Search

CN-122022299-A - Multi-mode port logistics self-adaptive scheduling system and method

CN122022299ACN 122022299 ACN122022299 ACN 122022299ACN-122022299-A

Abstract

The invention discloses a multi-mode port logistics self-adaptive scheduling system and a method, wherein the system comprises a data acquisition module, a semantic alignment module, a situation analysis module, a decision generation module, a multi-objective optimization module, an instruction generation module and an execution monitoring module; the multi-mode data acquired by the data acquisition module are generated to be uniformly represented by the semantic alignment module, the situation analysis module calculates KPIs based on the uniform representation and detects abnormality, the decision generation module generates scheduling decisions according to situation information, the multi-objective optimization module optimizes the decisions to obtain the multi-mode data, the instruction generation module converts the optimization scheme into executable instructions, the feedback of the monitoring module is returned to the data acquisition layer to update a data source, and meanwhile, the historical corpus is updated for continuous learning of the decision model. According to the method, the port operation self-adaptive scheduling is realized by constructing a port space-time semantic model, fusing multi-source heterogeneous data, analyzing the real-time situation and generating intelligent decisions.

Inventors

  • LI YIPENG
  • SHAO XINQING
  • WU HAO
  • ZHANG PING
  • ZHOU HONGWEI

Assignees

  • 江苏润和软件股份有限公司

Dates

Publication Date
20260512
Application Date
20260122

Claims (9)

  1. 1. A multi-mode port logistics self-adaptive scheduling system is characterized by comprising a data acquisition module, a semantic alignment module, a situation analysis module, a decision generation module, a multi-target optimization module, an instruction generation module and an execution monitoring module, wherein the multi-mode data are acquired by the data acquisition module Generating unified representations via semantic alignment module The situation analysis module is based on Calculating KPI and detecting abnormality, and generating scheduling decisions by a decision generation module according to situation information The multi-objective optimization module optimizes the decision to obtain The instruction generation module converts the optimization scheme into an executable instruction, the feedback of the execution monitoring module flows back to the data acquisition layer to update the data source, the historical corpus is updated for continuous learning of the decision model, and a complete closed loop is formed between the modules through the data flow and the feedback mechanism.
  2. 2. A multi-modal port logistics adaptive scheduling method based on the system of claim 1, characterized by comprising the following steps: S1, multi-source data acquisition and space-time alignment; s2, multi-mode semantic alignment driven by a port business body; S3, real-time analysis and anomaly detection of port situation; s4, generating an instruction type scheduling decision and checking constraint; S5, congestion-energy consumption collaborative optimization; S6, generating a bimodal scheduling instruction and cooperating with a man-machine; S7, executing closed loop and rolling optimization.
  3. 3. The method of claim 2, wherein in step S1, the port multi-mode data source including video data stream, AIS ship track data, unmanned aerial vehicle inspection data, ioT sensor data, voice intercom data, business system number, weather and tide data is accessed through the data acquisition layer, and the unified time axis is defined by performing time-space alignment processing first All data are time stamped based on UTC time Indexing, establishing a harbor district unified coordinate system for the space data Uniformly converting AIS longitude and latitude, video pixel coordinates and equipment GPS coordinates into A coordinate system; the multi-modal data after the time-space alignment is expressed as: Wherein, the For the moment of time Is a video frame set of (a); for the moment of time AIS trace point set; for the moment of time Is a set of sensor readings; for the moment of time Is a text data set of (2); for the moment of time Is a weather tide data.
  4. 4. The method for adaptive scheduling of multi-modal port logistics according to claim 3, wherein in step S2, a port service ontology library is constructed for achieving semantic fusion of collected multi-modal data The system comprises loading and unloading links, equipment types, goods attribute, operation state and relation thereof, wherein a body library is expressed in a form of triples: Wherein, the The entity set comprises equipment entities, cargo entities and regional entities; is a set of relationships, including "serving," "located," "dependent" relationships; Is an entity in the ontology library; Is an entity And (3) with A relationship between; Based on the ontology library, the multi-modal data after space-time alignment is aligned Scene level annotation for video frames Identifying equipment, container and vehicle entity in the picture by using the target detection model and associating the equipment, container and vehicle entity with the corresponding entity in the body library, and identifying the text data Extracting equipment numbers, box numbers and position entities by using a named entity recognition model, and mapping the equipment numbers, the box numbers and the position entities to an ontology library; multimodal semantic alignment is achieved by cross-modal embedding, defining visual encoders Text encoder Sensor encoder AIS encoder Meteorological encoder Mapping different modality data to a unified semantic embedded space : Wherein the method comprises the steps of Semantic embedded vectors of vision, text, sensor, AIS and weather respectively, Is an embedding dimension; The distances of different modes of the same entity embedded in the semantic space at the same moment are minimized and the embedding distances of different entities are maximized through a contrast learning mechanism, and the aligned multi-mode semantic representation is calculated through a multi-head attention fusion mechanism: Wherein, the Is a query matrix; is a key matrix; is a value matrix; is a learnable linear transformation matrix; is an embedding dimension; fused semantic representations Including time of day Semantic information of the port global situation is used as unified input for subsequent situation analysis and decision generation.
  5. 5. The method according to claim 4, wherein in the step S3, the multi-modal semantic representation is generated based on the multi-modal port logistics adaptive scheduling Constructing a port situation analysis module, and calculating key performance indexes KPIs in real time, wherein the KPIs extract the running states of all areas and equipment of the port from semantic representation, and quantitatively evaluate the port operation situation: Storage yard saturation defining storage yard region The saturation of (2) is: Wherein, the For the moment of time Storage yard area The number of occupied boxes; Is a storage yard area Is the total tank capacity of (2) Defining a shore bridge In the time window The internal utilization rate is as follows: Wherein, the Is a shore bridge Actual working time within the time window; for the total duration of the time window ; Defining a channel The congestion index of (2) is: Wherein, the Is a channel Current average vehicle speed; Is a channel Free flow vehicle speed of (2); Is a channel The number of currently queued vehicles; Is a channel Is a design capacity of (2); influencing the weight coefficient for the queuing; The anomaly detection module is based on the KPI and the multi-mode semantic representation The following anomalies are identified: abnormal stock yard congestion when Time triggered, wherein Is a saturation threshold; abnormal equipment failure when sensor data Triggering when the parameters of the medium equipment exceed the normal range; detecting whether the dangerous goods container is placed in a designated area or not through video identification and bill comparison; abnormal channel congestion when Time triggered, wherein Is a congestion threshold; Outputting risk level by abnormal detection result Vectors KPI And risk level And the result is transmitted to a scheduling decision module as key input for decision generation.
  6. 6. The adaptive scheduling method for port logistics in multiple modes according to claim 5, wherein in step S4, the scheduling decision generating module receives semantic representation Combining with situation analysis results, historical scheduling corpus to generate a scheduling decision scheme, adopting a large language model with instruction type fine tuning as a decision core, and inputting the model as follows: Wherein, the Is a multi-modal semantic representation; Saturation, utilization rate and congestion index vectors of each area/device respectively; is a risk level; a corpus of historical scheduling instructions and execution results (updated continuously by execution feedback); Model output scheduling decision Comprising: A shore bridge scheduling scheme is that a shore bridge and an operation period are allocated for each ship to be operated; a stacking position adjusting scheme is used for distributing stacking positions for the newly arrived containers or carrying out optimization adjustment on the existing stacking positions; The method comprises the steps of distributing transport tasks and paths for the collection cards; the barge connection scheme is to arrange the connection time and berth of the barge and the mother ship; To ensure feasibility and security of scheduling scheme, system makes decision And (3) performing constraint verification: safety distance constraint, that is, minimum safety distance between equipment and container is required to be met ; Equipment capacity constraint shore bridge Maximum weight of (2) Maximum stack height of field bridge ; The tide window constraint is that the entering and exiting port of the ship needs to meet the tide level requirement, namely Wherein For the moment of time Is used for the tide level of the water pump, For the purpose of the draft of the ship, Is a safety margin; carbon emission constraint-the total carbon emission of the scheduling scheme needs to be satisfied Wherein In order to estimate the carbon emissions from the fuel, Is carbon emission quota; constraint verification is completed cooperatively with an optimization solver through a rule engine, if a decision scheme violates a constraint, scheme parameters are automatically adjusted or a decision is regenerated until all constraint conditions are met, and a scheduling scheme passing the verification is verified Entering a multi-objective optimization link to further balance the operation efficiency and the energy consumption.
  7. 7. The method for adaptively scheduling the logistics of the multi-modal port according to claim 6, wherein in the step S5, energy consumption and carbon emission optimization are introduced on the basis of constraint verification to construct a multi-objective optimization model, and a scheduling scheme is defined The comprehensive objective function of (1) is: Wherein, the The total job time for the scheduling scheme; Is the total energy consumption of the scheduling scheme; Delay cost for the ship; is a weight coefficient, satisfies ; Total energy consumption The energy consumption of equipment such as a shore bridge, a field bridge, a collector card and the like is included, and the calculation formula is as follows: Wherein, the Respectively being a collection of a shore bridge, a field bridge and a collector card; Respectively, is a quay bridge Bridge for field Average power of (2); Respectively, is a quay bridge Bridge for field In the scheme The working time in (2); Respectively, are collecting cards Power at heavy load and no load; Respectively, are collecting cards Heavy load and no load travel distance Respectively, are collecting cards Average speed of heavy load and no load; by adjusting the weighting coefficients Can realize flexible balance between operation efficiency and energy consumption, and increase during off-peak time or shortage of carbon emission quota To reduce energy consumption, and increase when the ship is concentrated to port To improve the operation efficiency, and an optimized dispatching scheme And the scheduling instruction is transferred to the instruction generation module and converted into an executable scheduling instruction.
  8. 8. The method for adaptive scheduling of multi-modal port logistics according to claim 7, wherein in step S6, the optimized scheduling scheme is The instruction generation module supports text and voice bimodal output: the text instruction is a structured scheduling instruction and is pushed to the TOS/GOS system and the equipment control terminal; the voice command is converted into voice broadcasting by a text-to-voice technology, and the voice broadcasting command is sent to site operators through an intercom system; the dispatcher queries port situation, modifies scheduling scheme and confirms instruction execution in real time through intercom voice, and inputs voice input into decision model for understanding and responding after the voice input is transcribed into text through automatic voice recognition; when the dispatcher modifies the instruction generated by the model, the difference before and after the modification is automatically extracted, a new instruction sample is constructed, and the new instruction sample is stored in a historical corpus The method is used for continuous learning and optimization of the model, and the generated instruction is pushed to an execution layer to enter an execution monitoring and feedback link.
  9. 9. The method for adaptively scheduling the multi-modal port logistics according to claim 8, wherein in the step S7, after the scheduling instruction is pushed to the execution layer, the execution state of the instruction is tracked in real time through a TOS/GOS interface; Executing feedback data to return to the data acquisition layer as new sensor data Data of business system After semantic alignment, update port situation The system evaluates the actual effect of the scheduling scheme based on the execution feedback, calculates the prediction bias: Wherein, the The actual operation time vector of the equipment; a predicted job time vector for the decision model; Is a vector norm; When predicting deviation Exceeding a threshold value Triggering rolling optimization, re-executing situation analysis, decision generation, multi-objective optimization and instruction output to generate a new scheduling scheme, wherein the rolling optimization adopts a sliding time window mechanism, and is based on the latest port situation at fixed time intervals or when abnormal events occur Re-planning a scheduling scheme of a future period to realize dynamic self-adaptive scheduling; the system periodically extracts scheduling instructions, execution results and abnormal event information from the execution log, constructs instruction-response pairs and updates the historical corpus The method is used for incremental learning of the decision model, and continuously improves the scheduling decision capability of the model.

Description

Multi-mode port logistics self-adaptive scheduling system and method Technical Field The invention relates to the field of port logistics scheduling and artificial intelligence application, in particular to a multi-mode port logistics self-adaptive scheduling system and method, which aim to solve the problems of complex port collection and distribution scheduling, serious information island, difficult resource coordination and the like and realize intelligent, efficient and low-carbonization scheduling of port operation. Background With the continued expansion of global trade size, ports are increasingly important as international logistics hubs. The modern port relates to a plurality of operation links such as shore bridge loading and unloading, storage yard management, integrated card dispatching, molten iron intermodal transportation and the like, and the types of data generated by each link are various, including video monitoring, AIS ship track, ioT sensor data, voice intercom, text dispatching instructions and the like. However, the existing port scheduling system has the following technical problems: The information island is serious, the data formats among different systems (TOS wharf operating system, GOS gate system and equipment monitoring system) are not uniform, and cross-system data fusion and collaborative decision-making are difficult to realize. Video, sensor and text data of each operation link are stored in a scattered manner, and a unified space-time alignment mechanism is lacked. The traditional scheduling method mainly depends on manual experience and static rules, and cannot respond to dynamic changes such as storage yard congestion, equipment failure, weather mutation and the like in real time. The scheduling instruction has a long generation period, and is difficult to adapt to the high-frequency ship arrival and cargo circulation requirements. The resource coupling relation is complex, namely, a strong coupling relation exists among the resources such as a shore bridge, a field bridge, a collector card, a pile position and the like, and the scheduling change of a single resource can cause a chain reaction. The existing system lacks global modeling capability for resource coupling relation, which results in the problem of local optimization but global suboptimal. The abnormal recognition capability is insufficient, various abnormal conditions such as empty box accumulation, dangerous goods dislocation, channel congestion and the like exist in port operation, the existing system mainly depends on manual inspection, the recognition efficiency is low, the response speed is low, and potential safety hazards exist. And the energy consumption and the carbon management are lost, and in the background of a double-carbon target, the port scheduling needs to be compatible with the optimization of the operation efficiency and the energy consumption. The existing dispatching system mostly takes the shortest operation time as a single target, and does not incorporate indexes such as energy consumption, carbon emission and the like into a dispatching decision system. In the prior art, some ports are scheduled by adopting a rule-based expert system or a single optimization algorithm (such as a genetic algorithm and a particle swarm algorithm), but the methods have the following limitations: 1. the rule system is difficult to cover complex and changeable harbor scenes, and the generalization capability is poor; 2. the single optimization algorithm is difficult to process the multi-mode data fusion problem; 3. Lack of learning and multiplexing mechanisms for history scheduling experience; 4. Natural language generation and voice interaction of scheduling instructions cannot be achieved. Therefore, an intelligent scheduling system which can integrate multi-mode data, sense port situation in real time, adaptively generate scheduling decisions and support man-machine cooperation is urgently needed, so that port operation efficiency is improved, energy consumption is reduced, and safety is guaranteed. Disclosure of Invention In order to solve the problems, the invention provides a multi-mode port logistics self-adaptive scheduling system and a multi-mode port logistics self-adaptive scheduling method, which are used for realizing port operation self-adaptive scheduling by constructing a port space-time semantic model, fusing multi-source heterogeneous data, analyzing real-time situation and generating intelligent decisions. The system is accessed into multi-mode data sources such as a quay bridge operation video, an AIS track, an unmanned plane yard top view, an equipment (IoT) sensor, intercom voice, a ship company manifest, a railway/highway dispatching text, meteorological/tidal data and the like, and the heterogeneous data is mapped to a unified space-time semantic space through a multi-mode alignment mechanism driven by a port business body. On the basis, an intelligent decision model of instructio