism route planning method based on weather and natural disasters.

[0005] The core idea of the application is as follows: first, a dynamic risk assessment matrix is constructed by integrating multi-modal perception data and semantic enhancement technology, solving the problem of static environment model perception lag; second, a collaborative decision-making mechanism based on Nash equilibrium constraints and NSGA-II multi-objective optimization is designed to ensure the decision-making efficiency and stability of the multi-agent system in complex environments; third, a dynamic weight adjustment algorithm driven by user portraits is introduced to objectively quantify subjective preferences and achieve accurate balance between safety and experience; finally, a closed-loop decision-making and evolution framework is constructed through GIS, blockchain, and meta-learning technology to improve the autonomy and adaptability of the system.

[0006] To achieve the above purpose, the application adopts the following technical solutions: A mountain tourism route planning method based on weather and natural disasters, comprising: Step one (dynamic risk assessment and state perception module): Based on the POMDP model, a Markov game decision model for multi-agent collaborative optimization is constructed. The model integrates traffic along the line and natural disaster monitoring data of tourist destinations (such as landslide displacement sensor, rainfall radar) and ecological environment indicators (vegetation coverage, soil moisture), combines Mamba multi-modal feature representation learning method, and processes through multi-modal feature compression and RAG semantic enhancement to construct a dynamic risk matrix (output G, W, E, R). It provides state perception basis for multi-agent system in partially observable environment.

[0007] Step two (multi-agent collaborative optimization module): Through federated learning and NSGA-II algorithm, the unmanned aerial vehicle inspection strategy is optimized, and an optimization target model meeting the Nash equilibrium constraint is established. A special agent is designed for weather perception needs, and a lightweight YOLOv8 model is deployed to identify weather phenomena. Meanwhile, a communication and feature sharing mechanism among multiple agents is proposed. The NSGA-II algorithm is introduced into the Actor-Critic architecture to optimize the unmanned aerial vehicle inspection frequency and path adjustment strategy, ensuring that the dynamic communication protocol is automatically triggered when the rainfall intensity exceeds the threshold, and improving the collaborative perception efficiency in high-dimensional state space.

[0008] Step three (path planning decision module): A reward function that combines safety and experience benefits is designed, and an improved DDPG algorithm is used to output the final path instructions (θ, v, α). This module directly supports the claims of "improved DDPG for continuous action planning" and "reward function design combining safety and experience benefits".

[0009] Step four (user portrait and adjustment module): Based on user historical behavior data, a portrait is constructed, and dynamic weight adjustment and AR simulation preview are performed. The environment and user feedback loop in the figure embodies the features of "dynamic adjustment of weight balance benefits" and "AR simulation preview preference". When the user selects a specific mode, the corresponding benefit item weight is automatically increased, and the AR weather simulation function is developed to realize preference preview. Finally, through joint game, an "observation - behavior" mapping strategy is generated to strengthen geographic perception and recommendation ability.

[0010] Step five (closed-loop management and learning module): Based on the GIS geographic situation reasoning framework, a closed-loop decision engine with "perception input - cognitive output" capability is formed, and the feedback mechanism between agents and the environment is realized to achieve feature complementation. The blockchain technology is introduced to record emergency decision logs, and combined with the meta-learning framework, a disaster emergency case library is constructed, so that the system can quickly adapt to new climate regions. With the help of "reward - benefit" balance mechanism, spatio-temporal evolution intelligent reasoning is completed, and the robustness of group decision-making in continuous action space is improved, providing safer and more sustainable mountain tourism service support for tourists.

[0011] The technical solution further comprises: Further, the design dynamic risk assessment matrix comprises the following steps: Deploy geological monitoring sensor networks (crack displacement meters, pore water pressure gauges, etc.) to collect disaster-related data such as landslides and debris flows; access weather radar and satellite cloud data to obtain weather indicators such as rainfall intensity, visibility, and wind speed; through GIS geographic registration, unify different source data to the WGS84 coordinate system and perform spatial alignment; Use state space model to process time series data and build implicit representation of continuous time series Formula expression: Let the geological sensor sequence be Map to the implicit space through linear transformation W: Where, is the implicit state vector at time step t; is the linear transformation matrix of input data; is the linear transformation matrix of historical implicit states; is a bias vector with a dimension of 64; Finally output a 64-dimensional geological feature vector (original data dimension 1024→64); Build a disaster domain knowledge graph to map text reports (such as "local cracks in the mountain expand") to structured indicators; use LLM for event extraction, for example: Input text: "New cracks found on the northeast side of monitoring point A, width up to 3cm" Output structured data: {"crack displacement": 0.03, "location": "monitoring point A northeast side", "risk level": "high"}; Define matrix dimensions: Geological disaster probability (G): 0-100% (based on historical data and real-time sensor trends); Weather risk level (W): 1-5 levels (1 level safe, 5 level extreme weather); Ecological sensitivity (E): 0-100% (high sensitivity triggered when vegetation coverage <30%); Comprehensive risk value calculation formula: Trigger red alert when R≥75; Deploy lightweight ONNX inference engine on Raspberry Pi 4B device: Data preprocessing: 10ms / frame Feature compression: 8ms / frame Risk calculation: 2ms / frame The update cycle is strictly controlled within 5 minutes to ensure the timeliness of disaster response; the dimension weight is automatically adjusted according to seasonal changes; By compressing high-dimensional monitoring data into computable quantitative indicators and combining real-time semantic enhancement technology, a risk assessment system with dynamic adaptability is constructed, providing accurate environmental state input for subsequent multi-agent collaborative decision-making.

[0012] Further, the communication and feature sharing mechanism between multiple agents is constructed, and the NSGA-II algorithm is introduced into the Actor-Critic architecture, including the following steps: Based on the dynamic risk matrix output by step one, the communication threshold is set as: when the comprehensive risk value R>60, or the rainfall intensity >20mm / h, the trigger is triggered. When the trigger threshold is triggered, the agent automatically switches to high-frequency communication mode; a hybrid communication topology structure is adopted, which adopts distributed communication under normal conditions (to reduce the pressure of the center node), and automatically switches to centralized communication during disaster warning (to ensure the uniformity of instructions); JPEG2000 compression algorithm is used for high-definition images collected by unmanned aerial vehicles, which compresses the data size to 15% of the original size while maintaining 95% of the image quality;

[0013] Each agent (such as unmanned aerial vehicle, ground sensor) trains a lightweight YOLOv8 model on the local data set, and the output feature vector dimension is 256; the local data privacy is protected by differential privacy technology (adding noise variance σ=0.1), and the FedAvg algorithm is used to aggregate model parameters on the server side: Among them is the local model parameter of the kth agent, ∣ ∣ is the local data volume; The maximum mean difference (MMD) index is used to measure the consistency of feature distribution, ensuring that the difference in feature space after aggregation is less than 0.05; The visual features of unmanned aerial vehicles (256 dimensions), geological sensor features (64 dimensions), and meteorological data features (32 dimensions) are projected into a unified 128-dimensional space; self-attention mechanism is used to calculate the importance of each feature: Among them, is the attention weight of the i-th feature, is the energy score of the i-th feature, W is a learnable weight matrix, b is a bias term, is the transposed query vector, is the i-th feature vector, and the final fused feature is ; NSGA-II-Actor-Critic multi-objective optimization: Safety reward (G is the geological disaster probability); Efficiency reward (v is the actual inspection speed) Ecological reward (E is the ecological sensitivity) The Actor network generates an initial strategy π(s|θ), and then NSGA-II performs crossover and mutation operations on θ to generate a candidate strategy set, and then the Critic network evaluates the Q value of each strategy The optimal strategy is selected by non-dominated sorting and congestion calculation, and the Actor parameters are updated; Introduce constraint conditions when updating the strategy: Where, is the reward function of the ith agent; is the changed strategy of the ith agent; is the strategy set of other agents; so that any agent changing the strategy alone will not obtain higher rewards; The UAV adjusts the inspection path according to the fused features F, and when a new risk point is detected, it sends a priority interrupt signal to the ground agent through the V2X protocol; adopt cognitive radio technology, automatically switch between 2.4GHz / 5GHz frequency bands, avoid communication congestion (trigger switching when channel utilization> 85%).

[0014] Further, the design fuses the reward function of safety reward and experience reward, and processes the path planning problem in continuous action space through improved DDPG algorithm, including the following steps: Safety reward: Geological disaster risk: (G is the geological disaster probability output in step one); Weather risk: (W is the weather risk level); Ecological sensitivity: (E is the ecological sensitivity) Comprehensive safety reward: Experience reward: Landscape aesthetic value: score the landscape along the path such as terraces and cloud seas through the pre-trained CNN model; User preference matching degree: according to the weather sensitivity portrait generated in step four, calculate the similarity between the path and the user's preference, such as the preference weight of "rain scene shooting type" users for rain and fog landscape; Comprehensive experience reward: (A is the aesthetic score, S is the preference similarity) Total reward function: Where a is a dynamic weight coefficient, adjusted according to real-time risk value: When <0.7, a = 0.7 (forced priority safety) When ≥0.7, a = 0.3 (allow experience priority) The input of the Actor network is the dynamic risk matrix and the user preference vector output in step one, and the output is continuous action (path direction angle θ ∈ [0°, 360°) and speed v ∈ [0, 8m / s]); The Critic network uses double Q value estimation to reduce overestimation bias, and the input is state-action pairs and the output is Q value estimation; Adaptive Ornstein-Uhlenbeck noise is introduced, and the noise amplitude decays with the training round: (initial noise σ0=0.5, decay rate λ=0.001) According to the TD error, the sampling priority is calculated, and the samples with large error are given a 30% higher sampling probability; Store experience data of different agents in a shared replay pool to break data isolation; Generate a feasible region mask from DEM elevation data to prohibit paths from crossing dangerous areas such as cliffs; When the communication between the UAV and the ground agent is interrupted, automatically trigger the conservative path mode (reduce speed by 50% and limit direction angle to ±30°); When the comprehensive risk value R output in step one is greater than 75, set a = 0.7 forcibly, and prioritize path planning to avoid high-risk areas; When the user selects the "landscape priority" mode, a is dynamically adjusted to 0.3, allowing the path to detour to areas with high aesthetic scores, such as cloud-shrouded ridge lines; The mixed mode defaults to a = 0.5, balancing safety and experience; Deploy an asynchronous Actor-Learner architecture on 4 GPUs to improve training speed by 3 times; Pre-trained models complete initialization in a simulated mountain environment, reducing actual training time; Add a hard constraint layer after the Actor output to forcibly exclude all illegal actions that violate physical rules (such as air paths crossing mountains).

[0015] Further, the fusion of NSGA-II and Actor-Critic architecture establishes a multi-objective optimization model, accelerates the convergence of the Pareto frontier, and generates joint game decision results, including the following steps: 50% of the strategies are randomly generated by the Actor network, and 50% are generated by NSGA-II crossover and mutation, retaining the top 10% of high fitness strategies as the initial population; Output continuous action a (path direction angle θ and speed v) based on current state s; generate offspring strategies using SBX crossover (distribution index η = 15); explore new strategy space by applying polynomial mutation (mutation probability PM = 0.1); calculate Q value of each strategy: According to Q value and constraint conditions (such as path feasibility), divide Pareto front level; within the same front, select more diverse strategies by distance measurement; Update strategy parameters θ in the direction of Q value gradient using policy gradient method: Keep the top 20% elite strategies and combine them with newly generated strategies to form the next generation population; When the comprehensive risk value R ≥ 75, the safety benefit weight is increased by 30%; when the "landscape priority" request is detected, the experience benefit weight is increased by 20%; Under the federal learning framework, each agent shares the top 5% elite strategies and protects the strategy privacy through differential privacy (noise variance σ = 0.05); after adjusting the local fitness, the shared strategies participate in NSGA-II evolution again; Using repeated game mechanism, each agent proposes strategies in turn, and evaluates whether to reach equilibrium through Critic; when the benefit change rate of all agents is less than 2% after 5 consecutive strategy updates, it is determined that the game equilibrium is reached; Finally, generate a three-tuple decision including path direction angle θ, speed v, and risk aversion coefficient α: The decision satisfies both multi-objective optimization and Nash equilibrium constraints; Deploy asynchronous NSGA-II evolution on GPU cluster, and shorten the strategy evaluation time to 0.3 seconds per round; add a hard constraint layer after the Actor output to forcibly exclude all illegal actions that violate physical rules such as crossing mountains; dynamically adjust the multi-objective weight according to the real-time risk value and user preference.

[0016] Further, the user weather sensitivity profile is constructed, and the preference label is generated based on historical behavior data; a dynamic weight adjustment algorithm is used, including the following steps: Collect user historical behavior data, including: travel records (trip date, route selection, stay time), weather feedback (user actively labeled "comfortable / inconvenient" weather types), social media comments (such as "rainy day photo effect is amazing" and other weather-related descriptions); remove irrelevant information (such as advertisements, emojis) through text cleaning, and keep weather-related key expressions; User-initiated weather mode (e.g., "rain preference"), trip adjustment records (number of trips canceled due to rain), sentiment analysis of user reviews through NLP, such as "fog visibility is low, experience is poor" marked as negative sentiment; K-means algorithm is used to divide users into 3 categories: Rain-avoiding type: more than 70% of historical trips adjusted route due to rain; Rain-sky shooting type: actively choose rain trips and "rain" keyword frequency in comments is greater than 5 times; Climate adaptation type: no obvious weather preference, trip selection and weather correlation is less than 30%; After each user completes a trip, real-time feedback is collected through a questionnaire survey, and the portrait label is updated; when a user has not updated behavior data for 3 months, the label weight is reduced by 10% per month; Weight adjustment trigger conditions: User-initiated: when the user explicitly selects "rain preference" or "safety first" mode through the APP interface; Environmental change trigger: when the comprehensive risk value R≥75 output in step one, forced into safety first mode; Mixed mode: no explicit user instruction and environmental safety, use default balanced mode; The weight adjustment rule is that the safety benefit weight of rain-avoiding users is increased to 70% (original default 50%), and the experience benefit weight is reduced to 30%, and the path planning is prohibited to pass through areas with rainfall probability> 40%; The experience benefit weight of rain-sky shooting users is increased to 60% (original default 50%), and the safety benefit weight is reduced to 30%, and the path is allowed to detour to areas with rainfall probability> 50% (must meet ecological sensitivity< 50%); Climate adaptation type users maintain default weight (safety 50%, experience 50%), only when environmental risk value R≥80 is forced to adjust to safety mode; The weight adjustment constraint is that the safety benefit weight is not less than 30% (to ensure the basic safety bottom line), and the experience benefit weight is not more than 60% (to avoid excessive pursuit of experience and ignore risk), and the weight adjustment needs to pass through the Critic network to verify the feasibility of the strategy, such as whether the path is passable; After the user selects the preferred mode, the weather effect along the path is displayed in real time through AR technology: when "rain preference" mode is selected, virtual rainfall effect and possible shooting points are displayed, and when "safety first" mode is selected, geological disaster risk areas are highlighted, and users can adjust simulation parameters such as rainfall intensity and visibility through gesture interaction; Users can score the display effect in the simulation interface, and the score data is fed back to the portrait update module in real time to adjust the subsequent weight allocation strategy.

[0017] Further, the GIS-based geographic context reasoning framework forms a closed-loop decision engine, which complements features through feedback between agents and the environment, including the following steps: Integrate high-precision DEM (Digital Elevation Model) data, vegetation cover layer (NDVI index), water system distribution layer and historical geological disaster point data; Through GIS spatial analysis tools, generate a composite layer containing slope (marked as dangerous if greater than 30°), catchment area (area greater than 1km²), ecological red line area (prohibited development area); Superimpose the dynamic risk matrix (geology-weather-ecology three dimensions) of step one on the GIS layer to generate a real-time geographic context model; The model update cycle is 5 minutes, which is synchronized with the data refresh frequency of step one; Render the terrain using a three-dimensional GIS engine, highlighting high-risk areas (red), user preferred paths (blue), and the current location of the agent (green marker); Superimpose the virtual path onto the real environment through AR technology to support users to intuitively view the decision effect; Distribute the joint game decision (path direction angle θ, speed v, risk aversion coefficient α) generated in step three to the unmanned aerial vehicle and the ground agent; During the execution of the path by the agent, real-time environmental data such as new cracks and changes in rainfall intensity are collected; The environmental change data is returned to the decision engine through the V2X protocol to update the GIS geographic context model; According to the updated model, recalculate the path planning parameters to form a closed-loop adjustment; Dynamic adjustment trigger conditions: Environmental mutation: When the geological disaster probability G is greater than or equal to 80% or the weather risk level W is greater than or equal to 4, forcibly trigger path re-planning; User intervention: When the user manually adjusts the path preference (such as "avoid the current area") through the APP, immediately generate a new decision; Periodic optimization: Perform a light path optimization every 15 minutes to balance computing resources and decision timeliness; The aerial view of the unmanned aerial vehicle provides a high-altitude macroscopic view to identify large-scale geological changes (such as mountain crack expansion); Ground sensors provide microscopic data such as soil moisture and surface displacement; Weather station data supplements local weather changes such as sudden gusts and sudden visibility drops; Convert the orthographic image of the unmanned aerial vehicle and the ground sensor data to the same coordinate system through GIS coordinate conversion; Use PTP clock synchronization protocol to ensure that the timestamp error of multiple data sources is less than 10ms; According to the reliability of the data source, assign weights, such as laser radar data weight 0.8, visual data weight 0.6; The geological change area identified by the UAV needs to be confirmed by ground sensor data, such as when the crack displacement is greater than 5mm, triggering secondary verification; when a single data source shows an anomaly, such as a sudden temperature drop in a certain sensor but normal in surrounding sensors, self-checking is started and the data is marked as suspicious; The input data of each decision (environmental state), output action (path parameter), and execution result (user feedback) are packaged and chained; the data on the chain is stored in a Merkle tree structure to ensure data integrity and traceability; Extract historical decision cases from the blockchain and build a case library containing typical scenarios (such as "rainy mountain inspection" and "user landscape priority mode"); use the MAML (model-agnostic meta-learning) algorithm, which has the same structure as the Actor network in the improved DDPG algorithm. In the meta-training phase, the model learns to quickly adapt to different disaster scenarios (such as landslides under different rainfall and path adjustments under different wind speeds), and the loss function is the negative expectation of the cumulative reward of decisions in each task. Quickly adapt to new scenarios with a small number of samples, reducing online learning time; combine GIS spatio-temporal analysis tools to identify environmental change patterns, such as "30% increase in landslide probability after 5 consecutive days of rainfall"; encode the spatio-temporal rules into a rule engine to assist the decision engine in predicting environmental trends.

[0018] Further, the introduction of blockchain technology records emergency decision logs and combines a meta-learning framework to build a disaster emergency case library, including the following steps: Use environmental state snapshots to record the dynamic risk matrix (geological disaster probability G, weather risk level W, and ecological sensitivity E) at the time of decision-making and GIS geographic context models; decision action details include path direction angle θ, speed v, risk avoidance coefficient α, and multi-agent communication protocol state; execution result feedback includes user actual travel data (such as path completion, stop point selection), system monitored environmental changes such as new cracks and sudden rainfall intensity changes; user intervention records are user manual path preference adjustments (such as "avoid the current area") and system responses; Encapsulate the complete log of each decision as a blockchain transaction, add a timestamp and digital signature; use a Merkle tree structure to organize data, ensuring data integrity and fast verification capability; Permission management: Write permission: limited to the decision engine and user terminal (signed by private key); Read permission: open to authorized agencies (such as scenic area management departments) for post-analysis; Decision logs at any point in time can be queried through a blockchain browser, supporting multi-dimensional retrieval by user ID, time range, risk level, etc.; when a geological disaster occurs, the decision path and environmental state 5 minutes before the accident can be quickly located to assist in responsibility identification; Extract historical decision cases from the blockchain, and classify them by scene type: Typical disaster scenarios: such as "landslide warning after 7 consecutive days of rainfall", "typhoon path adjustment"; User preference scenarios: such as "rain scene shooting user itinerary", "safety first user risk avoidance"; Mixed scenarios: such as "user insists on landscape priority in high-risk environment"; Each case label includes key features such as environmental risk value R, user preference label, decision action sequence, and execution result score; Adopt MAML (Model-Agnostic Meta-Learning) algorithm to quickly adapt to new scenarios with few samples: Meta-training phase: simulate different disaster scenarios in the historical case library to train the model to learn "how to quickly learn"; Fine-tuning phase: when encountering a new scenario, such as the "hail + tourist dense" composite scenario, only 10-20 samples are needed to adjust the model parameters; Measure the similarity between new scenarios and case libraries by Euclidean distance, and directly call historical strategies when similarity is greater than 85%; When a new risk scenario is detected, retrieve the top 3 cases with the highest similarity from the case library to generate a candidate strategy set; evaluate the Q value of the candidate strategy through the Critic network, and select the optimal strategy to issue to the agent; after each emergency decision execution, add the new case to the case library, and eliminate obsolete cases that have not been accessed for more than 2 years; The decision log recorded in the blockchain provides real scenario data for meta-learning, and the strategy optimized by meta-learning is traceable and verifiable through the blockchain; when the meta-learning model generates a new strategy, the strategy version number and training sample set need to be recorded in the blockchain to ensure decision transparency; User behavior data such as path selection and preference labels are desensitized before being stored in the blockchain, such as user ID hashing; during meta-learning training, use a federated learning framework, each agent retains the original data locally, and only shares model parameter updates.

[0019] The present application has the following advantages: 8. Real-time and accurate perception: By using the dynamic risk matrix construction method based on Mamba multi-modal feature compression and RAG semantic enhancement, minute-level (within 5 minutes) update and fusion evaluation of geological disasters, weather changes and ecological sensitivity are realized, which greatly improves the timeliness and accuracy of risk perception compared with traditional methods relying on static maps.

[0020] 9. High-efficiency collaborative decision-making: Due to the proposed NSGA-II-Actor-Critic hybrid architecture and the communication mechanism with Nash equilibrium constraints, the unmanned aerial vehicle and the ground intelligent agent can not only efficiently approach the multi-objective Pareto frontier of safety, efficiency, ecology, etc. in path planning, but also avoid strategy conflicts between intelligent agents, ensuring the collaboration and stability of system decision-making.

[0021] 10. Personalized experience optimization: Due to the construction of user weather sensitivity portraits and the design of dynamic weight adjustment algorithms, the system can understand and quantify user subjective preferences, thereby generating a truly personalized travel route that meets user needs under the premise of ensuring basic safety, improving user experience.

[0022] 11. System autonomy and evolution: Due to the closed-loop decision engine based on GIS geographic context reasoning framework and the introduction of blockchain storage and meta-learning case library, the system has the ability to continuously learn and evolve from historical decisions, can adapt to new climate environments and disaster scenarios, and enhances the robustness and long-term effectiveness of the system. BRIEF DESCRIPTION OF DRAWINGS

[0023] Figure 1 A system working principle and flowchart of a mountain tourism route planning method based on weather and natural disasters is proposed. DETAILED DESCRIPTION

[0024] The technical solutions in the embodiments of the present application will be described in detail below with reference to the drawings in the embodiments of the present application. Obviously, the described embodiments are only a part of the embodiments of the present application, not all embodiments. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without creative labor are within the scope of protection of the present application.

[0025] Please refer to Figure 1 As shown in the system flowchart, the present application is a mountain tourism route planning method based on weather and natural disasters, which includes: Step one (dynamic risk assessment and state perception module): as Figure 1 As shown above, it describes the input of multi-source data (geology, meteorology, ecology). Based on the POMDP model, a Markov game decision model for multi-agent collaborative optimization is constructed. Through the fusion of natural disaster monitoring data along the traffic line and tourist destinations (such as landslide displacement sensor, rainfall radar) and ecological environment indicators (vegetation coverage, soil moisture), combined with the multi-modal feature representation learning method realized by Mamba, the process of constructing a dynamic risk matrix (output G, W, E, R) is finally realized through multi-modal feature compression and RAG semantic enhancement processing, providing state perception basis for multi-agent system in partially observable environment.

[0026] Step two (multi-agent collaborative optimization module): as Figure 1 The upper left shows that it describes the triggering of the communication protocol based on the risk matrix, and optimizes the UAV inspection strategy through federated learning and NSGA-II algorithm to establish an optimization target model that meets the Nash equilibrium constraint. And design a special agent for weather perception needs, deploy a lightweight YOLOv8 model to identify weather phenomena such as rain, fog, and hail. At the same time, a communication mechanism and feature sharing mechanism between multiple agents are proposed. In the Actor-Critic architecture, the NSGA-II algorithm is introduced to optimize the UAV inspection frequency and path adjustment strategy, ensuring that the dynamic communication protocol is automatically triggered when the rainfall intensity exceeds the threshold, and improving the collaborative perception efficiency in high-dimensional state space.

[0027] Step three (path planning decision module): as Figure 1 The middle part shows that after receiving the risk data and collaborative strategy, it designs a reward function that combines safety and experience benefits, and outputs the final path instruction (θ, v, α) through the improved DDPG algorithm. This module directly supports the claims of "improved DDPG for continuous action planning" and "reward function design combining safety and experience benefits".

[0028] Step four (user portrait and adjustment module): as Figure 1 The lower right shows that it describes the process of constructing a portrait based on user historical behavior data, dynamically adjusting weights, and simulating preview in AR. The environmental feedback and user feedback loop in the figure embodies the technical features of "dynamic adjustment of weight balance benefits" and "AR simulation preview preferences". When the user selects a specific mode, the corresponding benefit item weight is automatically increased, and the AR weather simulation function is developed to realize the preference preview. Finally, through joint game, the "observation-behavior" mapping strategy is generated, and the geographic perception and recommendation ability are strengthened.

[0029] Step five (closed-loop management and learning module): as Figure 1The lower left shows a description of the closed-loop process of path execution, environmental feedback, GIS context model update, blockchain record, and meta-learning case library construction. This module fully embodies the three core features of "GIS-based closed-loop decision engine", "blockchain record emergency log" and "meta-learning construction of disaster case library". Among them, based on the GIS geographic context reasoning framework, a closed-loop decision engine with "perception input-cognition output" capability is formed, and the feedback mechanism between the agent and the environment is realized to achieve complementary characteristics; Introducing blockchain technology to record emergency decision logs, and combining meta-learning framework to build disaster emergency case library, so that the system can quickly adapt to new climate regions; With the help of the "reward-revenue" balance mechanism, the spatio-temporal evolution of intelligent reasoning is completed, and the robustness of group decision-making under continuous action space is improved, providing safer and more sustainable mountain tourism service support for tourists.

[0030] In one embodiment, the design of the dynamic risk assessment matrix includes the following steps: Deploy geological monitoring sensor networks (crack displacement meters, pore water pressure meters, etc.) to collect disaster-related data such as landslides and debris flows; access meteorological radar and satellite cloud data to obtain weather indicators such as rainfall intensity, visibility, and wind speed; through GIS geographic registration, unify different source data to WGS84 coordinate system, and perform spatial alignment; Use state space model to process time series data and construct continuous time series implicit representation Formula expression: Let the geological sensor sequence be Map to the hidden space through linear transformation W: Where, is the hidden state vector at time step t; is the linear transformation matrix of input data; is the linear transformation matrix of historical hidden states; is a bias vector with a dimension of 64; Finally output a 64-dimensional geological feature vector (original data dimension 1024→64); Build a disaster domain knowledge graph to map text reports (such as "local cracks in the mountain expand") to structured indicators; use LLM for event extraction, for example: Input text: "New cracks found on the northeast side of monitoring point A, width up to 3 cm" Output structured data: {"crack displacement": 0.03, "location": "monitoring point A northeast side", "risk level": "high"}; Define matrix dimensions: Geological disaster probability (G): 0-100% (based on historical data and real-time sensor trends); Weather risk level (W): 1-5 levels (1 level safe, 5 level extreme weather); Ecological sensitivity (E): 0-100% (high sensitivity triggered when vegetation coverage <30%); Comprehensive risk value calculation formula: Trigger red alert when R≥75; Deploy lightweight ONNX inference engine on Raspberry Pi 4B device: Data preprocessing: 10ms / frame Feature compression: 8ms / frame Risk calculation: 2ms / frame The update cycle is strictly controlled within 5 minutes to ensure the timeliness of disaster response; adjust the dimension weight automatically according to the seasonal change; By compressing high-dimensional monitoring data into quantifiable indicators and combining real-time semantic enhancement technology, a risk assessment system with dynamic adaptability is constructed, providing accurate environmental state input for subsequent multi-agent collaborative decision-making.

[0031] In one embodiment, the communication and feature sharing mechanism between multiple agents is constructed, and the NSGA-II algorithm is introduced into the Actor-Critic architecture, including the following steps: Based on the dynamic risk matrix output by step one, the communication threshold is set as: when the comprehensive risk value R>60, or the rainfall intensity >20mm / h, trigger. When the trigger threshold is reached, the agent automatically switches to high-frequency communication mode; adopt hybrid communication topology structure, under normal state, adopt distributed communication (reduce the pressure of central node), in disaster warning, automatically switch to centralized communication (ensure the uniformity of instructions); for high-definition images collected by unmanned aerial vehicles, adopt JPEG2000 compression algorithm, under the premise of maintaining 95% image quality, compress the data size to 15% of the original size;

[0032] Each agent (such as unmanned aerial vehicle, ground sensor) trains a lightweight YOLOv8 model on the local data set, and the output feature vector dimension is 256; protect local data privacy through differential privacy technology (add noise variance σ=0.1), and use FedAvg algorithm to aggregate model parameters on the server side: Where is the local model parameter of the kth agent, and is the local data volume; Use the maximum mean difference (MMD) index to measure the consistency of feature distribution, ensure that the difference in feature space after aggregation is less than 0.05; The UAV visual features (256 dimensions), geological sensor features (64 dimensions) and meteorological data features (32 dimensions) are projected into a unified 128-dimensional space; the importance of each feature is calculated using a self-attention mechanism: wherein, is the attention weight of the i-th feature, is the energy score of the i-th feature, W is a learnable weight matrix, and b is a bias term, is the transposed query vector, is the i-th feature vector, and the final fused feature is ; NSGA-II-Actor-Critic multi-objective optimization: Safety benefit (G is the probability of geological disaster); Efficiency benefit (v is the actual inspection speed) Ecological benefit (E is the ecological sensitivity) The Actor network generates an initial strategy π(s|θ), and then the NSGA-II performs crossover and mutation operations on θ to generate a candidate strategy set, and then the Critic network evaluates the Q value of each strategy The winning strategy is selected through non-dominated sorting and congestion calculation, and the Actor parameters are updated; When updating the strategy, introduce the constraint condition: wherein, is the i-th agent's benefit function; is the changed strategy of the i-th agent; is the strategy set of other agents; so that any agent changing the strategy alone will not get higher benefits; The UAV adjusts the inspection path according to the fused features F, and when a new risk point is detected, it sends a priority interrupt signal to the ground agent through the V2X protocol; adopt cognitive radio technology, automatically switch between 2.4GHz / 5GHz frequency bands, avoid communication congestion (trigger switching when channel utilization rate > 85%).

[0033] In one embodiment, the design combines safety benefits and experience benefits into a reward function, and uses an improved DDPG algorithm to handle path planning problems in continuous action space, including the following steps: Safety benefit: Geological disaster risk: (G is the geological disaster probability output by step one); Weather risk: (W is the weather risk level); Ecological sensitivity: (E is the ecological sensitivity) Comprehensive safety benefit: Experience benefit: Landscape aesthetic value: score the landscape along the path such as terraces and cloud sea through a pre-trained CNN model; User preference matching degree: calculate the similarity between the path and user preference based on the weather sensitivity portrait generated in step four, such as the preference of "rain scene shooting type" users for rain and fog landscape; Comprehensive experience benefit: (A is the aesthetic score, S is the preference similarity) Total reward function: Where α is a dynamic weight coefficient, adjusted according to real-time risk value: When <0.7, α=0.7 (forced priority safety) When ≥0.7, α=0.3 (allow experience priority) The input of the Actor network is the dynamic risk matrix output by step one and the user preference vector, and the output is continuous action (path direction angle θ ∈ [0°, 360°) and speed v ∈ [0, 8m / s]); The Critic network uses double Q value estimation to reduce overestimation bias, and the input is state-action pair and the output is Q value estimation; Adaptive Ornstein-Uhlenbeck noise is introduced, and the noise amplitude decays with training rounds: (initial noise σ0=0.5, decay rate λ=0.001) According to the TD error, the sampling priority is calculated, and the samples with large error are given a 30% probability of being sampled; Store experience data of different agents in a shared replay pool to break data isolation; Generate feasible region mask from DEM elevation data to prohibit path from crossing dangerous areas such as cliffs; When the communication between the UAV and the ground agent is interrupted, automatically trigger the conservative path mode (reduce speed by 50% and limit direction angle to ±30°); When the comprehensive risk value R output by step one is greater than 75, set α=0.7 forcibly, and prioritize path planning to avoid high-risk areas; When the user selects "landscape priority" mode, dynamically adjust α to 0.3 to allow the path to detour to areas with high aesthetic score such as cloud-wrapped ridge lines; The mixed mode defaults to α=0.5, balancing safety and experience; The asynchronous Actor-Learner architecture is deployed on 4 GPUs to improve training speed by 3 times; the pre-trained model is initialized in a simulated mountain environment to reduce actual training time; a hard constraint layer is added after the Actor output to force the exclusion of all illegal actions that violate physical rules (such as path traversal through mountains in the air).

[0034] In one embodiment, the fusion of NSGA-II and Actor-Critic architecture establishes a multi-objective optimization model, accelerates the convergence of the Pareto frontier, and generates joint game decision results, including the following steps: 50% of the strategies are randomly generated by the Actor network, and 50% are generated by NSGA-II crossover and mutation, with the top 10% of high fitness strategies reserved as the initial population; Based on the current state s, output continuous actions a (path direction angle θ and speed v); use SBX crossover (distribution index η = 15) to generate offspring strategies; apply polynomial mutation (mutation probability PM = 0.1) to explore new strategy space; calculate the Q value of each strategy: , according to the Q value and the constraint conditions (such as path feasibility), divide the Pareto frontier level; within the same frontier, select the strategy with better diversity through distance measurement; Use the policy gradient method to update the strategy parameters θ along the Q value gradient direction: Keep the top 20% of elite strategies and combine them with newly generated strategies to form the next generation population; When the comprehensive risk value R ≥ 75, the safety benefit weight is increased by 30%; when the "landscape priority" request is detected, the experience benefit weight is increased by 20%; Under the federated learning framework, each agent shares the top 5% of elite strategies and protects the strategy privacy through differential privacy (noise variance σ = 0.05); the shared strategies are adjusted for local fitness and then participate in NSGA-II evolution again; Use the repeated game mechanism, each agent proposes a strategy in turn, and through the Critic to evaluate whether the equilibrium is reached; when the strategy is updated for 5 consecutive rounds, and the benefit change rate of all agents is less than 2%, it is determined that the game equilibrium is reached; Finally, a three-tuple decision is generated, including path direction angle θ, speed v, and risk aversion coefficient α: The decision satisfies the multi-objective optimization and Nash equilibrium constraints simultaneously; Deploying asynchronous NSGA-II evolution on GPU clusters, each round of strategy evaluation time is shortened to 0.3 seconds; After the output of the Actor, a hard constraint layer is added to force the exclusion of all illegal actions that violate physical rules such as crossing mountains; According to the real-time risk value and user preference, the multi-objective weight is dynamically adjusted.

[0035] In one embodiment, the construction of the user weather sensitivity portrait generates preference labels in combination with historical behavior data; A dynamic weight adjustment algorithm is used, including the following steps: Collecting user historical behavior data, including: travel records (trip date, route selection, stay time), weather feedback (user-initiated "comfort / inconvenience" weather type labeling), social media comments (such as "rainy day photo effect is perfect" and other weather-related descriptions); Remove irrelevant information (such as advertisements, emojis) and retain weather-related key expressions through text cleaning; User-initiated weather mode (such as "rain priority"), trip adjustment records (number of times trips were canceled due to rain), sentiment analysis of user comments through NLP, such as "foggy weather with low visibility, poor experience" marked as negative sentiment; Users are divided into 3 categories using K-means algorithm: Rain-avoiding type: more than 70% of historical trips were adjusted due to rain; Rain-photography type: actively choose rainy day trips and "rain scene" keyword frequency in comments is greater than 5 times; Climate-adapted type: no obvious weather preference, trip selection has less than 30% correlation with weather; After each user completes a trip, real-time feedback is collected through a questionnaire survey to update the portrait label; When a user has not updated behavior data for 3 months, the label weight is reduced by 10% per month; Weight adjustment trigger conditions: User-initiated selection: when the user explicitly selects "rain priority" or "safety first" mode through the APP interface; Environmental change trigger: when the comprehensive risk value R≥75 output by step one, forced into safety first mode; Mixed mode: no explicit user instruction and environmental safety, use the default balanced mode; The weight adjustment rule is that the safety benefit weight of rain-avoiding users is increased to 70% (original default 50%), and the experience benefit weight is reduced to 30%, and the path planning is prohibited to pass through areas with rainfall probability >40%; The experience benefit weight of rain-photography users is increased to 60% (original default 50%), and the safety benefit weight is reduced to 30%, allowing the path to detour to areas with rainfall probability >50% (must meet ecological sensitivity <50%); Climate-adaptive users maintain default weights (safety 50%, experience 50%) and are forced to adjust to safety mode only when environmental risk value R≥80; Weight adjustment constraints are that the safety benefit weight is not less than 30% (to ensure a basic safety bottom line) and the experience benefit weight is not more than 60% (to avoid excessive pursuit of experience and neglect of risk), and after weight adjustment, the feasibility of the strategy needs to be verified by the Critic network, such as whether the path is passable; After the user selects the preferred mode, the AR technology is used to display the weather effects along the path in real time: when the "rain scene priority" mode is selected, virtual rainfall effects and possible shooting points are displayed, and when the "safety priority" mode is selected, geological disaster risk areas are highlighted, and the user can adjust simulation parameters such as rainfall intensity and visibility through gesture interaction; The user can score the display effects in the simulation interface, and the scoring data is fed back to the portrait update module in real time to adjust the subsequent weight allocation strategy.

[0036] In one embodiment, the GIS-based geographic context reasoning framework forms a closed-loop decision engine that complements features through feedback between the agent and the environment, including the following steps: Integrate high-precision DEM (Digital Elevation Model) data, vegetation coverage layers (NDVI index), water system distribution layers, and historical geological disaster point data; use GIS spatial analysis tools to generate a composite layer containing slope (marked as dangerous if greater than 30°), catchment area (greater than 1km²), and ecological red line area (prohibited development area); Overlay the dynamic risk matrix (geological-weather-ecological three dimensions) of step one with the GIS layer to generate a real-time geographic context model; the model update cycle is 5 minutes, synchronized with the data refresh frequency of step one; Use a three-dimensional GIS engine to render the terrain, highlight high-risk areas (red), user preferred paths (blue), and the current location of the agent (green marker); use AR technology to superimpose the virtual path onto the real environment to allow the user to intuitively view the decision effect; Distribute the joint game decision (path direction angle θ, speed v, risk aversion coefficient α) generated in step three to the drone and the ground agent; during the execution of the path by the agent, real-time environmental data such as new cracks and changes in rainfall intensity are collected; the environmental change data is transmitted back to the decision engine through the V2X protocol to update the GIS geographic context model; According to the updated model, recalculate the path planning parameters to form a closed-loop adjustment; Dynamic adjustment trigger conditions: Environmental mutation: when the geological disaster probability G≥80% or the weather risk level W≥4 is detected, the path re-planning is forcibly triggered; User intervention: When the user manually adjusts the path preference (such as "avoid the current area") through the APP, a new decision is immediately generated; Periodic optimization: A light-weight path optimization is automatically performed every 15 minutes to balance the computing resources and decision timeliness; The UAV perspective provides a high-altitude macroscopic view to identify large-scale geological changes (such as mountain crack expansion); ground sensors provide microscopic data such as soil moisture and surface displacement; weather station data complements local weather changes such as sudden gusts and sudden visibility drops; Through GIS coordinate conversion, the UAV orthographic image and ground sensor data are unified into the same coordinate system; PTP clock synchronization protocol is adopted to ensure that the timestamp error of multiple data sources is less than 10ms; according to the reliability of the data source, the weight is allocated, such as laser radar data weight 0.8, visual data weight 0.6; The geological change area identified by the UAV needs to be confirmed by ground sensor data, such as when the crack displacement is greater than 5mm, secondary verification is triggered; when a single data source shows an anomaly, such as a sudden drop in sensor temperature but the surrounding sensors are normal, self-checking is started and the data is marked as suspicious; The input data (environmental state), output action (path parameters) and execution result (user feedback) of each decision are packaged and chained; the on-chain data is stored in a Merkle tree structure to ensure data integrity and traceability; Extract historical decision cases from the blockchain to build a case library containing typical scenarios (such as "rainy season mountain inspection" and "user landscape priority mode"); use MAML (model-agnostic meta-learning) algorithm to quickly adapt to new scenarios with a small number of samples, reducing online learning time; combine GIS spatio-temporal analysis tools to identify environmental change patterns, such as "landslide probability increases by 30% after 5 consecutive days of rainfall"; encode the spatio-temporal rules into a rule engine to assist the decision engine in predicting environmental trends.

[0037] In one embodiment, the introduction of blockchain technology records the emergency decision log, and the meta-learning framework builds a disaster emergency case library, including the following steps: Use environmental state snapshots to record the dynamic risk matrix (geological disaster probability G, weather risk level W, ecological sensitivity E) and GIS geographic context model at the time of decision; decision action details include path direction angle θ, speed v, risk avoidance coefficient α and multi-agent communication protocol state; execution result feedback includes user actual travel data (such as path completion, stop point selection), system monitored environmental changes such as new cracks and sudden rainfall intensity changes; user intervention records the user's manual adjustment of path preference (such as "avoid the current area") and the system's response; The complete log of each decision is encapsulated as a blockchain transaction, with a timestamp and digital signature; the data is organized using a Merkle tree structure to ensure data integrity and fast verification capabilities; Permission management: Write permission: limited to the decision engine and user terminals (signed by private key); Read permission: open to authorized agencies (such as scenic area management departments) for post-analysis; The decision log at any point in time can be queried through the blockchain browser, supporting multi-dimensional retrieval by user ID, time range, risk level, etc. When a geological disaster occurs, the decision path and environmental state within the previous 5 minutes can be quickly located to assist in responsibility identification; Extract historical decision cases from the blockchain and classify them by scenario type: Typical disaster scenarios: such as "landslide warning after 7 consecutive days of rainfall" and "typhoon path adjustment"; User preference scenarios: such as "rain scene shooting user itinerary" and "safety first user risk avoidance"; Mixed scenarios: such as "user insisting on landscape priority in high-risk environment"; Each case is labeled with key features including environmental risk value R, user preference label, decision action sequence, and execution result score; Use the MAML (Model-Agnostic Meta-Learning) algorithm to quickly adapt to new scenarios with a small number of samples: Meta-training phase: simulate different disaster scenarios in the historical case library to train the model to learn "how to quickly learn"; Fine-tuning phase: when encountering a new scenario, such as the "hail + tourist concentration" composite scenario for the first time, only 10-20 samples are needed to adjust the model parameters; Measure the similarity between the new scenario and the case library using the Euclidean distance. When the similarity is greater than 85%, directly call the historical strategy; When a new risk scenario is detected, retrieve the top 3 cases with the highest similarity from the case library to generate a candidate strategy set. Evaluate the Q value of the candidate strategies using the Critic network and select the optimal strategy to issue to the agent. After each emergency decision is executed, add the new case to the case library and eliminate obsolete cases that have not been accessed for more than 2 years; The decision log recorded in the blockchain provides real scenario data for meta-learning, and the strategies optimized by meta-learning are traceable and verifiable through the blockchain. When the meta-learning model generates a new strategy, the strategy version number and training sample set must be recorded in the blockchain to ensure decision transparency; User behavior data such as path selection and preference labels are desensitized before being stored in the blockchain, such as user ID hashing. During meta-learning training, use a federated learning framework, with each agent retaining the original data locally and only sharing model parameter updates.

[0038] While embodiments of the application have been shown and described, it is to be understood that the embodiments described are merely exemplary of the principles and application of the present application. Numerous modifications and adaptions can be effected without departing from the spirit and scope of the present application, which is not limited to the exact construction described. Accordingly, the scope of the present application is not intended to be limited to the specific embodiments disclosed.

Claims

1. A method for planning mountain tourism routes based on weather and natural disasters, characterized in that, include: Step 1: Based on the POMDP model, construct a Markov game decision model for multi-agent collaborative optimization, integrate natural disaster monitoring data and ecological environment indicators from transportation routes and tourist destinations, combine the Mamba multimodal feature representation learning method, and construct a dynamic risk matrix through multimodal feature compression and RAG semantic enhancement processing to provide a state perception basis for multi-agent systems in partially observable environments; Step 2: Optimize the UAV inspection strategy through federated learning and the NSGA-II algorithm, and establish an optimization target model that satisfies the Nash equilibrium constraint; design a dedicated agent for weather perception needs, deploy a lightweight YOLOv8 model to identify weather phenomena, and propose a communication and feature sharing mechanism among multiple agents. Introduce the NSGA-II algorithm into the Actor-Critic architecture to optimize the UAV inspection frequency and path adjustment strategy, so that the dynamic communication protocol is automatically triggered when the rainfall intensity exceeds the threshold. Step 3: Design a reward function that integrates security and user experience benefits, and output the final path instruction through an improved DDPG algorithm; Step 4: Build user profiles based on historical user behavior data, perform dynamic weight adjustments and AR simulation previews; automatically increase the weight of corresponding benefit items when users select specific modes, develop AR weather simulation function for preference preview, and finally generate an "observation-behavior" mapping strategy through joint game theory. Step 5: Based on the GIS geographic context reasoning framework, a closed-loop decision engine with "perceptual input - cognitive output" capability is formed, and feature complementarity is achieved through the feedback mechanism between the agent and the environment; blockchain technology is introduced to record emergency decision logs, and a disaster emergency case library is built in combination with the meta-learning framework, enabling the system to quickly adapt to new climate regions; and a "reward-benefit" balance mechanism is used to conduct spatiotemporal evolution intelligent reasoning, thereby improving the robustness of group decision-making in continuous action space.

2. The method for planning mountain tourism routes based on weather and natural disasters according to claim 1, characterized in that, The design of the dynamic risk assessment matrix includes the following steps: Deploy a geological monitoring sensor network to collect relevant data on landslides and debris flows; access meteorological radar and satellite cloud image data to obtain weather indicators including rainfall intensity, visibility, and wind speed; and use GIS geographic registration to unify data from different sources into the WGS84 coordinate system for spatial alignment. A state-space model is used to process time series data, and an implicit representation of continuous time series is constructed. Let the geological sensor sequence be Mapped to the latent space by a linear transformation W: in, Let be the hidden state vector at time step t; The linear transformation matrix of the input data; Let be the linear transformation matrix of the historical hidden states; It is a bias vector with dimension 64; The final output is a 64-dimensional geological feature vector; Construct a disaster domain knowledge graph to map text reports into structured indicators; use LLM for event extraction; The definition includes the matrix dimensions of geological disaster probability G, weather risk level W, and ecological sensitivity E; Overall risk value calculation: A red alert is triggered when R ≥ 75; Deploy a lightweight ONNX inference engine to perform data preprocessing, feature compression, and risk calculation on a Raspberry Pi 4B device; The update cycle is strictly controlled within 5 minutes, and the dimension weights are automatically adjusted according to seasonal changes. By compressing high-dimensional monitoring data into computable quantitative indicators and combining it with real-time semantic enhancement technology, a risk assessment system with dynamic adaptability is constructed, providing accurate environmental state input for subsequent multi-agent collaborative decision-making.

3. The method for planning mountain tourism routes based on weather and natural disasters according to claim 1, characterized in that, The mechanism for communication and feature sharing among multiple agents, which introduces the NSGA-II algorithm into the Actor-Critic architecture, includes the following steps: Based on the dynamic risk matrix output in step one, the communication threshold is set as follows: it is triggered when the comprehensive risk value R is greater than 60 or the rainfall intensity is greater than 20 mm / h. When the threshold is triggered, the agent automatically switches to high-frequency communication mode. A hybrid communication topology is adopted, using distributed communication under normal conditions and automatically switching to centralized communication during disaster warning. The high-definition images collected by the UAV are compressed using the JPEG2000 compression algorithm, compressing the data volume to 15% of the original size while maintaining 95% of the image quality. Each agent trains a lightweight YOLOv8 model on a local dataset, protects local data privacy using differential privacy techniques, and aggregates model parameters on the server side using the FedAvg algorithm. in, These are the model parameters after global aggregation. For the local model parameters of the k-th agent, | | represents the local data volume; The maximum mean difference metric is used to measure the consistency of feature distributions. The visual features of the UAV, geological sensor features, and meteorological data features are projected into a unified 128-dimensional space; a self-attention mechanism is used to calculate the importance of each feature. in, Let be the attention weight for the i-th feature. Let be the energy score of the i-th feature, W be the learnable weight matrix, and b be the bias term. This is the transposed query vector. Let i be the i-th feature vector, and the final fused features are: ; NSGA-II-Actor-Critic Multi-Objective Optimization: Safety benefits G represents the probability of geological disasters; Efficiency Benefits v represents the actual inspection speed; Ecological benefits E represents ecological sensitivity; The Actor network generates an initial policy π(s|θ), then NSGA-II performs crossover and mutation operations on θ to generate a set of candidate policies. Finally, the Critic network evaluates the Q-value of each policy. The winning strategy is selected through non-dominated sorting and crowding calculation, and the Actor parameters are updated. Introduce constraints during policy updates: in, Let i be the reward function for the i-th agent; The changed policy for the i-th agent; It is a set of policies for other agents; ensuring that no agent can gain a higher benefit by unilaterally changing its policy. The drone adjusts its inspection path based on the fusion feature F. When a new risk point is detected, it sends a priority interruption signal to the ground agent via the V2X protocol. It uses cognitive radio technology to automatically switch between the 2.4GHz and 5GHz frequency bands to avoid communication congestion.

4. The method for planning mountain tourism routes based on weather and natural disasters according to claim 1, characterized in that, The design incorporates a reward function that combines safety and experience benefits, and uses an improved DDPG algorithm to handle path planning problems in a continuous action space, including the following steps: Geological disaster risk: G represents the probability of geological disasters output in step one; Weather risk: W represents the weather risk level; Ecological sensitivity: E represents ecological sensitivity; Overall security benefits: ; The landscape along the route is scored using a pre-trained CNN model; the similarity between the route and user preferences is calculated based on the weather sensitivity profile generated in step four. Overall experience benefits: Where A is the landscape aesthetic score output by the pre-trained CNN model, with a value range of [0, 1]; S is the path preference matching degree calculated based on user profile, using cosine similarity, with a value range of [0, 1]. Total reward function: Where α is the dynamic weighting coefficient, adjusted according to the real-time risk value: when When <0.7, α=0.7; when When ≥0.7, α=0.3; The Actor network takes the dynamic risk matrix and user preference vector output from step one as input and outputs continuous actions; the Critic network uses double Q-value estimation to reduce overestimation bias, takes state-action pairs as input and outputs Q-value estimates; adaptive Ornstein-Uhlenbeck noise is introduced, with the noise amplitude decreasing with training epochs. The initial noise σ0 = 0.5 and the attenuation rate λ = 0.

001. Based on the TD error calculation sampling priority, the sampling probability of samples with large errors is increased by 30%; the experience data of different agents are stored in a shared replay pool; A feasible area mask is generated using DEM elevation data to prevent paths from crossing dangerous cliff areas; a conservative path mode is automatically triggered when communication between the UAV and the ground agent is interrupted. When the overall risk value R output in step one is greater than 75, α is forcibly set to 0.7, and the route planning prioritizes avoiding high-risk areas; when the user selects the "landscape priority" mode, α is dynamically adjusted to 0.3, allowing the route to detour to areas with high aesthetic scores; the hybrid mode defaults to α=0.5, balancing safety and experience. An asynchronous Actor-Learner architecture pre-trained model is deployed on 4 GPUs and initialized in a simulated mountainous environment. After the Actor output, a hard constraint layer is added to forcibly exclude all illegal actions that violate the physical rules.

5. The method for planning mountain tourism routes based on weather and natural disasters according to claim 1, characterized in that, The multi-objective optimization model established by integrating NSGA-II and the Actor-Critic architecture accelerates Pareto front convergence and generates joint game decision results, including the following steps: 50% of the strategies were randomly generated through the Actor network, and 50% were generated through NSGA-II crossover mutation. The top 10% of high-fitness strategies were retained as the initial population. Based on the current state s, output continuous actions a; use SBX crossover to generate offspring policies; apply polynomial mutation to explore the new policy space; calculate the Q-value of each policy: The Pareto front is divided into levels based on Q-values ​​and constraints; within the same front, strategies with better diversity are selected using distance metrics. The policy parameter θ is updated along the Q-value gradient direction using the policy gradient method: The strategy of retaining the top 20% of elites is combined with newly generated strategies to form the next generation of the population; When the overall risk value R ≥ 75, the weight of safety benefits increases by 30%; when a "landscape priority" request is detected, the weight of experience benefits increases by 20%. Under the federated learning framework, each agent shares the top 5% of elite policies, and the privacy of the policies is protected through differential privacy; after the shared policies are adjusted by local fitness, they re-participate in NSGA-II evolution. A repeated game mechanism is adopted, in which each agent proposes a strategy in turn, and the Critic evaluates whether an equilibrium has been reached; when the rate of change of all agents' payoffs is less than 2% after 5 consecutive rounds of strategy updates, the game equilibrium is determined to have been reached. The final decision is a triplet containing the path direction angle θ, velocity v, and risk aversion coefficient α. The decision simultaneously satisfies multi-objective optimization and Nash equilibrium constraints.

6. The method for planning mountain tourism routes based on weather and natural disasters according to claim 1, characterized in that, The process involves constructing a user's weather sensitivity profile and generating preference tags by combining historical behavioral data. The dynamic weight adjustment algorithm includes the following steps: Collect user historical behavior data, including: travel records, weather feedback, and social media comments; remove irrelevant information through text cleaning, retaining key weather-related statements; User-selected weather modes and travel itinerary records were analyzed using NLP to assess sentiment in user comments; users were then categorized into three groups using the K-means algorithm. Rain-sheltered type: Over 70% of the historical itineraries have had their routes adjusted due to rainfall; Rain scene photography type: Actively choose rainy day trips and the keyword "rain scene" appears more than 5 times in the comments; Climate-adaptive type: No obvious weather preference, and the correlation between itinerary selection and weather is less than 30%; After each user completes a trip, real-time feedback is collected through questionnaires to update user profile tags; if a user's behavioral data has not been updated for 3 months, the tag weight decreases by 10% each month. The weight adjustment is triggered when the user explicitly selects "Rain Scene Priority" or "Safety Priority" mode through the APP interface; or when the comprehensive risk value R output in step one is greater than 75, the user is forced to enter the safety priority mode; when there is no explicit user instruction and the environment is safe, the default balance mode is used. The weighting adjustment rules are as follows: for rain-avoidance users, the safety benefit weight is increased to 70%, the experience benefit weight is reduced to 30%, and route planning is prohibited from passing through areas with a rainfall probability greater than 40%; for rain scene photography users, the user experience benefit weight is increased to 60%, the safety benefit weight is reduced to 30%, and route detours to areas with a rainfall probability greater than 50% are allowed; for climate-adaptive users, the default weight is maintained, and they are only forcibly adjusted to safety mode when the environmental risk value R is greater than 80. The weighting adjustment constraints are that the minimum weighting of safety benefits is no less than 30%, and the maximum weighting of experience benefits is no more than 60%. After the weighting adjustment, the feasibility of the strategy must be verified through the Critic network. After the user selects a preferred mode, AR technology is used to display the weather effects along the route in real time: when the "Rain Scene Priority" mode is selected, the virtual rainfall effect and possible shooting locations are displayed; when the "Safety Priority" mode is selected, areas at risk of geological disasters are highlighted. Users can adjust the simulation parameters through gesture interaction. Users can rate the display effect in the simulation interface, and the rating data is fed back to the profile update module in real time to adjust the subsequent weight allocation strategy.

7. The method for planning mountain tourism routes based on weather and natural disasters according to claim 1, characterized in that, The GIS-based geographic context reasoning framework forms a closed-loop decision engine, which complements features through feedback between the agent and the environment, including the following steps: Integrate high-precision DEM data, vegetation cover layers, water system distribution layers, and historical geological disaster point data; generate composite layers containing slope, catchment areas, and ecological red line areas using GIS spatial analysis tools; The dynamic risk matrix from step one is overlaid with the GIS layer to generate a real-time geographic scenario model; the model is updated every 5 minutes, keeping pace with the data refresh rate in step one. The system uses a 3D GIS engine to render the terrain, highlighting high-risk areas, user-preferred paths, and the current location of the agent; it also uses AR technology to overlay virtual paths onto the real environment, allowing users to intuitively view the effects of their decisions. The joint game decision generated in step three is sent to the drone and the ground intelligent agent; during the execution of the path by the intelligent agent, environmental data is collected in real time; environmental change data is sent back to the decision engine via the V2X protocol to update the GIS geographic context model. Based on the updated model, the path planning parameters are recalculated to form a closed-loop adjustment. Dynamically adjust trigger conditions: Environmental mutation: When a geological disaster probability G ≥ 80% or a weather risk level W ≥ 4 is detected, route replanning will be forcibly triggered; User intervention: When a user manually adjusts their path preferences through the app, a new decision is generated immediately; Periodic optimization: Lightweight path optimization is automatically performed every 15 minutes to balance computing resources and decision-making timeliness; Drones provide a high-altitude macroscopic view to identify large-scale geological changes; ground sensors provide microscopic data; and weather station data supplements local weather changes. By using GIS coordinate transformation, the UAV orthophotos and ground sensor data are unified into the same coordinate system; the PTP clock synchronization protocol is adopted, and weights are allocated according to the reliability of the data source. The geological change areas identified by the drone are confirmed by ground sensor data; when a single data source shows an anomaly, a self-check is initiated and the data is marked as suspicious. Each decision's input data, output actions, and execution results are packaged and uploaded to the blockchain; the on-chain data is stored using a Merkle tree structure, ensuring data integrity and traceability. Historical decision-making cases are extracted from the blockchain to build a case library containing typical scenarios; the MAML algorithm is used to quickly adapt to new scenarios with a small number of samples, reducing online learning time; combined with GIS spatiotemporal analysis tools, environmental change patterns are identified; and spatiotemporal patterns are encoded into a rule engine to assist the decision engine in predicting environmental trends.

8. The method for planning mountain tourism routes based on weather and natural disasters according to claim 1, characterized in that, The process of introducing blockchain technology to record emergency decision logs and building a disaster emergency case library using a meta-learning framework includes the following steps: The system uses environmental state snapshots to record dynamic risk matrices and GIS geographic context models during decision-making; decision action details include path direction angle θ, speed v, risk avoidance coefficient α, and multi-agent communication protocol status; execution result feedback includes actual user travel data and environmental changes monitored by the system; user intervention records include user manual adjustments to path preferences and system responses. Each decision is fully logged and encapsulated as a blockchain transaction, with timestamps and digital signatures added, and the data is organized using a Merkle tree structure. Access control: Write permissions: Decision engine and user terminal only; Read access: Granted to authorized organizations for post-event analysis; The blockchain explorer allows users to query decision logs at any point in time, supporting multi-dimensional searches by user ID, time range, and risk level. In the event of a geological disaster, it can quickly pinpoint the decision-making path and environmental conditions 5 minutes before the incident, assisting in liability determination. Historical decision-making cases are extracted from the blockchain and categorized by scenario type into typical disaster scenarios, user preference scenarios, and mixed scenarios. Each case is labeled with key features including environmental risk value R, user preference tags, decision action sequence, and execution result score. The MAML algorithm is used to quickly adapt to new scenarios with a small number of samples. Meta-training phase: Simulate different disaster scenarios in the historical case library to train the model to learn "how to learn quickly"; Fine-tuning phase: When encountering new scenarios, adjust the model parameters using 10-20 samples; The matching degree between the new scenario and the case library is measured by Euclidean distance. When the similarity is greater than 85%, the historical strategy is directly invoked. When a new risk scenario is detected, the three most similar cases in the case library are quickly retrieved to generate a candidate strategy set; the Q-value of the candidate strategies is evaluated through the Critic network, and the optimal strategy is selected and sent to the agent; after each emergency decision is executed, the new case is added to the case library, and outdated cases that have not been accessed for more than two years are eliminated. The decision logs recorded by the blockchain provide real-world data for meta-learning, and the strategies optimized by meta-learning are traceable and verifiable through the blockchain. When the meta-learning model generates a new strategy, the strategy version number and training sample set are recorded in the blockchain to improve decision transparency. User behavior data is anonymized before being stored in the blockchain; during meta-learning training, a federated learning framework is used, where each agent retains the original data locally and only shares the model parameter update amount.

Images