Transportation hoisting intelligent scheduling optimization method based on multi-dimensional screening and trajectory analysis

Abstract

The application discloses a transport hoisting intelligent scheduling optimization method based on multi-dimensional screening and trajectory analysis, which collects and analyzes multi-dimensional data of transport trajectories, task nodes, environmental parameters, operation timeliness and safety events and the like of historical hoisting operations, and constructs a hoisting operation path knowledge base containing space-time characteristics, risk weight and efficiency index; by using big data mining and machine learning algorithm, the optimal path mode and dynamic adjustment strategy under the typical operation scene are extracted; before actual operation, the optimal hoisting operation path is matched or generated from the knowledge base in combination with current task parameters and real-time environmental data, and dynamic optimization and real-time correction are supported. The application solves the problems that the traditional hoisting operation path relies on artificial experience, lacks data-driven decision and cannot adapt to complex dynamic environment, and significantly improves the scientificity, safety and efficiency of path planning, and reduces transport cost and accident risk.

Description

Technical Field

[0001] This invention relates to the field of intelligent management and control technology for transportation and hoisting operations, specifically to an intelligent scheduling and optimization method for transportation and hoisting operations based on multi-dimensional screening and trajectory analysis. Background Technology

[0002] In traditional transport and hoisting operations, route planning primarily relies on human experience or simple navigation tools. Dispatchers often choose transport routes for hoisting operations based on their own experience, using main roads recommended by map software or previously worked routes. Meanwhile, some drivers also use common in-vehicle navigation systems, which typically offer route recommendations based on "shortest distance" or "fastest time."

[0003] However, hoisting operations have highly unique characteristics. On the one hand, hoisting operations frequently require traversing narrow roads, construction sites, and areas with height and weight restrictions, which ordinary navigation routes easily overlook. For example, ordinary navigation may not indicate the minimum turning radius allowed for hoisting equipment, and the actual turning radius of some roads may not meet the passage requirements of hoisting equipment. On the other hand, the equipment involved in hoisting operations is usually large and heavy, placing high demands on the load-bearing capacity of the road surface and the gradient of the road, factors that traditional route planning methods cannot fully consider.

[0004] Inadequate handling of scenario complexity: Lifting operations involve extremely complex scenarios, with different locations exhibiting varying geographical environments and road conditions. Lifting operations within cities may require navigating narrow alleys and dense building clusters, while cross-regional operations may face differences in road standards and traffic rules across different areas. Traditional route planning methods are unable to provide tailored solutions for these complex scenarios and cannot adequately adapt to various unique geographical and operational environments.

[0005] Poor dynamic risk response capability: Lifting operations face numerous dynamic risks, such as weather changes (severe weather such as heavy rain and strong winds can affect the safety of lifting operations and the selection of travel routes), traffic control (temporary road closures and traffic congestion can affect the timeliness of operations), and unexpected situations at the work site (such as cargo deviation and on-site equipment failure). Traditional methods cannot perceive these dynamic risks in real time and adjust routes in a timely manner. When encountering emergencies, it often leads to operation delays, increased transportation costs, and even safety accidents.

[0006] Difficulty in transferring experience: Excellent dispatchers accumulate rich experience in long-term hoisting operation scheduling, enabling them to rationally plan routes based on different task situations. However, this experience is often personalized and difficult to quantify, making it difficult for new employees or personnel working across regions to quickly learn and master it. This leads to situations in new work scenarios where unfamiliarity with road conditions can easily result in detours, delays, or safety accidents, affecting the overall efficiency and quality of hoisting operations.

[0007] Insufficient Data Utilization: During hoisting operations, a large amount of historical data is generated, including transport trajectories, task milestones, environmental parameters, operation timelines, and safety incidents. However, this data is typically stored in scattered locations across different systems, such as transport management systems and safety logs, without forming a structured knowledge base. Traditional route planning methods do not fully utilize this historical data, failing to extract patterns and experiences from past operations and thus unable to provide effective data support for current route planning.

[0008] Therefore, we propose an intelligent scheduling optimization method for transportation and hoisting based on multi-dimensional screening and trajectory analysis. Summary of the Invention

[0009] To achieve the above objectives, this invention provides the following technical solution: a transportation and hoisting intelligent scheduling optimization method based on multi-dimensional screening and trajectory analysis. This method systematically integrates multi-source heterogeneous data from historical hoisting operations, constructs a dynamically adaptable knowledge base, and combines real-time environment and task parameters to achieve scientific planning and dynamic optimization of hoisting operation paths. Specifically, it includes the following steps:

[0010] Historical data collection steps: Multi-dimensional historical data throughout the entire lifecycle of the hoisting operation is collected using various types of data acquisition terminals. This historical data includes, but is not limited to:

[0011] S1 transport trajectory data: vehicle latitude and longitude coordinate sequence (positioning accuracy ≤10 meters), driving speed, acceleration, direction angle, driving timestamp, and vehicle attitude data (such as tilt angle and vibration amplitude) collected by the vehicle-mounted high-precision GPS / BeiDou dual-mode positioning terminal.

[0012] S2 Task Node Data: Based on the structured task information recorded by the transportation and hoisting operation visualization management platform, including loading point (latitude and longitude, site type, loading and unloading equipment configuration), unloading point (latitude and longitude, site type, loading and unloading equipment configuration), route points (such as necessary bridges, checkpoints), task time window (planned start time, planned end time, allowed delay time), and task priority (urgent / normal / low priority).

[0013] S3 environmental parameter data: Multi-source environmental information collected through vehicle-mounted meteorological sensors (collecting real-time wind speed, wind direction, temperature, and humidity), road meteorological monitoring stations (collecting road section-level rainfall, snowfall, and visibility), traffic management department interfaces (obtaining real-time traffic flow, congestion index, and speed limit adjustment information), and construction site management systems (obtaining terrain features of the work site, such as slope, turning radius restrictions, and distribution of surrounding obstacles, including high-voltage line height, building height restrictions, and underground pipeline locations);

[0014] S4 Operation Timeliness Data: Task execution time parameters automatically recorded by the platform, including actual travel time (total travel time from loading point to unloading point), travel time for each road segment (statistically divided by electronic map), waiting time (such as loading wait, unloading wait, traffic control wait), total task completion time (time span from task creation to final completion), and delay time of key nodes (such as loading timeout, unloading timeout).

[0015] S5 safety incident data: Safety-related incidents recorded through vehicle video monitoring (identifying abnormal scenes such as collisions, scratches, and cargo deviation), sensor alarms (such as torque limiter over-limit and brake malfunction), and driver manual reporting (accident descriptions submitted through the vehicle terminal). These incidents include the type of accident (equipment damage / personnel injury / cargo loss), location (latitude and longitude + road segment markings), time of occurrence, risk factors (such as narrow roads, insufficient turning radius, speeding, and weather conditions), and degree of loss (minor / moderate / severe).

[0016] Data preprocessing steps: Cleaning, transforming, and feature engineering the collected raw historical data, specifically including:

[0017] S1 Data Cleaning: Remove data with location drift (such as more than 10 consecutive coordinate points with a straight-line distance exceeding the theoretical value of the vehicle's maximum speed), duplicate records, and data with missing key fields (such as missing latitude and longitude of loading points);

[0018] S2 data standardization: unify the time format (convert to UTC timestamp), coordinate system (unify to WGS84 coordinate system), and units (e.g., speed is unified to km / h, weight is unified to tons) of different data sources.

[0019] S3 Feature Extraction: Based on business logic and statistical methods, extract key features from raw data, including but not limited to:

[0020] Transportation scenario characteristics: Operation area type (urban core area / suburbs / inter-regional expressway), road type (main road / secondary road / local road / dedicated hoisting channel), traffic time (weekdays / holidays, peak hours / off-peak hours);

[0021] Task characteristics: Cargo type (general building materials / heavy equipment / precision instruments), cargo weight class (≤10 tons / 10-30 tons / 30-50 tons / >50 tons), vehicle type (light crane / medium crane / heavy crane / customized crane), lifting equipment configuration (whether auxiliary lifting tools are carried).

[0022] Route characteristics: total route length (km), total number of turns, maximum turning radius (m), gradient variation range (minimum gradient - maximum gradient), number of restricted areas (e.g., schools, hospitals, narrow sections);

[0023] Risk characteristics: historical accident density (number of accidents per unit mileage), high-risk road section markings (such as bridges with a gradient > 8° and no anti-skid treatment), and weather sensitivity (accident incidence rate on rainy / snowy / windy days).

[0024] Efficiency characteristics: average driving speed (km / h), on-time completion rate (number of tasks completed on time / total number of tasks), fuel consumption per unit mile (liters / km, related to fuel cost).

[0025] S4 data classification and storage: Preprocessed data is classified and stored according to operation scenarios (such as short-distance hoisting in the city (one way ≤ 50 kilometers), long-distance hoisting across regions (one way > 50 kilometers), hoisting in special terrain (mountainous / tunnel / bridge-dense areas)), task priority, and cargo weight class, to build a structured historical database of hoisting operation paths;

[0026] Knowledge base construction steps: Based on the preprocessed historical database, big data mining technology with multi-algorithm fusion is used to extract path optimization patterns under typical operation scenarios, and a hoisting operation path knowledge base containing path features, applicable conditions, risk weights, and efficiency indicators is constructed, specifically including:

[0027] S1 Cluster Analysis: Using K-means++ or DBSCAN algorithms, historical tasks are clustered based on transportation scenario characteristics (such as region type + road type + traffic time period) and task characteristics (such as cargo weight + vehicle type) to identify a set of typical operation scenarios with similar characteristics (such as "urban core area + weekday peak hours + 30-50 tons of cargo + medium-sized crane" as a typical scenario).

[0028] S2 Association Rule Mining: Using Apriori or FP-Growth algorithms, we mine strong association rules between "path features - efficiency indicators" and "path features - risk indicators" in typical scenarios. For example, "when the maximum turning radius of the path is >20 meters and the number of restricted areas passed through is ≤2, the probability of accidents is reduced by 40%" and "the transportation time of choosing a secondary road (width ≥10 meters) is reduced by an average of 25% compared to the main road, but the fuel consumption increases by 10%".

[0029] S3 Temporal Pattern Recognition: Utilizes LSTM (Long Short-Term Memory Network) or ARIMA (Autoregressive Integral Moving Average Model) to analyze the time series characteristics of historical tasks (such as traffic flow patterns and seasonal weather effects during specific time periods) and extracts the optimal path selection tendencies for different time periods (such as morning and evening rush hours on holidays, rainy season / winter).

[0030] S4 Efficiency and Risk Assessment Model: Construct a comprehensive evaluation model with efficiency weights (e.g., time saving rate = theoretical shortest time / actual driving time), risk weights (e.g., accident probability = number of historical accidents / total number of trips on the route), and cost weights (e.g., fuel cost = fuel consumption per unit mile × driving mileage) as core indicators. A multi-objective optimization algorithm (e.g., NSGA-II non-dominated sorting genetic algorithm) is used to calculate the comprehensive performance score of different paths (comprehensive performance = efficiency weight × α + risk weight × β + cost weight × γ, where α, β, and γ are configurable weight coefficients, with default values ​​of α = 0.6, β = 0.3, and γ = 0.1). The path with the optimal comprehensive performance is then selected as the recommended path in the knowledge base.

[0031] The S5 knowledge base uses structured storage: it stores the optimal path features (such as origin and destination coordinates, key points along the route, and road type combinations), applicable conditions (such as cargo weight range and weather conditions), risk weights (such as the proportion of high-risk road sections), and efficiency indicators (such as average travel time and on-time completion rate) corresponding to each typical scenario in the knowledge base in a structured data format (such as JSON / XML), and establishes a mapping index between scenario features and path solutions to support fast retrieval;

[0032] Current task parsing steps: When a new hoisting operation task is received, the system obtains detailed parameters and real-time environmental data of the current task through the visual management and control platform, specifically including:

[0033] S1 Task Parameter Analysis: Extract the core information of the task, including the precise coordinates of the origin and destination (loading point and unloading point), cargo type (associated with attributes through material code), cargo weight (tons), planned time window (earliest allowed departure time and latest allowed arrival time), vehicle / lifting equipment configuration (vehicle type, maximum lifting capacity, boom length), and task priority (urgency level).

[0034] S2 Real-time Environmental Data Acquisition: Collects current environmental information in real time through multi-source data interfaces, including:

[0035] Real-time traffic data: Current traffic flow (vehicles / hour), congestion index (0-10), real-time speed limits (km / h) for each road segment, and accident hotspots (e.g., a rear-end collision 2 kilometers ahead causing lane closure).

[0036] Real-time meteorological data: real-time wind speed (m / s, with a focus on strong winds >8m / s), rainfall (mm / h, with a focus on heavy rain >10mm / h), snowfall (cm / h, with a focus on snow accumulation >2cm / h), and visibility (meters, with a focus on low visibility <500 meters) along the operation path.

[0037] Work site data: Real-time site status of unloading / loading points (e.g., whether they are occupied by other equipment, whether the ground is slippery), and surrounding temporary restriction information (e.g., temporary road closures due to road construction, and newly added height restriction barriers);

[0038] S3 Task Constraint Generation: Based on task parameters and environmental data, generate path planning constraints for the current task, including:

[0039] Physical constraints: maximum permissible driving gradient for vehicles (e.g., heavy cranes ≤ 15°), minimum turning radius (e.g., ≥ 25 meters when the boom is extended), and the degree of matching between cargo weight and road load-bearing capacity (e.g., bridge weight limit > total cargo weight).

[0040] Time constraints: Transportation must be completed within the planned time window (e.g., departure time from loading point ≥ 8:00 and arrival time at unloading point ≤ 10:00).

[0041] Safety constraints: Avoid high-risk road sections (such as road sections with a historical accident density > 0.5 times / km) and avoid passing through areas with a risk of falling objects from heights in severe weather (such as wind speed > 12m / s);

[0042] Path matching and optimization steps: Based on the current task parsing results, the optimal path is intelligently matched from the knowledge base and dynamically adjusted, specifically including:

[0043] S1 Similarity Matching: Using a multi-dimensional similarity calculation algorithm, the current task parameters (such as cargo weight class, planned time window, origin and destination areas, and traffic hours) are compared with historical scene features (such as similar cargo transportation time periods, task time distribution in the same area, and similar road type combinations). A similarity score is calculated, and typical scenes with similarity scores higher than a preset threshold (such as ≥0.8) and their corresponding optimal paths are selected as a set of candidate paths. The similarity calculation can use weighted Euclidean distance (logarithmic feature) or cosine similarity (categorical feature), and the weights are dynamically adjusted according to the importance of the features (such as task priority weight > cargo weight weight).

[0044] S2 route initial screening: Based on the constraints of the current task (such as physical constraints, time constraints, and safety constraints), the candidate route set is initially filtered to remove routes that do not meet the conditions (such as routes containing road sections with a gradient greater than the maximum allowable gradient for vehicles, or routes whose estimated travel time exceeds the planned time window).

[0045] S3 Dynamic Optimization: For candidate paths that pass the initial screening, real-time environmental data (such as current traffic congestion index and weather warnings) and real-time vehicle status (such as current speed and remaining fuel) are combined with reinforcement learning models (such as Deep Q-Network, DQN) or dynamic programming algorithms (such as an improved version of Dijkstra's algorithm) to adjust the paths in real time.

[0046] If a road segment is detected to have a real-time congestion index greater than level 7 (severe congestion), it will be automatically replaced with a low-congestion alternative route in the same area from the knowledge base.

[0047] If there is a temporary weather risk ahead (such as a sudden increase in wind speed to 10m / s), then avoid that section of road and replan the detour route.

[0048] If the vehicle's current remaining fuel is insufficient to support the original route (estimated based on historical fuel consumption data), then the route that passes through a gas station or service area will be selected first.

[0049] S4 Comprehensive Performance Evaluation: The optimized path set is scored comprehensively (based on a multi-objective optimization model of efficiency, risk, and cost), and the path with the highest comprehensive performance score is selected as the final recommended optimal hoisting operation path; if multiple paths have similar scores (e.g., difference < 5%), the path with the highest historical on-time completion rate is selected first.

[0050] Output and Feedback Steps: The finalized optimal route is pushed to the driver's navigation device (such as an in-vehicle tablet or mobile app) and the monitoring management interface through the visual management platform. Simultaneously, a real-time feedback mechanism for route execution is established, specifically including:

[0051] S1 Route Output: Provides detailed route navigation instructions to the driver (including turn prompts, speed limit reminders, and risk section warnings), and generates a route execution plan to the monitoring terminal (including estimated travel time for each segment and time predictions for key nodes).

[0052] S2 Real-time Data Acquisition: During the route execution process, real-time driving data is continuously collected, including the actual driving trajectory (real-time positioning via GPS / BeiDou), the actual driving time of each road segment, the actual arrival time, whether any risk events occur (such as deviation from the planned route or encountering a sudden accident), and driver operation behavior (such as the frequency of rapid acceleration and sudden braking).

[0053] S3 Knowledge Base Iteration: Compare and analyze the actual execution data with the original task parameters and recommended paths, calculate the path execution deviation (such as the difference rate between actual driving time and predicted time, and the difference between the actual number of risk events and the predicted number), and supplement the deviation data and new risk event records to the historical database; through periodic (such as daily / weekly) incremental learning, update the path features, efficiency indicators and risk weights in the knowledge base to optimize the accuracy of subsequent path planning.

[0054] S4 Visual Feedback: The monitoring interface of the visual management platform displays the real-time status of the route execution (such as current location, remaining distance, and estimated arrival time), and provides early warnings for behaviors that deviate from the optimal route (such as drivers manually changing routes), assisting managers in timely intervention.

[0055] Preferably, in the historical data acquisition step, the transportation trajectory data also includes distance information of surrounding obstacles (such as the minimum distance to roadside guardrails and buildings) obtained by vehicle-mounted LiDAR, to supplement the limiting features of the path passage space; the task node data further includes specific requirements for loading / unloading operations (such as whether auxiliary hoisting equipment is needed, and the way the goods are fixed), to associate the site adaptability requirements of the path selection.

[0056] Preferably, in the data preprocessing step, the feature extraction further includes feature importance analysis based on machine learning algorithms (such as random forests) to screen out the key features that have the greatest impact on path efficiency and risk (such as "cargo weight > 30 tons and turning radius < 20 meters" having an impact weight of 0.7 on the probability of accidents), and prioritize retaining these high-impact features for subsequent knowledge base construction.

[0057] Preferably, in the knowledge base construction step, the comprehensive performance evaluation model also includes user preference parameters (such as enterprises prioritizing time costs over fuel costs), and dynamically adjusts the proportions of efficiency, risk, and cost through configurable weight coefficients (α, β, γ) to meet the personalized needs of different enterprises.

[0058] Preferably, in the path matching and optimization step, the similarity matching algorithm further includes a deep learning-based embedding representation method (such as converting textual features (e.g., operation area description, road name) into vectors using the BERT model, and then calculating the similarity with the textual features of the current task) to improve the matching accuracy of complex scene descriptions.

[0059] Preferably, in the path matching and optimization step, the dynamic optimization further includes predicting the future congestion trend of road segments based on a real-time traffic prediction model (such as LSTM-based 15-minute traffic flow prediction) and adjusting the path in advance to avoid impending congestion.

[0060] Preferably, the feedback of the actual execution data also includes the actual measured values ​​of environmental parameters (such as the actual wind speed and rainfall during the driving process) to correct the prediction model of environmental risks in the knowledge base (such as adjusting the historical statistical value of "accident probability when wind speed > 10m / s").

[0061] Preferably, in the output and feedback steps, the knowledge base iteration also includes filtering abnormal data (such as erroneous wind speed records caused by sensor failure), and removing obviously unreasonable data by setting a reasonable threshold range (such as the normal wind speed range of 0-50m / s) to ensure the reliability of the knowledge base data.

[0062] Preferably, in the output and feedback steps, the visual feedback also includes providing managers with a detailed analysis report on path optimization (such as the composition of the comprehensive performance score of the recommended path, a comparative analysis with alternative paths, and a reference for the execution of similar historical tasks), to help decision-makers understand the scientific basis for path selection.

[0063] Compared with existing technologies, this invention provides an intelligent scheduling and optimization method for transportation and hoisting based on multi-dimensional screening and trajectory analysis, which has the following beneficial effects:

[0064] 1. This intelligent scheduling and optimization method for transportation and hoisting based on multi-dimensional screening and trajectory analysis completely changes the traditional path planning model that relies on human experience by building a knowledge base through in-depth mining of historical data. By collecting and analyzing multi-dimensional data such as historical transportation trajectories, task nodes, environmental parameters, operation timeliness, and safety events of hoisting operations, the optimal path patterns for typical scenarios are extracted using big data mining and machine learning algorithms. This makes path planning no longer based on subjective judgment, but supported by a large amount of real-world data, enabling accurate identification of efficient path characteristics in different operation scenarios (such as short-distance hoisting within cities, long-distance hoisting across regions, and hoisting in special terrain). For example, the system can analyze the optimal driving route for a specific time period, specific area, certain type of goods, and vehicle configuration, avoiding the limitations and errors that may exist in human experience, providing a scientific and quantifiable decision-making basis for hoisting operations, and significantly improving the scientific nature and accuracy of path planning.

[0065] 2. This intelligent scheduling and optimization method for transportation and hoisting, based on multi-dimensional screening and trajectory analysis, possesses strong dynamic adaptability and can respond to changes in the operating environment in real time. By combining real-time environmental data (such as current traffic flow, weather warnings, and road construction information) with task parameters (such as origin and destination, cargo type, time window, and vehicle / hoisting equipment configuration), the system can intelligently match or generate the optimal hoisting operation path from the knowledge base and dynamically adjust it according to real-time changes. In the face of emergencies, such as severe weather like heavy rain and strong winds, temporary traffic control, or road construction, it can quickly identify risks and replan the route, effectively avoiding risky sections. For example, when it detects that there is a risk of water accumulation or landslides on the road ahead due to heavy rain, the system can immediately adjust the route, guiding vehicles to avoid dangerous areas and choose safer routes. This dynamic adjustment capability greatly improves the safety of hoisting operations, reduces the probability of accidents caused by emergencies, and also ensures the timeliness of operations, reducing losses caused by delays.

[0066] 3. This intelligent scheduling and optimization method for transportation and hoisting based on multi-dimensional screening and trajectory analysis achieves continuous optimization of path planning capabilities through a feedback mechanism. During the process of outputting the optimal path and guiding vehicles to perform operations, the system collects real-time path execution data during the actual operation, including actual driving trajectories, time deviations, risk events, and driver operation behavior data, and feeds this data back to the historical database. Through analysis and learning of this feedback data, the knowledge base and optimization model can be continuously iterated and updated, allowing the path planning strategy to continuously evolve with the accumulation of operational experience. For example, with the accumulation of operational data in more different scenarios, the system can more accurately predict the comprehensive effectiveness of different paths under various conditions, further optimizing path selection. Furthermore, this method is deeply integrated with a transportation and hoisting operation visualization and control platform, allowing path instructions to be directly pushed to the driver and monitoring terminals, realizing closed-loop management of "planning-execution-feedback," promoting synergistic efficiency among various links, and comprehensively improving the overall efficiency, safety, and economic benefits of hoisting operations. Detailed Implementation

[0067] The technical solutions in the embodiments of the present invention will be clearly and completely described below. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0068] Example

[0069] An Example of an Intelligent Scheduling and Optimization Method for Transportation and Lifting Based on Multi-Dimensional Screening and Trajectory Analysis

[0070] A smart scheduling and optimization method for transportation and hoisting based on multi-dimensional screening and trajectory analysis is proposed. This method systematically integrates multi-source heterogeneous data from historical hoisting operations, constructs a dynamically adaptable knowledge base, and combines real-time environment and task parameters to achieve scientific planning and dynamic optimization of hoisting operation paths. Specifically, it includes the following steps:

[0071] Historical data collection steps: Multi-dimensional historical data throughout the entire lifecycle of the hoisting operation is collected using various types of data acquisition terminals. This historical data includes, but is not limited to:

[0072] S1 transport trajectory data: vehicle latitude and longitude coordinate sequence (positioning accuracy ≤10 meters), driving speed, acceleration, direction angle, driving timestamp, and vehicle attitude data (such as tilt angle and vibration amplitude) collected by the vehicle-mounted high-precision GPS / BeiDou dual-mode positioning terminal.

[0073] S2 Task Node Data: Based on the structured task information recorded by the transportation and hoisting operation visualization management platform, including loading point (latitude and longitude, site type, loading and unloading equipment configuration), unloading point (latitude and longitude, site type, loading and unloading equipment configuration), route points (such as necessary bridges, checkpoints), task time window (planned start time, planned end time, allowed delay time), and task priority (urgent / normal / low priority).

[0074] S3 environmental parameter data: Multi-source environmental information collected through vehicle-mounted meteorological sensors (collecting real-time wind speed, wind direction, temperature, and humidity), road meteorological monitoring stations (collecting road section-level rainfall, snowfall, and visibility), traffic management department interfaces (obtaining real-time traffic flow, congestion index, and speed limit adjustment information), and construction site management systems (obtaining terrain features of the work site, such as slope, turning radius restrictions, and distribution of surrounding obstacles, including high-voltage line height, building height restrictions, and underground pipeline locations);

[0075] S4 Operation Timeliness Data: Task execution time parameters automatically recorded by the platform, including actual travel time (total travel time from loading point to unloading point), travel time for each road segment (statistically divided by electronic map), waiting time (such as loading wait, unloading wait, traffic control wait), total task completion time (time span from task creation to final completion), and delay time of key nodes (such as loading timeout, unloading timeout).

[0076] S5 safety incident data: Safety-related incidents recorded through vehicle video monitoring (identifying abnormal scenes such as collisions, scratches, and cargo deviation), sensor alarms (such as torque limiter over-limit and brake malfunction), and driver manual reporting (accident descriptions submitted through the vehicle terminal). These incidents include the type of accident (equipment damage / personnel injury / cargo loss), location (latitude and longitude + road segment markings), time of occurrence, risk factors (such as narrow roads, insufficient turning radius, speeding, and weather conditions), and degree of loss (minor / moderate / severe).

[0077] Data preprocessing steps: Cleaning, transforming, and feature engineering the collected raw historical data, specifically including:

[0078] S1 Data Cleaning: Remove data with location drift (such as more than 10 consecutive coordinate points with a straight-line distance exceeding the theoretical value of the vehicle's maximum speed), duplicate records, and data with missing key fields (such as missing latitude and longitude of loading points);

[0079] S2 data standardization: unify the time format (convert to UTC timestamp), coordinate system (unify to WGS84 coordinate system), and units (e.g., speed is unified to km / h, weight is unified to tons) of different data sources.

[0080] S3 Feature Extraction: Based on business logic and statistical methods, extract key features from raw data, including but not limited to:

[0081] Transportation scenario characteristics: Operation area type (urban core area / suburbs / inter-regional expressway), road type (main road / secondary road / local road / dedicated hoisting channel), traffic time (weekdays / holidays, peak hours / off-peak hours);

[0082] Task characteristics: Cargo type (general building materials / heavy equipment / precision instruments), cargo weight class (≤10 tons / 10-30 tons / 30-50 tons / >50 tons), vehicle type (light crane / medium crane / heavy crane / customized crane), lifting equipment configuration (whether auxiliary lifting tools are carried).

[0083] Route characteristics: total route length (km), total number of turns, maximum turning radius (m), gradient variation range (minimum gradient - maximum gradient), number of restricted areas (e.g., schools, hospitals, narrow sections);

[0084] Risk characteristics: historical accident density (number of accidents per unit mileage), high-risk road section markings (such as bridges with a gradient > 8° and no anti-skid treatment), and weather sensitivity (accident incidence rate on rainy / snowy / windy days).

[0085] Efficiency characteristics: average driving speed (km / h), on-time completion rate (number of tasks completed on time / total number of tasks), fuel consumption per unit mile (liters / km, related to fuel cost).

[0086] S4 data classification and storage: Preprocessed data is classified and stored according to operation scenarios (such as short-distance hoisting in the city (one way ≤ 50 kilometers), long-distance hoisting across regions (one way > 50 kilometers), hoisting in special terrain (mountainous / tunnel / bridge-dense areas)), task priority, and cargo weight class, to build a structured historical database of hoisting operation paths;

[0087] Knowledge base construction steps: Based on the preprocessed historical database, big data mining technology with multi-algorithm fusion is used to extract path optimization patterns under typical operation scenarios, and a hoisting operation path knowledge base containing path features, applicable conditions, risk weights, and efficiency indicators is constructed, specifically including:

[0088] S1 Cluster Analysis: Using K-means++ or DBSCAN algorithms, historical tasks are clustered based on transportation scenario characteristics (such as region type + road type + traffic time period) and task characteristics (such as cargo weight + vehicle type) to identify a set of typical operation scenarios with similar characteristics (such as "urban core area + weekday peak hours + 30-50 tons of cargo + medium-sized crane" as a typical scenario).

[0089] S2 Association Rule Mining: Using Apriori or FP-Growth algorithms, we mine strong association rules between "path features - efficiency indicators" and "path features - risk indicators" in typical scenarios. For example, "when the maximum turning radius of the path is >20 meters and the number of restricted areas passed through is ≤2, the probability of accidents is reduced by 40%" and "the transportation time of choosing a secondary road (width ≥10 meters) is reduced by an average of 25% compared to the main road, but the fuel consumption increases by 10%".

[0090] S3 Temporal Pattern Recognition: Utilizes LSTM (Long Short-Term Memory Network) or ARIMA (Autoregressive Integral Moving Average Model) to analyze the time series characteristics of historical tasks (such as traffic flow patterns and seasonal weather effects during specific time periods) and extracts the optimal path selection tendencies for different time periods (such as morning and evening rush hours on holidays, rainy season / winter).

[0091] S4 Efficiency and Risk Assessment Model: Construct a comprehensive evaluation model with efficiency weights (e.g., time saving rate = theoretical shortest time / actual driving time), risk weights (e.g., accident probability = number of historical accidents / total number of trips on the route), and cost weights (e.g., fuel cost = fuel consumption per unit mile × driving mileage) as core indicators. A multi-objective optimization algorithm (e.g., NSGA-II non-dominated sorting genetic algorithm) is used to calculate the comprehensive performance score of different paths (comprehensive performance = efficiency weight × α + risk weight × β + cost weight × γ, where α, β, and γ are configurable weight coefficients, with default values ​​of α = 0.6, β = 0.3, and γ = 0.1). The path with the optimal comprehensive performance is then selected as the recommended path in the knowledge base.

[0092] The S5 knowledge base uses structured storage: it stores the optimal path features (such as origin and destination coordinates, key points along the route, and road type combinations), applicable conditions (such as cargo weight range and weather conditions), risk weights (such as the proportion of high-risk road sections), and efficiency indicators (such as average travel time and on-time completion rate) corresponding to each typical scenario in the knowledge base in a structured data format (such as JSON / XML), and establishes a mapping index between scenario features and path solutions to support fast retrieval;

[0093] Current task parsing steps: When a new hoisting operation task is received, the system obtains detailed parameters and real-time environmental data of the current task through the visual management and control platform, specifically including:

[0094] S1 Task Parameter Analysis: Extract the core information of the task, including the precise coordinates of the origin and destination (loading point and unloading point), cargo type (associated with attributes through material code), cargo weight (tons), planned time window (earliest allowed departure time and latest allowed arrival time), vehicle / lifting equipment configuration (vehicle type, maximum lifting capacity, boom length), and task priority (urgency level).

[0095] S2 Real-time Environmental Data Acquisition: Collects current environmental information in real time through multi-source data interfaces, including:

[0096] Real-time traffic data: Current traffic flow (vehicles / hour), congestion index (0-10), real-time speed limits (km / h) for each road segment, and accident hotspots (e.g., a rear-end collision 2 kilometers ahead causing lane closure).

[0097] Real-time meteorological data: real-time wind speed (m / s, with a focus on strong winds >8m / s), rainfall (mm / h, with a focus on heavy rain >10mm / h), snowfall (cm / h, with a focus on snow accumulation >2cm / h), and visibility (meters, with a focus on low visibility <500 meters) along the operation path.

[0098] Work site data: Real-time site status of unloading / loading points (e.g., whether they are occupied by other equipment, whether the ground is slippery), and surrounding temporary restriction information (e.g., temporary road closures due to road construction, and newly added height restriction barriers);

[0099] S3 Task Constraint Generation: Based on task parameters and environmental data, generate path planning constraints for the current task, including:

[0100] Physical constraints: maximum permissible driving gradient for vehicles (e.g., heavy cranes ≤ 15°), minimum turning radius (e.g., ≥ 25 meters when the boom is extended), and the degree of matching between cargo weight and road load-bearing capacity (e.g., bridge weight limit > total cargo weight).

[0101] Time constraints: Transportation must be completed within the planned time window (e.g., departure time from loading point ≥ 8:00 and arrival time at unloading point ≤ 10:00).

[0102] Safety constraints: Avoid high-risk road sections (such as road sections with a historical accident density > 0.5 times / km) and avoid passing through areas with a risk of falling objects from heights in severe weather (such as wind speed > 12m / s);

[0103] Path matching and optimization steps: Based on the current task parsing results, the optimal path is intelligently matched from the knowledge base and dynamically adjusted, specifically including:

[0104] S1 Similarity Matching: Using a multi-dimensional similarity calculation algorithm, the current task parameters (such as cargo weight class, planned time window, origin and destination areas, and traffic hours) are compared with historical scene features (such as similar cargo transportation time periods, task time distribution in the same area, and similar road type combinations). A similarity score is calculated, and typical scenes with similarity scores higher than a preset threshold (such as ≥0.8) and their corresponding optimal paths are selected as a set of candidate paths. The similarity calculation can use weighted Euclidean distance (logarithmic feature) or cosine similarity (categorical feature), and the weights are dynamically adjusted according to the importance of the features (such as task priority weight > cargo weight weight).

[0105] S2 route initial screening: Based on the constraints of the current task (such as physical constraints, time constraints, and safety constraints), the candidate route set is initially filtered to remove routes that do not meet the conditions (such as routes containing road sections with a gradient greater than the maximum allowable gradient for vehicles, or routes whose estimated travel time exceeds the planned time window).

[0106] S3 Dynamic Optimization: For candidate paths that pass the initial screening, real-time environmental data (such as current traffic congestion index and weather warnings) and real-time vehicle status (such as current speed and remaining fuel) are combined with reinforcement learning models (such as Deep Q-Network, DQN) or dynamic programming algorithms (such as an improved version of Dijkstra's algorithm) to adjust the paths in real time.

[0107] If a road segment is detected to have a real-time congestion index greater than level 7 (severe congestion), it will be automatically replaced with a low-congestion alternative route in the same area from the knowledge base.

[0108] If there is a temporary weather risk ahead (such as a sudden increase in wind speed to 10m / s), then avoid that section of road and replan the detour route.

[0109] If the vehicle's current remaining fuel is insufficient to support the original route (estimated based on historical fuel consumption data), then the route that passes through a gas station or service area will be selected first.

[0110] S4 Comprehensive Performance Evaluation: The optimized path set is scored comprehensively (based on a multi-objective optimization model of efficiency, risk, and cost), and the path with the highest comprehensive performance score is selected as the final recommended optimal hoisting operation path; if multiple paths have similar scores (e.g., difference < 5%), the path with the highest historical on-time completion rate is selected first.

[0111] Output and Feedback Steps: The finalized optimal route is pushed to the driver's navigation device (such as an in-vehicle tablet or mobile app) and the monitoring management interface through the visual management platform. Simultaneously, a real-time feedback mechanism for route execution is established, specifically including:

[0112] S1 Route Output: Provides detailed route navigation instructions to the driver (including turn prompts, speed limit reminders, and risk section warnings), and generates a route execution plan to the monitoring terminal (including estimated travel time for each segment and time predictions for key nodes).

[0113] S2 Real-time Data Acquisition: During the route execution process, real-time driving data is continuously collected, including the actual driving trajectory (real-time positioning via GPS / BeiDou), the actual driving time of each road segment, the actual arrival time, whether any risk events occur (such as deviation from the planned route or encountering a sudden accident), and driver operation behavior (such as the frequency of rapid acceleration and sudden braking).

[0114] S3 Knowledge Base Iteration: Compare and analyze the actual execution data with the original task parameters and recommended paths, calculate the path execution deviation (such as the difference rate between actual driving time and predicted time, and the difference between the actual number of risk events and the predicted number), and supplement the deviation data and new risk event records to the historical database; through periodic (such as daily / weekly) incremental learning, update the path features, efficiency indicators and risk weights in the knowledge base to optimize the accuracy of subsequent path planning.

[0115] S4 Visual Feedback: The monitoring interface of the visual management platform displays the real-time status of the route execution (such as current location, remaining distance, and estimated arrival time), and provides early warnings for behaviors that deviate from the optimal route (such as drivers manually changing routes), assisting managers in timely intervention.

[0116] Specifically, in the historical data collection step, the transportation trajectory data also includes distance information of surrounding obstacles (such as the minimum distance to roadside guardrails and buildings) obtained through vehicle-mounted LiDAR, which is used to supplement the limiting features of the path passage space; the task node data further includes the specific requirements of loading / unloading operations (such as whether auxiliary hoisting equipment is needed and the way the goods are fixed) to associate the site adaptability requirements of the path selection.

[0117] Specifically, in the data preprocessing step, feature extraction also includes feature importance analysis based on machine learning algorithms (such as random forests) to screen out the key features that have the greatest impact on path efficiency and risk (such as "cargo weight > 30 tons and turning radius < 20 meters" has an impact weight of 0.7 on the probability of accidents), and prioritize retaining these high-impact features for subsequent knowledge base construction.

[0118] Specifically, in the knowledge base construction process, the comprehensive performance evaluation model also includes user preference parameters (such as enterprises prioritizing time costs over fuel costs). The proportions of efficiency, risk, and cost are dynamically adjusted through configurable weight coefficients (α, β, γ) to meet the personalized needs of different enterprises.

[0119] Specifically, in the path matching and optimization steps, the similarity matching algorithm also includes deep learning-based embedding representation methods (such as converting textual features (e.g., operation area descriptions, road names) into vectors using the BERT model, and then calculating the similarity with the textual features of the current task) to improve the matching accuracy of complex scene descriptions.

[0120] Specifically, in the path matching and optimization steps, dynamic optimization also includes predicting the future congestion trend of road segments based on real-time traffic prediction models (such as LSTM-based 15-minute traffic flow prediction) and adjusting the path in advance to avoid impending congestion.

[0121] Specifically, the feedback from actual execution data also includes actual measured values ​​of environmental parameters (such as actual wind speed and rainfall during driving) to correct the environmental risk prediction model in the knowledge base (such as adjusting the historical statistical value of "accident probability when wind speed > 10m / s").

[0122] Specifically, in the output and feedback steps, the knowledge base iteration also includes filtering abnormal data (such as erroneous wind speed records caused by sensor failure). By setting a reasonable threshold range (such as the normal wind speed range of 0-50m / s), obviously unreasonable data is removed to ensure the reliability of the knowledge base data.

[0123] Specifically, in the output and feedback steps, visual feedback also includes providing managers with detailed analysis reports on path optimization (such as the composition of the comprehensive effectiveness score of the recommended path, comparative analysis with alternative paths, and execution references for similar historical tasks), to help decision-makers understand the scientific basis for path selection.

[0124] Through the above technical solution, this invention, based on in-depth mining of historical data to construct a knowledge base, completely changes the traditional path planning model that relies on human experience. By collecting and analyzing multi-dimensional data such as historical transportation trajectories, task nodes, environmental parameters, operation timeliness, and safety events of hoisting operations, big data mining and machine learning algorithms are used to extract the optimal path patterns in typical scenarios. This makes path planning no longer based on subjective judgment, but supported by a large amount of actual data, enabling accurate identification of efficient path characteristics in different operation scenarios (such as short-distance hoisting within cities, long-distance hoisting across regions, and hoisting in special terrain). For example, the system can analyze the optimal driving route for a specific time period, specific area, certain type of goods, and vehicle configuration, avoiding the limitations and errors that may exist in human experience, providing a scientific and quantifiable decision-making basis for hoisting operations, and significantly improving the scientific nature and accuracy of path planning. This method has strong dynamic adaptability and can respond to changes in the operation environment in real time. By combining real-time environmental data (such as current traffic flow, weather warnings, and road construction information) with task parameters (such as origin and destination, cargo type, time window, and vehicle / lifting equipment configuration), the system can intelligently match or generate the optimal lifting operation route from the knowledge base and dynamically adjust it according to real-time changes. In the face of emergencies, such as severe weather like heavy rain and strong winds, temporary traffic control, or road construction, the system can quickly identify risks and replan the route, effectively avoiding risky sections. For example, when it detects that the road ahead is prone to flooding or landslides due to heavy rain, the system can immediately adjust the route, guiding vehicles to avoid dangerous areas and choose a safer route. This dynamic adjustment capability greatly improves the safety of lifting operations, reduces the probability of accidents caused by emergencies, and also ensures the timeliness of operations, reducing losses due to delays. A feedback mechanism enables continuous optimization of the route planning capability. During the process of outputting the optimal route and guiding vehicles to execute the operation, the system collects real-time route execution data during the actual operation, including actual driving trajectory, time deviation, risk events, and driver operation behavior data, and feeds this data back to the historical database. By analyzing and learning from this feedback data, the knowledge base and optimization model can be continuously iterated and updated, allowing the route planning strategy to evolve continuously with the accumulation of operational experience. For example, with the accumulation of operational data in more different scenarios, the system can more accurately predict the comprehensive effectiveness of different routes under various conditions, further optimizing route selection. Furthermore, this method is deeply integrated with a transportation and hoisting operation visualization and control platform, allowing route instructions to be directly pushed to the driver and monitoring terminals, achieving closed-loop management of "planning-execution-feedback," promoting synergistic efficiency among various links, and comprehensively improving the overall efficiency, safety, and economic benefits of hoisting operations.

[0125] This specific implementation uses a large construction group and its cooperating hoisting and transportation companies as an application scenario. The group's business covers urban high-rise building construction, cross-regional bridge projects, and industrial plant hoisting, with an average of more than 50 hoisting operations per day, involving various complex scenarios such as narrow urban roads, cross-river bridges, and mountainous construction sites. Traditional route planning relies on dispatcher experience and ordinary navigation tools, often resulting in detours, delays, and even safety accidents due to unfamiliarity with site restrictions (such as a 4.5-meter height limit at a construction site entrance), sudden traffic control (such as road closures around examination centers during the college entrance examination period), or severe weather (such as typhoon season with gusts reaching level 10). (In the past year, improper route selection has caused 3 equipment scratch accidents, with an average loss of over 50,000 yuan per accident). To address this, the group has deployed the intelligent scheduling optimization method for transportation and hoisting based on multi-dimensional screening and trajectory analysis described in this invention, and deeply integrated it with its self-developed "Transportation and Hoisting Operation Visual Management and Control Platform" to form a closed-loop management system of "data collection - intelligent planning - real-time execution - feedback optimization".

[0126] I. Historical Data Acquisition: Comprehensive Perception of Multi-Source Heterogeneous Data

[0127] This method deploys multiple types of data acquisition terminals to cover key information throughout the entire hoisting operation process, ensuring the integrity and accuracy of historical data. Specific data collection content is as follows:

[0128] 1. Transportation trajectory data

[0129] Terminal equipment: Each hoisting and transport vehicle (including cranes, flatbed trucks, etc.) is equipped with a dual-mode positioning terminal (GPS + Beidou, positioning accuracy ≤ 5 meters, supporting differential positioning correction), an inertial measurement unit (IMU, sampling frequency 10Hz, collecting acceleration and angular velocity) and an on-board OBD (on-board diagnostic system, acquiring auxiliary information such as engine speed and fuel consumption).

[0130] Data collected includes: real-time vehicle latitude and longitude coordinates (uploaded every 5 seconds), driving speed (km / h), steering angle (°), driving timestamp (UTC time), vehicle attitude (tilt angle ≤3° is normal, >5° triggers an abnormal alarm), and rapid acceleration / sudden braking events (acceleration >2m / s² is recorded as abnormal driving behavior).

[0131] Example: In a certain tower crane component transportation task, the trajectory data of the vehicle from the suburban building materials market (loading point) to the construction site in the city center (unloading point) contains 2345 coordinate points, recording the speed changes when passing through 3 main roads, 2 branch roads and 1 bridge with a height limit of 8 meters (such as decelerating to 20km / h before the bridge and accelerating to 40km / h after passing it).

[0132] 2. Task node data

[0133] Data source: Structured information is input by the dispatcher or project manager through the task entry module of the transportation and hoisting operation visualization management platform, including:

[0134] Loading / unloading point: precise latitude and longitude (error ≤ 3 meters), site type (hardened ground / temporary dirt road / indoor factory), loading and unloading equipment configuration (e.g., whether a 25-ton auxiliary crane is provided), site restrictions (e.g., entrance width ≥ 4 meters, height limit ≤ 5 meters).

[0135] Route points: key nodes that must be passed (such as checkpoints on cross-river bridges, the starting point of traffic restrictions on municipal roads), and "suggested detour points" reported by drivers (such as sections of road under construction recently).

[0136] Time window: planned loading time (e.g., 8:00-8:30), planned unloading time (e.g., 10:00-10:30), allowable delay time (e.g., no more than 30 minutes), task priority (urgent tasks should be assigned the optimal path first).

[0137] Example: The task node data for a concrete pump truck transportation mission clearly requires that the loading point be mixing plant A (latitude and longitude: XX.XXXX, XX.XXXX, site is hardened cement ground, equipped with a 10-ton crane), and the unloading point be construction site B (latitude and longitude: XX.XXXX, XX.XXXX, entrance height limit 4.8 meters, unloading must be completed before 12:00 to avoid concrete solidification).

[0138] 3. Environmental parameter data

[0139] Multi-source acquisition:

[0140] Vehicle-mounted sensors: Each vehicle is equipped with a weather station (wind speed sensor accuracy ±0.5m / s, range 0-60m / s; rainfall sensor accuracy ±0.1mm / h, range 0-200mm / h; temperature sensor accuracy ±1℃, range -20℃-60℃) and an inclinometer (measurement range ±15°, accuracy ±0.2°).

[0141] External interfaces: Access to real-time traffic API from traffic management departments (obtain congestion index levels of 0-10 for each road segment, with level 10 indicating complete blockage; real-time speed limit adjustment information), regional weather forecasts from meteorological bureaus (probability of rainfall and wind speed change trends for the next 2 hours), and on-site data from construction site management systems (such as temporary closure notices for internal roads within the construction site and locations of newly added obstacles).

[0142] Example: During a hoisting operation, the onboard weather station detected a sudden increase in wind speed to 12 m / s 5 kilometers ahead of the route (exceeding the safe operating wind speed threshold of 10 m / s for cranes). At the same time, the traffic API showed that the congestion index of this section of the road reached level 8 (average vehicle speed < 10 km / h) due to a traffic accident.

[0143] 4. Work timeliness data

[0144] Automatic recording: The platform marks key nodes with timestamps, including:

[0145] Actual travel time: The time from departure from the loading point to arrival at the unloading point (accurate to the second), with travel time recorded in segments for each road segment (e.g., "urban section: 15 minutes, highway section: 20 minutes").

[0146] Waiting time: loading wait (e.g., goods not ready), unloading wait (e.g., equipment not in place at the construction site), traffic control wait (e.g., red lights or temporary road closures).

[0147] Total task completion time: The total time from task creation (dispatcher enters order) to final completion (driver uploads unloading photos).

[0148] Example: The actual travel time for a certain equipment transportation task was 42 minutes (planned time was 35 minutes). The time spent on securing the goods at the loading point was 10 minutes (5 minutes longer than planned), resulting in an overall delay of 8 minutes for the task.

[0149] 5. Security Incident Data

[0150] Multi-dimensional recording:

[0151] Video surveillance: The vehicle-mounted camera (1080P front view, 720P side view, wide-angle view from the cab) records driving footage in real time and uses AI algorithms to identify collisions (such as contact with other vehicles / objects), cargo deviation (such as the spreader swaying and causing the cargo to go beyond the edge of the cargo compartment), and road obstacles (such as potholes).

[0152] Sensor alarms: Risk events triggered by the crane's torque limiter (overload alarm), brakes (abnormal wear alarm), and tilt sensor (vehicle tilt > 5° alarm).

[0153] Manual reporting: Accident descriptions (such as "scratched the roadside guardrail while turning due to the narrow road") and on-site photos submitted by the driver through the vehicle terminal.

[0154] Example: During a hoisting and transportation operation, video surveillance captured a vehicle scraping its right rearview mirror against a utility pole while passing through a narrow side road (damage level: minor, repair cost < 1000 yuan). At the same time, the tilt sensor recorded that the vehicle tilted at an angle of 6° (exceeding the safety threshold of 5°).

[0155] II. Data Preprocessing: From Raw Data to Structured Features

[0156] The collected raw data suffers from noise, missing data, and inconsistent formats. It needs to be cleaned, transformed, and feature-engineered to extract features valuable for path optimization. The specific process is as follows:

[0157] 1. Data Cleaning

[0158] Location data correction: Remove abnormal points where the straight-line distance between three or more consecutive coordinate points exceeds the vehicle's maximum speed multiplied by the time interval (e.g., when the vehicle speed is 40km / h, the theoretical maximum travel distance within 10 seconds is about 11.1 meters; if the distance between coordinate points is greater than 15 meters, it is determined to be drift).

[0159] Missing value filling: For task records with missing key fields (such as latitude and longitude of loading point), the dispatcher manually fills them in or links them with historical similar task data; for data with missing non-key fields (such as occasional failure of wind speed sensor), the average value of the previous and next time points is used to fill them in.

[0160] Outlier removal: For example, data with a driving speed >150km / h (exceeding the crane's design limits) or fuel consumption <0L / h (sensor error) are directly deleted.

[0161] 2. Data Standardization

[0162] Unified format: Convert all times to UTC timestamps (e.g., “2025-02-02 20:40:34” → 1738471234), coordinate system 1 is WGS84 (globally used GPS coordinate system), and units are standardized as km / h (speed), tons (weight), and meters (distance).

[0163] Encoding categorical variables: Textual features (such as region type: "urban core area" → 1, "suburbs" → 2, "inter-regional expressway" → 3; road type: "main road" → 1, "secondary road" → 2, "local road" → 3) are converted into numerical codes to facilitate subsequent calculations.

[0164] 3. Feature Extraction

[0165] Based on business logic and statistical methods, the following key features were extracted from the raw data (partial examples):

[0166] Characteristics of transportation scenarios:

[0167] Regional types (Urban core area: population density > 10,000 people / km² and road width < 20 meters; Suburbs: population density < 5,000 people / km²; Inter-regional expressways: national highways / expressways connecting different cities).

[0168] Traffic hours (peak hours: weekdays 7:00-9:00 and 17:00-19:00; off-peak hours: other times).

[0169] Task characteristics:

[0170] Cargo weight class (≤10 tons: light equipment; 10-30 tons: medium equipment; 30-50 tons: heavy equipment; >50 tons: extra-heavy equipment).

[0171] Vehicle type (light crane: maximum lifting capacity < 25 tons; medium crane: 25-50 tons; heavy crane: > 50 tons).

[0172] Path characteristics:

[0173] Total path length (calculated by summing Euclidean distances from a sequence of coordinate points, unit: kilometers).

[0174] Maximum turning radius (calculated based on vehicle type and boom extension status; for example, the minimum turning radius of a medium-sized crane with its boom fully extended must be ≥20 meters).

[0175] Number of restricted areas (such as school areas, hospital areas, narrow road sections (width < 8 meters)).

[0176] Risk characteristics:

[0177] Historical accident density (the number of accidents that occur in the total mileage of a certain route, in units of accidents / km).

[0178] High-risk weather correlation (Rainy day accident rate = number of accidents in rainy weather / total number of trips in rainy weather).

[0179] Efficiency characteristics:

[0180] Average driving speed (total driving distance / total driving time, unit: km / h).

[0181] On-time completion rate (number of tasks completed on time / total number of tasks in this path, unit: %).

[0182] 4. Data classification and storage

[0183] The preprocessed data is categorized and stored in a relational database (MySQL) and a time-series database (InfluxDB) according to the following dimensions:

[0184] Classified by scenario: short-distance hoisting within the city (one-way ≤ 50 km), long-distance hoisting across regions (one-way > 50 km), and hoisting in special terrain (mountainous / tunnel / bridge-dense areas).

[0185] By task priority: Urgent tasks (must be completed within 2 hours), routine tasks (2-24 hours), low priority tasks (>24 hours).

[0186] According to cargo weight class: light (≤10 tons), medium (10-30 tons), heavy (>30 tons).

[0187] III. Knowledge Base Construction: Transforming Data into Intelligent Decision-Making

[0188] Based on the preprocessed historical database, path optimization patterns in typical scenarios are mined through the fusion of multiple algorithms, and a knowledge base supporting fast retrieval and dynamic adaptation is constructed. The specific steps are as follows:

[0189] 1. Cluster analysis: Identifying typical work scenarios

[0190] An improved DBSCAN (density-based clustering algorithm) is used, comprehensively considering transportation scenario characteristics (region type + road type + traffic time), task characteristics (cargo weight + vehicle type), and environmental characteristics (weather type) to divide historical tasks into several typical scenario clusters. For example:

[0191] Cluster 1: Urban core area + weekday peak hours (7:30-8:30) + 10-30 tons of goods + medium-sized crane + rainy days;

[0192] Cluster 2: Cross-regional highway + off-peak hours + 30-50 tons of cargo + heavy crane + sunny weather;

[0193] Cluster 3: Special terrain (mountainous areas) + any time period + super heavy equipment (>50 tons) + special cranes + non-rainy days.

[0194] Each cluster represents a class of job scenarios with similar constraints, providing a foundation for subsequent path pattern extraction.

[0195] 2. Association rule mining: Discovering strong correlations between paths and performance.

[0196] By using FP-Growth (Frequent Pattern Growth Algorithm), we can uncover the association rules between "path features and efficiency metrics" and "path features and risk metrics" in typical scenarios. For example:

[0197] Rule 1 (Efficiency Oriented): If a route contains ≥2 secondary arterial roads (width ≥10 meters) and passes through ≤1 restricted area, the average travel time is 20%-30% shorter than a route containing arterial roads (92% confidence level).

[0198] Rule 2 (Risk-Oriented): If the maximum turning radius of the route is less than 15 meters and the cargo weight is greater than 30 tons, the probability of an accident increases to 8% (far higher than the baseline of 2%) (confidence level 89%).

[0199] These rules reveal the business logic behind "choosing secondary roads to avoid peak traffic hours can improve efficiency" or "avoiding small-turn-radius sections of road when transporting heavy equipment can reduce risk".

[0200] 3. Temporal pattern recognition: Adapting to dynamic environmental changes

[0201] By using LSTM (Long Short-Term Memory) networks to analyze the time-series characteristics of historical tasks, the optimal path selection tendency at different time periods can be identified. For example:

[0202] During weekday morning rush hour (7:00-9:00), the average congestion index of the city's main roads reaches level 7. At this time, choosing to detour via a secondary road (adding 2 kilometers but with a congestion index of only level 3) is more efficient.

[0203] During the rainy season (June-August), the accident rate on routes that pass through bridges increases by 40%, and the system will prioritize recommending alternative routes that do not pass through bridges.

[0204] 4. Comprehensive Performance Evaluation Model: Quantifying the Advantages and Disadvantages of Different Paths

[0205] A multi-objective optimization model is constructed, using efficiency (time saving rate), risk (accident probability), and cost (fuel consumption) as core indicators to calculate the overall efficiency score of the route. The formula is as follows:

[0206] Overall efficiency = Efficiency weight (α) × Time saving rate + Risk weight (β) × (-accident probability) + Cost weight (γ) × (-fuel consumption per unit mile)

[0207] Among them, α, β, and γ are configurable parameters (the default values ​​are α=0.6, β=0.3, and γ=0.1, which can be adjusted by enterprises according to their needs. For example, cost-sensitive enterprises can increase γ to 0.3).

[0208] This model scores candidate paths for each typical scenario, for example:

[0209] Route A (Secondary Road): Time saving rate 15% (score +9), accident probability 1% (score +2.7), fuel consumption per unit mileage 0.8L / km (score +0.8), overall efficiency = 0.6×9 + 0.3×2.7 + 0.1×0.8 ≈ 6.21;

[0210] Route B (Main Road): Time saving rate 5% (score +3), accident probability 3% (score +0.9), fuel consumption per unit mileage 1.0L / km (score +0), overall efficiency = 0.6×3 + 0.3×0.9 + 0.1×0 ≈ 2.07.

[0211] Clearly, path A has higher overall efficiency and will be recommended as the optimal path in this scenario.

[0212] IV. Current Task Analysis: Extraction of Precise Requirements and Real-Time Constraints

[0213] When a new hoisting task is received, the system obtains task parameters and real-time environmental data through a visual management platform and generates path planning constraints. The specific process is as follows:

[0214] 1. Task Parameter Analysis

[0215] Extract core information from the platform's task entry module:

[0216] Origin and destination coordinates: Loading point (XX.XXXX, XX.XXXX, building materials market, coordinate accuracy ±2 meters), unloading point (XX.XXXX, XX.XXXX, construction site entrance, coordinate accuracy ±2 meters).

[0217] Cargo information: Type (precast concrete beam, heavy equipment), weight (45 tons), dimensions (12 meters long × 2.5 meters wide × 3 meters high).

[0218] Vehicle / Equipment Configuration: Model (Heavy-duty crane, maximum lifting capacity 60 tons, minimum turning radius 22 meters when boom is extended).

[0219] Time window: planned loading time (8:00-8:30), planned unloading time (10:00-10:30), and allowed delay time (not exceeding 20 minutes).

[0220] Task priority: Urgent (due to tight construction schedule, priority must be given).

[0221] 2. Real-time environmental data acquisition

[0222] Real-time data acquisition via multi-source interfaces:

[0223] Traffic data: Congestion index along the current route (congestion index level 6 on the main road from loading point to unloading point, average speed 25km / h), real-time speed limit (temporary speed limit of 30km / h on some sections due to construction).

[0224] Meteorological data: There is a 60% probability of rain in the next 2 hours (current wind speed is 5 m / s, no strong wind warning), but the meteorological bureau suggests that there may be short-term light rain after 11:00 (rainfall < 5 mm / h).

[0225] Work site data: The entrance to the unloading point is currently unoccupied (normal passage is possible), but there are temporary protective sheds nearby (height limit of 4.5 meters, it needs to be confirmed whether the height of the goods is compliant).

[0226] 3. Task Constraint Generation

[0227] Based on task parameters and environmental data, the following constraints are generated:

[0228] Physical constraints: Minimum turning radius of vehicles is 22 meters (the turning radius of the route must be ≥22 meters), cargo height is 3 meters (bridges / tunnels with a clearance of <3.5 meters must be avoided), and maximum permissible slope is 12° (safety restrictions for heavy cranes).

[0229] Time constraints: Must depart before 8:30 (loading point) and arrive before 10:30 (unloading point), with a total travel time of ≤120 minutes (30 minutes allowed for loading / unloading).

[0230] Safety constraints: Avoid high-risk road sections (such as side roads with a historical accident density of >0.3 times / km) and avoid passing through sections prone to water accumulation during rainy weather (such as under a low-lying bridge).

[0231] V. Path Matching and Optimization: Intelligent Decision Making from Candidates to the Optimal

[0232] Based on the current task parsing results, the system matches the optimal path from the knowledge base and makes dynamic adjustments. The specific process is as follows:

[0233] 1. Similarity matching

[0234] A weighted cosine similarity algorithm is used to calculate the similarity between the current task parameters (cargo weight 45 tons, urban core area, weekday off-peak hours (after 9:00 AM), moderate rain) and historical scene features. For example:

[0235] The current task has a similarity score of 0.82 (higher than the threshold of 0.8) to the scenario of "urban core area + off-peak hours + 30-50 tons of goods + heavy crane + light rain". The optimal path in the corresponding knowledge base is "secondary arterial road combination (width ≥ 12 meters) + detour through 1 school area".

[0236] 2. Initial path screening

[0237] Filter candidate paths based on current constraints:

[0238] Exclude routes that include sections with a gradient greater than 12° (such as a mountain road route);

[0239] Exclude routes with an estimated travel time exceeding 120 minutes (including loading / unloading time) (such as detours on alternative highways).

[0240] Retain routes that meet physical constraints (turning radius ≥ 22 meters), time constraints (arrive before 10:30), and safety constraints (avoid high-risk road sections).

[0241] 3. Dynamic optimization

[0242] Further adjustments based on real-time environmental data:

[0243] Traffic dynamics: If the original planned secondary arterial road is found to have a congestion index of level 7 due to a sudden accident when departure, the system will automatically switch to a low-congestion backup branch road (width ≥ 10 meters, congestion index level 3) in the same area from the knowledge base.

[0244] Weather Update: If rainfall increases to 10 mm / h after 10:00 (exceeding the light rain threshold), avoid low-lying and easily flooded sections of the road (such as under a bridge where the historical water depth has reached 0.5 meters) and replan your detour route.

[0245] Vehicle status: If the driver reports that the vehicle has only enough fuel to travel 80 kilometers (the original route is 70 kilometers long but emergency fuel consumption needs to be reserved), then the route that passes a gas station should be selected first (such as adding 2 kilometers but ensuring that the fuel is sufficient).

[0246] 4. Comprehensive Performance Evaluation

[0247] The comprehensive performance of the two optimized candidate routes (Route X: combination of secondary arterial roads; Route Y: ring expressway + short-distance branch road) was evaluated (parameter configuration: α=0.6, β=0.3, γ=0.1):

[0248] Path X:

[0249] Time saving rate: 18% shorter than planned (score +10.8);

[0250] Accident probability: Historical data shows 0.8% (score +2.4);

[0251] Fuel consumption per unit distance: 0.75L / km (score +0.75);

[0252] Overall efficiency = 0.6 × 10.8 + 0.3 × 2.4 + 0.1 × 0.75 ≈ 7.29.

[0253] Path Y:

[0254] Time saving rate: 10% shorter than planned (score +6);

[0255] Accident probability: Historical data shows 1.5% (score +1.35);

[0256] Fuel consumption per unit mileage: 0.85L / km (score +0.45);

[0257] Overall efficiency = 0.6 × 6 + 0.3 × 1.35 + 0.1 × 0.45 ≈ 4.23.

[0258] Finally, route X (overall efficiency 7.29) was selected as the optimal hoisting operation route and pushed to the driver's navigation (displaying "Recommended route: via XX Road → XX Branch Road, estimated travel time 45 minutes, passing through 1 section with a speed limit of 30km / h").

[0259] VI. Output and Feedback: Closed-Loop Optimization and Continuous Improvement

[0260] During path execution, the system collects feedback data in real time and optimizes the knowledge base. The specific process is as follows:

[0261] 1. Path output and execution

[0262] Driver's side: Navigation instructions are pushed via the in-vehicle tablet (including voice prompts such as "Turn right into XX branch road 500 meters ahead" and speed limit reminders such as "Speed ​​limit of 30km / h in the current section") and risk warnings such as "School area 200 meters ahead, slow down to 20km / h".

[0263] Monitoring terminal: The visual management platform displays the vehicle's location in real time (accuracy ≤ 5 meters), estimated arrival time (dynamically updated), and route deviation (such as whether the driver is following the recommended route).

[0264] 2. Actual data collection

[0265] Execution data: actual driving trajectory (records 23 coordinate points along the way), driving time for each road segment (e.g., "XX Road: 8 minutes, XX Branch Road: 12 minutes"), actual arrival time (10:12, 18 minutes ahead of the planned 10:30).

[0266] Risk event: No collision, cargo deviation or other abnormalities (no alarms were triggered by video surveillance or sensors).

[0267] Actual environmental values: During operation, the actual wind speed was 6 m / s (below the crane's safety threshold) and the rainfall was 2 mm / h (light rain, consistent with the forecast).

[0268] 3. Knowledge base iteration

[0269] Deviation analysis: The actual comprehensive efficiency of the recommended route was calculated (actual driving time 42 minutes, 3 minutes shorter than the predicted 45 minutes; no accidents, actual accident probability 0%). It was found that there was a slight difference between the original predicted "fuel consumption per unit mile 0.75L / km" and the actual 0.72L / km, but the overall efficiency was better than expected.

[0270] Data Supplementation: Add the actual data of this task (such as "the actual driving time of this route during light rain and off-peak hours is 42 minutes") to the historical database, and update the knowledge base entries for the corresponding typical scenarios (such as adjusting the average driving time of this scenario to "23-25 ​​minutes").

[0271] Model optimization: Through daily incremental learning (automatically analyzing the actual execution data of all tasks from the previous day), optimize the parameters of the clustering algorithm (such as adjusting the density threshold of DBSCAN) and the weight coefficients of the comprehensive performance model (such as adjusting γ from 0.1 to 0.2 if the enterprise is more focused on cost recently).

[0272] 4. Visual feedback

[0273] On the "Route Optimization Analysis" interface of the visual management platform, managers can view a detailed report of this task:

[0274] Recommended route efficiency: Overall score 7.29 (ranked first), time saving rate 18%, accident probability 0%;

[0275] Alternative routes comparison: Route Y has an overall score of 4.23 (ranked second), with a time saving rate of 10%;

[0276] Historical reference: In the past 3 months, in similar scenarios (30-50 tons + urban core area + light rain), 85% of the tasks chose the secondary arterial road combination route, with an average on-time completion rate of 91%.

[0277] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their equivalents.

Claims

1. A method for intelligent scheduling and optimization of transportation and hoisting based on multi-dimensional screening and trajectory analysis, characterized by: This method systematically integrates multi-source heterogeneous data from historical hoisting operations, constructs a dynamically adaptable knowledge base, and combines real-time environment and task parameters to achieve scientific planning and dynamic optimization of hoisting operation paths. Specifically, it includes the following steps: Historical data collection steps: Multi-dimensional historical data throughout the entire lifecycle of the hoisting operation is collected using various types of data acquisition terminals. This historical data includes, but is not limited to: S1 transport trajectory data: vehicle latitude and longitude coordinate sequence (positioning accuracy ≤10 meters), driving speed, acceleration, direction angle, driving timestamp, and vehicle attitude data (such as tilt angle and vibration amplitude) collected by the vehicle-mounted high-precision GPS / BeiDou dual-mode positioning terminal. S2 Task Node Data: Based on the structured task information recorded by the transportation and hoisting operation visualization management platform, including loading point (latitude and longitude, site type, loading and unloading equipment configuration), unloading point (latitude and longitude, site type, loading and unloading equipment configuration), route points (such as necessary bridges, checkpoints), task time window (planned start time, planned end time, allowed delay time), and task priority (urgent / normal / low priority). S3 environmental parameter data: Multi-source environmental information collected through vehicle-mounted meteorological sensors (collecting real-time wind speed, wind direction, temperature, and humidity), road meteorological monitoring stations (collecting road section-level rainfall, snowfall, and visibility), traffic management department interfaces (obtaining real-time traffic flow, congestion index, and speed limit adjustment information), and construction site management systems (obtaining terrain features of the work site, such as slope, turning radius restrictions, and distribution of surrounding obstacles, including high-voltage line height, building height restrictions, and underground pipeline locations); S4 Operation Timeliness Data: Task execution time parameters automatically recorded by the platform, including actual travel time (total travel time from loading point to unloading point), travel time for each road segment (statistically divided by electronic map), waiting time (such as loading wait, unloading wait, traffic control wait), total task completion time (time span from task creation to final completion), and delay time of key nodes (such as loading timeout, unloading timeout). S5 safety incident data: Safety-related incidents recorded through vehicle video monitoring (identifying abnormal scenes such as collisions, scratches, and cargo deviation), sensor alarms (such as torque limiter over-limit and brake malfunction), and driver manual reporting (accident descriptions submitted through the vehicle terminal). These incidents include the type of accident (equipment damage / personnel injury / cargo loss), location (latitude and longitude + road segment markings), time of occurrence, risk factors (such as narrow roads, insufficient turning radius, speeding, and weather conditions), and degree of loss (minor / moderate / severe). Data preprocessing steps: Cleaning, transforming, and feature engineering the collected raw historical data, specifically including: S1 Data Cleaning: Remove data with location drift (such as more than 10 consecutive coordinate points with a straight-line distance exceeding the theoretical value of the vehicle's maximum speed), duplicate records, and data with missing key fields (such as missing latitude and longitude of loading points); S2 data standardization: unify the time format (convert to UTC timestamp), coordinate system (unify to WGS84 coordinate system), and units (e.g., speed is unified to km / h, weight is unified to tons) of different data sources. S3 Feature Extraction: Based on business logic and statistical methods, extract key features from raw data, including but not limited to: Transportation scenario characteristics: Operation area type (urban core area / suburbs / inter-regional expressway), road type (main road / secondary road / local road / dedicated hoisting channel), traffic time (weekdays / holidays, peak hours / off-peak hours); Task characteristics: Cargo type (general building materials / heavy equipment / precision instruments), cargo weight class (≤10 tons / 10-30 tons / 30-50 tons / >50 tons), vehicle type (light crane / medium crane / heavy crane / customized crane), lifting equipment configuration (whether auxiliary lifting tools are carried). Route characteristics: total route length (km), total number of turns, maximum turning radius (m), gradient variation range (minimum gradient - maximum gradient), number of restricted areas (e.g., schools, hospitals, narrow sections); Risk characteristics: historical accident density (number of accidents per unit mileage), high-risk road section markings (such as bridges with a gradient > 8° and no anti-skid treatment), and weather sensitivity (accident incidence rate on rainy / snowy / windy days). Efficiency characteristics: average driving speed (km / h), on-time completion rate (number of tasks completed on time / total number of tasks), fuel consumption per unit mile (liters / km, related to fuel cost). S4 data classification and storage: Preprocessed data is classified and stored according to operation scenarios (such as short-distance hoisting in the city (one way ≤ 50 kilometers), long-distance hoisting across regions (one way > 50 kilometers), hoisting in special terrain (mountainous / tunnel / bridge-dense areas)), task priority, and cargo weight class, to build a structured historical database of hoisting operation paths; Knowledge base construction steps: Based on the preprocessed historical database, big data mining technology with multi-algorithm fusion is used to extract path optimization patterns under typical operation scenarios, and a hoisting operation path knowledge base containing path features, applicable conditions, risk weights, and efficiency indicators is constructed, specifically including: S1 Cluster Analysis: Using K-means++ or DBSCAN algorithms, historical tasks are clustered based on transportation scenario characteristics (such as region type + road type + traffic time period) and task characteristics (such as cargo weight + vehicle type) to identify a set of typical operation scenarios with similar characteristics (such as "urban core area + weekday peak hours + 30-50 tons of cargo + medium-sized crane" as a typical scenario). S2 Association Rule Mining: Using Apriori or FP-Growth algorithms, we mine strong association rules between "path features - efficiency indicators" and "path features - risk indicators" in typical scenarios. For example, "when the maximum turning radius of the path is >20 meters and the number of restricted areas passed through is ≤2, the probability of accidents is reduced by 40%" and "the transportation time of choosing a secondary road (width ≥10 meters) is reduced by an average of 25% compared to the main road, but the fuel consumption increases by 10%". S3 Temporal Pattern Recognition: Utilizes LSTM (Long Short-Term Memory Network) or ARIMA (Autoregressive Integral Moving Average Model) to analyze the time series characteristics of historical tasks (such as traffic flow patterns and seasonal weather effects during specific time periods) and extracts the optimal path selection tendencies for different time periods (such as morning and evening rush hours on holidays, rainy season / winter). S4 Efficiency and Risk Assessment Model: Construct a comprehensive evaluation model with efficiency weights (e.g., time saving rate = theoretical shortest time / actual driving time), risk weights (e.g., accident probability = number of historical accidents / total number of trips on the route), and cost weights (e.g., fuel cost = fuel consumption per unit mile × driving mileage) as core indicators. A multi-objective optimization algorithm (e.g., NSGA-II non-dominated sorting genetic algorithm) is used to calculate the comprehensive performance score of different paths (comprehensive performance = efficiency weight × α + risk weight × β + cost weight × γ, where α, β, and γ are configurable weight coefficients, with default values ​​of α = 0.6, β = 0.3, and γ = 0.1). The path with the optimal comprehensive performance is then selected as the recommended path in the knowledge base. The S5 knowledge base uses structured storage: it stores the optimal path features (such as origin and destination coordinates, key points along the route, and road type combinations), applicable conditions (such as cargo weight range and weather conditions), risk weights (such as the proportion of high-risk road sections), and efficiency indicators (such as average travel time and on-time completion rate) corresponding to each typical scenario in the knowledge base in a structured data format (such as JSON / XML), and establishes a mapping index between scenario features and path solutions to support fast retrieval; Current task parsing steps: When a new hoisting operation task is received, the system obtains detailed parameters and real-time environmental data of the current task through the visual management and control platform, specifically including: S1 Task Parameter Analysis: Extract the core information of the task, including the precise coordinates of the origin and destination (loading point and unloading point), cargo type (associated with attributes through material code), cargo weight (tons), planned time window (earliest allowed departure time and latest allowed arrival time), vehicle / lifting equipment configuration (vehicle type, maximum lifting capacity, boom length), and task priority (urgency level). S2 Real-time Environmental Data Acquisition: Collects current environmental information in real time through multi-source data interfaces, including: Real-time traffic data: Current traffic flow (vehicles / hour), congestion index (0-10), real-time speed limits (km / h) for each road segment, and accident hotspots (e.g., a rear-end collision 2 kilometers ahead causing lane closure). Real-time meteorological data: real-time wind speed (m / s, with a focus on strong winds >8m / s), rainfall (mm / h, with a focus on heavy rain >10mm / h), snowfall (cm / h, with a focus on snow accumulation >2cm / h), and visibility (meters, with a focus on low visibility <500 meters) along the operation path. Work site data: Real-time site status of unloading / loading points (e.g., whether they are occupied by other equipment, whether the ground is slippery), and surrounding temporary restriction information (e.g., temporary road closures due to road construction, and newly added height restriction barriers); S3 Task Constraint Generation: Based on task parameters and environmental data, generate path planning constraints for the current task, including: Physical constraints: maximum permissible driving gradient for vehicles (e.g., heavy cranes ≤ 15°), minimum turning radius (e.g., ≥ 25 meters when the boom is extended), and the degree of matching between cargo weight and road load-bearing capacity (e.g., bridge weight limit > total cargo weight). Time constraints: Transportation must be completed within the planned time window (e.g., departure time from loading point ≥ 8:00 and arrival time at unloading point ≤ 10:00). Safety constraints: Avoid high-risk road sections (such as road sections with a historical accident density > 0.5 times / km) and avoid passing through areas with a risk of falling objects from heights in severe weather (such as wind speed > 12m / s); Path matching and optimization steps: Based on the current task parsing results, the optimal path is intelligently matched from the knowledge base and dynamically adjusted, specifically including: S1 Similarity Matching: Using a multi-dimensional similarity calculation algorithm, the current task parameters (such as cargo weight class, planned time window, origin and destination areas, and traffic hours) are compared with historical scene features (such as similar cargo transportation time periods, task time distribution in the same area, and similar road type combinations). A similarity score is calculated, and typical scenes with similarity scores higher than a preset threshold (such as ≥0.8) and their corresponding optimal paths are selected as a set of candidate paths. The similarity calculation can use weighted Euclidean distance (logarithmic feature) or cosine similarity (categorical feature), and the weights are dynamically adjusted according to the importance of the features (such as task priority weight > cargo weight weight). S2 route initial screening: Based on the constraints of the current task (such as physical constraints, time constraints, and safety constraints), the candidate route set is initially filtered to remove routes that do not meet the conditions (such as routes containing road sections with a gradient greater than the maximum allowable gradient for vehicles, or routes whose estimated travel time exceeds the planned time window). S3 Dynamic Optimization: For candidate paths that pass the initial screening, real-time environmental data (such as current traffic congestion index and weather warnings) and real-time vehicle status (such as current speed and remaining fuel) are combined with reinforcement learning models (such as Deep Q-Network, DQN) or dynamic programming algorithms (such as an improved version of Dijkstra's algorithm) to adjust the paths in real time. If a road segment is detected to have a real-time congestion index greater than level 7 (severe congestion), it will be automatically replaced with a low-congestion alternative route in the same area from the knowledge base. If there is a temporary weather risk ahead (such as a sudden increase in wind speed to 10m / s), then avoid that section of road and replan the detour route. If the vehicle's current remaining fuel is insufficient to support the original route (estimated based on historical fuel consumption data), then the route that passes through a gas station or service area will be selected first. S4 Comprehensive Performance Evaluation: The optimized path set is scored comprehensively (based on a multi-objective optimization model of efficiency, risk, and cost), and the path with the highest comprehensive performance score is selected as the final recommended optimal hoisting operation path; if multiple paths have similar scores (e.g., difference < 5%), the path with the highest historical on-time completion rate is selected first. Output and Feedback Steps: The finalized optimal route is pushed to the driver's navigation device (such as an in-vehicle tablet or mobile app) and the monitoring management interface through the visual management platform. Simultaneously, a real-time feedback mechanism for route execution is established, specifically including: S1 Route Output: Provides detailed route navigation instructions to the driver (including turn prompts, speed limit reminders, and risk section warnings), and generates a route execution plan to the monitoring terminal (including estimated travel time for each segment and time predictions for key nodes). S2 Real-time Data Acquisition: During the route execution process, real-time driving data is continuously collected, including the actual driving trajectory (real-time positioning via GPS / BeiDou), the actual driving time of each road segment, the actual arrival time, whether any risk events occur (such as deviation from the planned route or encountering a sudden accident), and driver operation behavior (such as the frequency of rapid acceleration and sudden braking). S3 Knowledge Base Iteration: Compare and analyze the actual execution data with the original task parameters and recommended paths, calculate the path execution deviation (such as the difference rate between actual driving time and predicted time, and the difference between the actual number of risk events and the predicted number), and supplement the deviation data and new risk event records to the historical database; through periodic (such as daily / weekly) incremental learning, update the path features, efficiency indicators and risk weights in the knowledge base to optimize the accuracy of subsequent path planning. S4 Visual Feedback: The monitoring interface of the visual management platform displays the real-time status of the route execution (such as current location, remaining distance, and estimated arrival time), and provides early warnings for behaviors that deviate from the optimal route (such as drivers manually changing routes), assisting managers in timely intervention.

2. The intelligent scheduling optimization method for transportation and hoisting based on multi-dimensional screening and trajectory analysis according to claim 1, characterized in that: In the historical data collection step, the transportation trajectory data also includes distance information of surrounding obstacles (such as the minimum distance to roadside guardrails and buildings) obtained by vehicle-mounted LiDAR, which is used to supplement the limiting features of the path passage space. The task node data further includes specific requirements for loading / unloading operations (such as whether auxiliary hoisting equipment is needed and the method of securing the goods) to associate with the site adaptability requirements of the path selection.

3. The intelligent scheduling optimization method for transportation and hoisting based on multi-dimensional screening and trajectory analysis according to claim 1, characterized in that: In the data preprocessing step, the feature extraction also includes feature importance analysis based on machine learning algorithms (such as random forests) to screen out the key features that have the greatest impact on path efficiency and risk (such as "cargo weight > 30 tons and turning radius < 20 meters" has an impact weight of 0.7 on the probability of accidents), and prioritize retaining these high-impact features for subsequent knowledge base construction.

4. The intelligent scheduling optimization method for transportation and hoisting based on multi-dimensional screening and trajectory analysis according to claim 1, characterized in that: In the knowledge base construction step, the comprehensive performance evaluation model also includes user preference parameters (such as enterprises prioritizing time costs over fuel costs), and dynamically adjusts the proportions of efficiency, risk, and cost through configurable weight coefficients (α, β, γ) to meet the personalized needs of different enterprises.

5. The intelligent scheduling optimization method for transportation and hoisting based on multi-dimensional screening and trajectory analysis according to claim 1, characterized in that: In the path matching and optimization steps, the similarity matching algorithm also includes a deep learning-based embedding representation method (such as converting textual features (e.g., operation area description, road name) into vectors using the BERT model, and then calculating the similarity with the textual features of the current task) to improve the matching accuracy of complex scene descriptions.

6. The intelligent scheduling optimization method for transportation and hoisting based on multi-dimensional screening and trajectory analysis according to claim 1, characterized in that: In the path matching and optimization step, the dynamic optimization also includes predicting the future congestion trend of road segments based on real-time traffic prediction models (such as LSTM-based 15-minute traffic flow prediction) and adjusting the path in advance to avoid impending congestion.

7. The intelligent scheduling optimization method for transportation and hoisting based on multi-dimensional screening and trajectory analysis according to claim 1, characterized in that: The feedback of the actual execution data also includes the actual measured values ​​of environmental parameters (such as the actual wind speed and rainfall during the driving process), which are used to correct the prediction model of environmental risks in the knowledge base (such as adjusting the historical statistical value of "accident probability when wind speed > 10m / s").

8. The intelligent scheduling optimization method for transportation and hoisting based on multi-dimensional screening and trajectory analysis according to claim 1, characterized in that: In the output and feedback steps, the knowledge base iteration also includes filtering abnormal data (such as erroneous wind speed records caused by sensor failure). By setting a reasonable threshold range (such as the normal wind speed range of 0-50m / s), obviously unreasonable data is removed to ensure the reliability of the knowledge base data.

9. The intelligent scheduling optimization method for transportation and hoisting based on multi-dimensional screening and trajectory analysis according to claim 1, characterized in that: In the output and feedback steps, the visual feedback also includes providing managers with a detailed analysis report on path optimization (such as the composition of the comprehensive performance score of the recommended path, a comparative analysis with alternative paths, and a reference for the execution of similar historical tasks), to help decision-makers understand the scientific basis for path selection.