Bulk cargo wharf intelligent equipment scheduling system and method, equipment and storage medium

Abstract

The invention provides a bulk cargo wharf intelligent equipment scheduling system and method, equipment and a storage medium. The scheduling system comprises a data acquisition module; an information analysis and strategy generation module; a joint scheduling calculation module; a scheduling execution module; an execution feedback module; a model updating module; and a rule induction module. According to the method, a closed-loop liberalization system fusing semantic understanding, strategy generation and scheduling execution is constructed, so that the problem that unstructured information cannot be effectively utilized by a scheduling system is solved; scheduling parameters can be accurately extracted from unstructured data, even scheduling characteristic parameters are directly extracted from the unstructured data and structured scheduling strategy suggestions are generated, the data threshold of system intelligent transformation is reduced, a scheduling model is driven together with the structured parameters, a feedback learning mechanism is introduced, and the scheduling efficiency is improved. And efficient, intelligent and self-learning management of equipment scheduling is realized, so that the system can adapt to complex and variable port operation environments, and has good expansibility and generalization ability.

Description

Technical Field

[0001] The present invention relates to the technical field of artificial intelligence and intelligent port scheduling, and in particular to an intelligent equipment scheduling system and method, equipment and storage medium for bulk cargo terminals. Background Art

[0002] The current scheduling systems for container terminals mostly rely on structured TOS system (Terminal Operating System) data. However, due to the diversity of operations and complex data forms in bulk cargo terminals, there is generally a lack of a unified scheduling information system. The scheduling process more relies on manual experience and it is difficult to achieve automation.

[0003] In addition, the data recording methods in traditional bulk cargo terminals are highly unstructured, such as handwritten operation logs, voice scheduling records, dispatcher operation notes, etc., resulting in the difficulty of systematically extracting and utilizing scheduling information, which hinders the realization of scheduling intelligence.

[0004] In existing methods, rule extraction, keyword recognition, and templatization processing methods are used to attempt to extract scheduling information. However, due to strong context dependence and complex semantics, these methods have low accuracy and poor generalization ability and cannot support complex intelligent scheduling optimization.

[0005] Furthermore, at present, most scheduling algorithms and information extraction modules are independently deployed and lack a cooperation mechanism, resulting in that the information extraction cannot be automatically adjusted according to the optimization target, and the system cannot continuously self-optimize. Summary of the Invention

[0006] In order to overcome the defect that the prior art can only use structured data to realize port intelligence and it is difficult to improve the accuracy and rationality of scheduling results, the present invention provides an intelligent equipment scheduling system and method, equipment and storage medium for bulk cargo terminals.

[0007] The present invention constructs a closed-loop liberalized system integrating semantic understanding, strategy generation and scheduling execution to solve the problem that unstructured information cannot be effectively utilized by the scheduling system; it can accurately extract scheduling parameters from unstructured data, and even directly extract scheduling feature parameters and generate structured scheduling strategy suggestions from unstructured data, reducing the data threshold for system intelligent transformation, driving the scheduling model together with structured parameters, and introducing a feedback learning mechanism to achieve efficient, intelligent and self-learning management of equipment scheduling, so as to be adaptable to complex and changeable port (bulk cargo terminal) operation environments and have good scalability and generalization ability.

[0008] The present invention solves the above technical problems through the following technical solutions: The present invention provides an intelligent equipment scheduling system for break-bulk terminals, which includes: a data acquisition module for collecting unstructured data and performing standardized preprocessing to obtain standardized text; an information parsing and strategy generation module for semantically understanding the standardized text through a large language model and extracting structured scheduling parameters; simultaneously generating strategy suggestions and converting them into structured scheduling strategies; a joint scheduling calculation module for inputting input data into a scheduling model for resource allocation and path planning to generate scheduling instructions; the input data includes system structured data, the structured scheduling parameters, and the structured scheduling strategies; a scheduling execution module for sending the scheduling instructions to the on-site execution system and collecting scheduling execution effect data; an execution feedback module for quantitatively evaluating the scheduling execution effect data to obtain a feedback score; a model update module for optimizing the large language model according to the feedback score by using a fine-tuning strategy and a reinforcement learning method; a rule induction module for inducing strategy rules based on the feedback score and the records of the scheduling instructions, and constructing a dynamically evolving rule base to guide the large language model to generate strategy suggestions.

[0009] In the present invention, by placing the large language model on the core path of the scheduling system, a direct linkage between strategy generation and task optimization is formed, and a feedback iteration closed-loop is further constructed.

[0010] In some embodiments, the data acquisition module, the information parsing and strategy generation module, the joint scheduling calculation module, and the scheduling execution module are connected in sequence; the output end of the scheduling execution module is respectively connected to the input end of the model update module and the input end of the rule induction module; the output ends of the model update module and the rule induction module are independently connected to the input end of the information parsing and strategy generation module.

[0011] In some embodiments, in the data acquisition module, the unstructured data is at least one of operation logs, voice instructions, handwritten records, and operation photo records.

[0012] In some embodiments, in the data acquisition module, the standardized preprocessing includes: text conversion, cleaning, denoising, and entity recognition.

[0013] In some embodiments, in the data acquisition module, the process of the standardized preprocessing is completed through OCR, ASR, or NLP to form a corpus basis for parsing; where OCR is optical character recognition technology, ASR is speech recognition technology, and NLP is natural language processing technology.

[0014] In the present invention, the standardized text is the structured pre-corpus obtained through collection and preprocessing, which is the input basis for subsequent information extraction and strategy generation by the large language model.

[0015] In some embodiments, in the information parsing and strategy generation module, the structured scheduling parameters are at least one of operation time, equipment number, cargo type, and operation area.

[0016] In the present invention, feeding the strategy suggestions generated by the large language model into the scheduling model in a structured manner belongs to the collaborative modeling of cross-semantics and optimization.

[0017] In some embodiments, in the information parsing and strategy generation module, the generation method of the strategy suggestions includes combining context and scenario semantics.

[0018] In some embodiments, in the information parsing and strategy generation module, the strategy suggestions are at least one of path preference, operation sequence, and equipment selection tendency.

[0019] In the present invention, the structured scheduling strategy may include priority, target area, path preference, and resource constraint.

[0020] In some embodiments, in the information parsing and strategy generation module, the structured scheduling strategy is in JSON format to form a structured expression, which is convenient for the system to identify and use.

[0021] In the joint scheduling calculation module of the present invention, task allocation and path planning can be performed based on graph optimization, reinforcement learning, or heuristic algorithms, and the optimal equipment scheduling and operation instruction plan can be output.

[0022] In some embodiments, in the joint scheduling calculation module, the system structured data is at least one of the existing operation plans, equipment status, and yard information in the TOS system. This system structured data exists in the operation system for a long time and is the first type of data source for the joint input of the scheduling model of the present invention, which is complementary to the structured data extracted or generated by the large language model.

[0023] In the present invention, the system structured data can also be used as part of the input for the large language model to generate strategy suggestions in the information parsing and strategy generation module, belonging to the "scenario semantics" therein.

[0024] In some embodiments, the joint scheduling calculation module includes a scheduling engine, and the scheduling engine includes a graph optimization model, a reinforcement learning model, or a genetic algorithm model.

[0025] In some embodiments, in the joint scheduling calculation module, the scheduling instruction includes a device scheduling plan and an execution order.

[0026] In the present invention, the on-site execution system may include a device central control platform.

[0027] In some embodiments, in the scheduling execution module, the scheduling execution effect data is at least one of the job completion time, the equipment idling rate, the number of path conflicts, and the frequency of manual intervention.

[0028] In some embodiments, the scheduling execution module is further configured to record the device feedback status in real time, and monitor the execution deviation, the response delay, and the conflict risk.

[0029] In the present invention, based on the scheduling execution effect data, a scoring model can be constructed to quantify the effectiveness of the current scheduling strategy, and used as the feedback basis for the strategy fine-tuning of the large language model and the optimization of the prompt template, so as to improve the goal orientation of the subsequent generated strategy.

[0030] In some embodiments, in the execution feedback module, the calculation method of the feedback score is: Score = α × T f + β × E idle + γ × C + δ × H; Wherein, T f represents the job completion time; E idle represents the equipment idling rate; C represents the number of path conflicts; H represents the frequency of manual intervention; α, β, γ, δ are adjustable weight coefficients.

[0031] In some embodiments, in the model update module, the optimization method of the large language model includes updating the large language model parameter θ in the form of policy gradient: ; Wherein, θ represents the parameter set of the large language model; η is the learning rate; r(x) is the scheduling task score for the given input x; P θ (x) represents the probability distribution of generating the instruction x under the model parameter θ; represents taking the gradient of the parameter θ.

[0032] In the present invention, according to the historical feedback score and the record of the scheduling instruction, efficient strategy rules are automatically summarized, a dynamically evolving rule base is constructed, and used as the guiding information to participate in the subsequent strategy generation process, forming an intelligent scheduling closed-loop system in which the scheduling model and the large language model are coordinated and optimized and self-evolved.

[0033] The present invention also provides an intelligent device scheduling method for break-bulk terminals, which includes the following steps: S1. Collect unstructured data and perform standardized preprocessing to obtain standardized text; S2. Perform semantic understanding on the standardized text through a large language model, and extract structured scheduling parameters; at the same time, generate policy suggestions and convert them into structured scheduling policies; S3. Input the input data into a scheduling model for resource allocation and path planning to generate scheduling instructions; the input data includes system structured data, the structured scheduling parameters, and the structured scheduling policies; S4. Send the scheduling instructions to the on-site execution system and collect scheduling execution effect data; S5. Quantitatively evaluate the scheduling execution effect data to obtain a feedback score; S6. Optimize the large language model according to the feedback score by using a fine-tuning strategy and reinforcement learning method; S7. According to the feedback score and the record of the scheduling instructions, summarize policy rules and construct a dynamically evolving rule library to guide the large language model to generate policy suggestions; wherein, there is no sequential order between step S6 and step S7.

[0034] In some embodiments, in step S1, the unstructured data is at least one of operation logs, voice instructions, handwritten records, and operation photo records.

[0035] In some embodiments, in step S1, the standardized preprocessing includes: text conversion, cleaning, denoising, and entity recognition.

[0036] In some embodiments, in step S1, the process of the standardized preprocessing is completed by OCR, ASR, or NLP.

[0037] In some embodiments, in step S2, the structured scheduling parameters are at least one of operation time, equipment number, cargo type, and operation area.

[0038] In some embodiments, in step S2, the generation method of the policy suggestions includes combining context and scenario semantics.

[0039] In some embodiments, in step S2, the policy suggestions are at least one of path preference, operation sequence, and equipment selection tendency.

[0040] In some embodiments, in step S2, the structured scheduling policy is in JSON format.

[0041] In some embodiments, in step S3, the system structured data is at least one of the existing operation plans, equipment status, and yard information in the TOS system.

[0042] In the present invention, the system structured data can also be input as part of the large language model generation strategy suggestions in step S2 and belongs to the "scenario semantics" therein.

[0043] In some embodiments, in step S3, the scheduling model includes a scheduling engine, and the scheduling engine includes a graph optimization model, a reinforcement learning model, or a genetic algorithm model.

[0044] In some embodiments, in step S3, the scheduling instruction includes a device scheduling scheme and an execution order.

[0045] In some embodiments, in step S4, the scheduling execution effect data is at least one of the job completion time, the equipment idle running rate, the number of path conflicts, and the frequency of manual intervention.

[0046] In some embodiments, step S4 further includes the following process: real-time recording of the device feedback status, monitoring of the execution deviation, response delay, and conflict risk.

[0047] In some embodiments, in step S5, the calculation method of the feedback score is as follows: Score = α × T f + β × E idle + γ × C + δ × H; wherein, T f represents the job completion time; E idle represents the equipment idle running rate; C represents the number of path conflicts; H represents the frequency of manual intervention; α, β, γ, δ are adjustable weight coefficients.

[0048] In some embodiments, in step S6, the optimization method of the large language model includes updating the large language model parameter θ in the form of policy gradient: ; wherein, θ represents the parameter set of the large language model; η is the learning rate; r(x) is the scheduling task score for the given input x; P θ (x) represents the probability distribution of generating the instruction x under the model parameter θ; represents taking the gradient with respect to the parameter θ.

[0049] The present invention also provides an electronic device, which includes a memory, a processor, and a computer program stored on the memory and executable on the processor. The characteristic is that when the processor executes the computer program, it implements the intelligent device scheduling method for break-bulk terminals as described above.

[0050] The present invention also provides a computer-readable storage medium, on which a computer program is stored, characterized in that when the computer program is executed by a processor, the intelligent device scheduling method for break-bulk terminals described above is implemented.

[0051] The present invention also provides a computer program product, which includes a computer program, characterized in that when the computer program is executed by a processor, the intelligent device scheduling method for break-bulk terminals described above is implemented.

[0052] The present invention has the following technical effects: 1. By leveraging large language models to achieve semantic parsing and structured transformation of unstructured data, the dependence on structured data in traditional scheduling systems is broken through, significantly reducing the threshold for intelligent transformation of port systems; moreover, policy suggestions are directly generated from semantics, realizing the leap from "understanding" to "decision-making"; 2. The present invention proposes a mechanism for jointly driving the scheduling engine with three types of data: "system structured data + model-extracted parameters + policy inference results", enriching the input dimension of the scheduling model and making the scheduling results more accurate and reasonable; 3. In the present invention, the collaborative mechanism between the scheduling optimization algorithm and the large model realizes closed-loop optimization; the scheduling execution results can be used for fine-tuning of the large model policy and Prompt adaptation after quantitative scoring, enabling the model to have learning ability and self-improving ability, and the system and method continuously evolve during use and become more intelligent with use.

[0053] 4. The present invention can be adapted to complex and changing port (break-bulk terminal) operating environments and has good scalability and generalization ability. BRIEF DESCRIPTION OF THE DRAWINGS

[0054] Figure 1 It is a flowchart of the intelligent device scheduling method for break-bulk terminals in Embodiment 2 of the present invention. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0055] The present invention will be further described below by way of embodiments, but the present invention is not limited to the scope of the described embodiments. For the experimental methods without specific conditions indicated in the following embodiments, they are carried out according to conventional methods and conditions, or selected according to the product specifications.

[0056] Embodiment 1 This embodiment provides an intelligent device scheduling system for break-bulk terminals, which includes: A data acquisition module, which is used to collect unstructured data at the terminal operation site (such as operation logs (paper, electronic), voice instruction records, scheduling text messages, handwritten instructions of on-site operators, operation photo records, etc.) and perform standardized preprocessing (text conversion, cleaning, denoising, and entity recognition) using OCR, ASR, or NLP to obtain standardized text; An information parsing and strategy generation module, which is used to perform semantic understanding on standardized text through a large language model, extract structured scheduling parameters (such as: job time, equipment number, cargo type, and operation area, etc.), and at the same time, combine the context and scenario semantics to generate strategy suggestions (such as: path preference, job sequence, and equipment selection tendency), and convert them into structured scheduling strategies; A joint scheduling calculation module, which is used to perform resource allocation and path planning by inputting the input data into the scheduling model based on models such as graph optimization, reinforcement learning, and genetic algorithms, and generate scheduling instructions (including: equipment scheduling plan and execution order); The input data includes system structured data (such as: existing job plans, equipment status, and yard information in the TOS system), the above-mentioned structured scheduling parameters, and structured scheduling strategies; The structured scheduling strategy is obtained by converting the strategy suggestions through a scheduling engine and exists in JSON format; A scheduling execution module, which is used to send the scheduling instructions to the on-site execution system, and collect scheduling execution effect data (such as: job completion time, equipment idle running rate, number of path conflicts, and frequency of manual intervention); And record the equipment feedback status in real time, monitor execution deviation, response delay, and conflict risk; An execution feedback module, which is used to quantitatively evaluate the scheduling execution effect data to obtain a feedback score; The calculation method of the feedback score is: Score = α × T f + β × E idle + γ × C + δ × H; Where, T f represents the job completion time; E idle represents the equipment idle running rate; C represents the number of path conflicts; H represents the frequency of manual intervention; α, β, γ, δ are adjustable weight coefficients; A model update module, which is used to optimize the large language model according to the feedback score by using the fine-tuning strategy and reinforcement learning method; This optimization method updates the parameters θ of the large language model in the form of policy gradient: ; Where, θ represents the parameter set of the large language model; η is the learning rate; r(x) is the scheduling task score for the given input x; P θ (x) represents the probability distribution of generating the instruction x under the model parameters θ; represents taking the gradient of the parameter θ.

[0057] A rule induction module, which is used to induce strategy rules according to the feedback score and the record of scheduling instructions, and construct a dynamically evolving rule library to guide the large language model to generate strategy suggestions.

[0058] The intelligent scheduling system for break bulk terminal equipment of the present invention has the characteristics of closed-loop, self-adaptive and sustainable optimization.

[0059] Embodiment 2 This embodiment discloses an intelligent equipment scheduling method for break bulk terminals. Figure 1 It is a flowchart of the intelligent equipment scheduling method for break bulk terminals in this embodiment. This scheduling method adopts the intelligent scheduling system for break bulk terminal equipment as described in Embodiment 1, and includes the following steps: S1. Unstructured data collection and standardized preprocessing, the specific steps are as follows: S1.1. Collect unstructured data at the terminal operation site. The types of this unstructured data include: operation logs (paper, electronic), voice instruction records, scheduling text messages, handwritten instructions of on-site operators, operation photo records, etc.; S1.2. Use technologies such as OCR, ASR, and NLP to convert the above unstructured data into a unified JSON text format, and complete cleaning, denoising, and entity recognition to obtain standardized text.

[0060] S2. Information parsing and scheduling strategy generation driven by large language models S2.1. Use a large language model to perform semantic understanding on the standardized JSON text, and extract structured scheduling parameters, such as operation time, equipment number, cargo type, operation area, etc.; Among them, "semantic understanding" is completed by a large language model (LLM). The core process includes: (1) Prompt construction Encapsulate the standardized text into the input Prompt of the LLM, and clearly require "extract the following fields", that is, tell the model which key information to find.

[0061] (2) Model parsing The LLM performs semantic analysis on the clean_text field in the JSON according to the instructions in the Prompt, combines the built-in language and domain knowledge, locates the required entities and outputs structured results.

[0062] (3) Post-processing verification Perform simple verification on the content returned by the LLM (such as time format, legality of enumerated fields) to ensure that the parameters can be directly used in the scheduling engine.

[0063] S2.2. The Prompt payload combined by these three parts of "structured scheduling parameters + on-site real-time context + rule prompts" is used as the input of the model; and combined with the context and scenario semantics, generate strategy suggestions, such as path preferences, operation sequences, equipment selection tendencies, etc., and convert them into structured scheduling strategies; S3. Joint scheduling calculation Input the input data into scheduling models such as graph optimization, reinforcement learning, and genetic algorithms for resource allocation and path planning to generate scheduling instructions (including: equipment scheduling plans and execution orders). Among them, the input data includes: (1) System structured data includes: existing job plans, equipment status, yard information, etc. in the TOS system; (2) Structured scheduling parameters extracted in step S2.1; (3) Structured scheduling strategies generated in step S2.2; S4. Scheduling execution and process monitoring Send the scheduling instructions in step S3 to the on-site operation terminal (such as unmanned forklift and crane scheduling systems) through the central control system; and record the equipment feedback status in real time, monitor execution deviations, response delays, conflict risks, etc., and collect scheduling execution effect data.

[0064] S5. Execution feedback and scoring evaluation After the scheduling instructions are completed, according to the scheduling execution effect data (such as: job completion time, equipment idle rate, conflict rate, number of job interventions, etc.), construct a scoring model to quantitatively evaluate the scheduling result, obtain a feedback score, and use it as the basis for subsequent model learning.

[0065] Suppose the output indicators of a certain scheduling task are: job completion time T f 、equipment idle rate E idle 、number of path conflicts C, and frequency of manual intervention H. Then the comprehensive scheduling score can be calculated as: Score = α × T f + β × E idle + γ × C + δ × H Among them, T f represents the job completion time (Total finish time); E idle represents the equipment idle rate (Equipment idle rate); C represents the number of path conflicts (Conflict count); H represents the frequency of manual intervention (Human intervention frequency); α, β, γ, δ are adjustable weight coefficients.

[0066] S6. Model update According to the feedback score and the accumulated samples, adopt a fine-tuning strategy and reinforcement learning method to optimize the large language model in step S2, so as to optimize the scheduling strategy generated by the large language model; For example, define the scheduling score as the reward r, then the parameters θ of the large language model can be updated in the following form of policy gradient:

[0067] Among them, θ represents the parameter set of the large model; η is the learning rate; r(x) is the scheduling task score (reward) for the given input x; log P θ (x) represents the log probability of generating the instruction x under the model parameter θ; denotes taking the gradient with respect to the parameter θ, and E[·] is the expectation: taking the weighted average of all possible model outputs x according to the current policy distribution P θ (x); E[r(x) · log P θ (x)] means sampling the output x with probability P θ (x), multiplying the corresponding reward r(x) by the log probability of this output, and then summing over all x; refers to the policy gradient.

[0068] Among them, the accumulated samples are composed of the following parameters: (1) Input context: normalized text, structured scheduling parameters, real-time on-site status, evolutionary rule hints; (2) Model decision: policy suggestions; (3) Execution feedback: as the completion time T f and the equipment idle rate E idle , the number of path conflicts C, the frequency of manual intervention H, and the comprehensive scheduling score Score.

[0069] S7. Rule induction According to the feedback metrics and historical record data, generalize the common rules of efficient scheduling behavior; for example: in peak-hour operations, if "Equipment A is given priority to complete the operation in Area C" frequently appears and brings higher efficiency, then the system automatically generalizes "Area C gives priority to dispatching Equipment A during peak hours"; The generalized rules will be written into the executable rule library for subsequent scheduling tasks to call, and can be used as prompt control input to guide the policy generation of the large language model; this large language module has a rule update and conflict detection mechanism to ensure the stability and consistency during the long-term policy evolution process.

[0070] This embodiment further lists the application of the intelligent equipment scheduling method for break-bulk terminals in the example scenario of "sand and gravel break-bulk unloading". Specifically, In step S1.1, initial information collection: The operation log is transcribed by OCR, such as: { "timestamp":"2025-05-14T08:00:00+08:00", "source":"log", "clean_text":"8:00 start work, give priority to unloading sand and gravel to Area C, pay attention to congestion" } In step S1.1, initial information collection: The voice command is transcribed by ASR, e.g.: { "timestamp":"2025-05-14T08:01:00+08:00", "source":"Voice command", "clean_text":"Crane2 unloads compartment A first, and then Crane1 provides support" } In steps S1.2 and S2.1, standardized text and scheduling parameter extraction, e.g.: { "Operation time": "08:00", "Equipment numbers": ["Crane2", "Crane1"], "Cargo type": "Sand and gravel", "Unloading area": "Area C", "Congestion reminder": true } In steps S2.2 and S3, directly generate a structured scheduling strategy, e.g.: Use the structured scheduling parameters, real-time status, and induction rules together as a Prompt, and require the model to directly output JSON consumable by the downstream scheduling engine: { "edge_weights": { "South entrance of Area C": 1, "North entrance of Area C": 5 }, "task_order": ["Crane2", "Unload compartment A"], ["Crane1", "Compensate to unload the remaining sand and gravel"] , "device_flags": { "Crane2": true, "Crane1": false } } In steps S4 and S5, execution feedback and scoring, e.g.: Collected after on-site execution: Completion time: 120 min, Empty running rate: 5%, Path conflicts: 2 times, Intervention times: 1 time; Calculate the scheduling score: Score = 0.4 × 120 + 0.3 × 5 + 0.2 × 2 + 0.1 × 1 = 48.7 In step S6, the model is optimized, for example: Use the current "state–action–reward" sample for policy gradient update: (Score·log P θ (policy suggestion)) In step S7, rule induction, for example: Induce new rules from the feedback: "When the congestion index at the north entrance of yard C > 70%, give priority to using single-sided equipment when unloading sand and gravel." And add it to the prompt, the policy example generated next time: { "edge_weights": { "South Entrance of Area C": 1, "North Entrance of Area C": 10 }, "task_order": ["Crane2", "Continuously unload hold A"] , "device_flags": { "Crane2": true, "Crane1": false } } The method of this embodiment realizes semantic parsing and structured transformation of unstructured data with the help of a large language model, breaks through the dependence of traditional scheduling systems on structured data, and significantly reduces the threshold of intelligent transformation of port systems; moreover, directly generates policy suggestions from semantics, realizing the leap from "understanding" to "decision-making"; furthermore, enriches the input dimension of the scheduling model, making the scheduling results more accurate and reasonable; In this embodiment, the collaborative mechanism of the scheduling optimization algorithm and the large model realizes closed-loop optimization; the scheduling execution results can be used for large model policy fine-tuning and Prompt adaptation after quantitative scoring, enabling the model to have learning ability and self-improving ability, and the system and method continuously evolve during use and become more intelligent; it can be adapted to complex and changing port (bulk cargo terminal) operation environments and has good scalability and generalization ability.

[0071] Embodiment 3 This embodiment discloses an electronic device, which includes a memory, a processor, and a computer program stored on the memory and configured to run on the processor. When the processor executes the computer program, it implements the intelligent device scheduling method for break-bulk terminals provided in the above-mentioned Embodiment 2. The electronic device is merely an example and should not impose any limitations on the functions and usage scope of the embodiments of the present disclosure.

[0072] The electronic device may be presented in the form of a general-purpose computing device. For example, it may be a server device. The components of the electronic device may include, but are not limited to: the above-mentioned at least one processor, the above-mentioned at least one memory, and a bus connecting different system components (including the memory and the processor).

[0073] The bus includes a data bus, an address bus, and a control bus.

[0074] The memory may include volatile memory, such as random access memory (RAM) and / or cache memory, and may further include read-only memory (ROM).

[0075] The memory may also include program tools (or utilities) having a set (at least one) of program modules. Such program modules include, but are not limited to: an operating system, one or more application programs, other program modules, and program data. The implementation of a network environment may be included in each or some combination of these examples.

[0076] The processor executes various functional applications and data processing by running the computer program stored in the memory, such as the intelligent device scheduling method for break-bulk terminals provided in the above-mentioned Embodiment 2.

[0077] The electronic device may also communicate with one or more external devices (such as a keyboard, a pointing device, etc.). Such communication may be carried out through an input / output (I / O) interface. In addition, the electronic device may also communicate with one or more networks (such as a local area network (LAN), a wide area network (WAN), and / or a public network, such as the Internet) through a network adapter. As shown in the figure, the network adapter communicates with other modules of the electronic device through the bus. It should be understood that other hardware and / or software modules may be used in combination with the electronic device, including but not limited to: microcode, device drivers, redundant processors, external disk drive arrays, RAID (Redundant Array of Independent Disks) systems, tape drives, and data backup storage systems, etc.

[0078] It should be noted that although several units / modules or sub-units / modules of the electronic device are mentioned in the above detailed description, this division is merely exemplary and not mandatory. In fact, according to the embodiments of the present disclosure, the features and functions of two or more units / modules described above can be embodied in one unit / modules. Conversely, the features and functions of one unit / modules described above can be further divided and embodied by multiple units / modules.

[0079] Embodiment 4 The embodiments of the present disclosure also provide a computer-readable storage medium, on which a computer program is stored, and when the program is executed by a processor, it implements the bulk cargo terminal intelligent device scheduling method provided in the above Embodiment 2.

[0080] Among them, the readable storage medium can more specifically include but is not limited to: portable disks, hard disks, random access memories, read-only memories, erasable programmable read-only memories, optical storage devices, magnetic storage devices, or any suitable combination of the above.

[0081] Embodiment 5 The embodiments of the present disclosure also provide a computer program product, including a computer program, and when the computer program is executed by a processor, it implements the bulk cargo terminal intelligent device scheduling method provided in the above Embodiment 2.

[0082] Among them, the program code for executing the computer program product of the present disclosure can be written in any combination of one or more programming languages, and the program code can be completely executed on the user device, partially executed on the user device, executed as an independent software package, partially executed on the user device and partially executed on a remote device, or completely executed on a remote device.

[0083] Although the specific embodiments of the present disclosure have been described above, those skilled in the art should understand that this is only an example, and the protection scope of the present disclosure is defined by the appended claims. Without departing from the principles and essence of the present disclosure, those skilled in the art can make various changes or modifications to these embodiments, but these changes and modifications all fall within the protection scope of the present disclosure.

Claims

1. An intelligent equipment scheduling system for break-bulk terminals, characterized in that, It includes: A data acquisition module, which is used to acquire unstructured data and perform standardized preprocessing to obtain standardized text; The process of the standardized preprocessing is completed through OCR, ASR or NLP; An information parsing and strategy generation module, which is used to perform semantic understanding on the standardized text through a large language model and extract structured scheduling parameters; meanwhile, generate strategy suggestions and convert them into structured scheduling strategies; The structured scheduling strategy is in JSON format; A joint scheduling calculation module, which is used to input input data into a scheduling model for resource allocation and path planning to generate scheduling instructions; the input data includes system structured data, the structured scheduling parameters and the structured scheduling strategy; A scheduling execution module, which is used to send the scheduling instructions to the on-site execution system and collect scheduling execution effect data; An execution feedback module, which is used to quantitatively evaluate the scheduling execution effect data to obtain a feedback score; A model update module, which is used to optimize the large language model according to the feedback score by using a fine-tuning strategy and a reinforcement learning method; A rule induction module, which is used to induce policy rules according to the feedback score and the record of the scheduling instructions, and construct a dynamically evolving rule library to guide the large language model to generate policy suggestions.

2. The intelligent equipment scheduling system for break-bulk terminals according to claim 1, characterized in that, The data acquisition module, the information parsing and strategy generation module, the joint scheduling calculation module, and the scheduling execution module are connected in sequence; The output end of the scheduling execution module is respectively connected to the input end of the model update module and the input end of the rule induction module; The output end of the model update module and the output end of the rule induction module are independently connected to the input end of the information parsing and strategy generation module.

3. The general cargo terminal intelligent equipment scheduling system according to claim 1, characterized in that The intelligent equipment scheduling system for break-bulk terminals meets one or more of the following conditions: ① In the data acquisition module, the unstructured data is at least one of operation logs, voice instructions, handwritten records, and operation photo records; ② In the data acquisition module, the standardized preprocessing includes: text conversion, cleaning, denoising, and entity recognition; ③ In the information parsing and strategy generation module, the structured scheduling parameters are at least one of operation time, equipment number, cargo type, and operation area; ④ In the information parsing and strategy generation module, the generation method of the strategy suggestions includes combining context and scenario semantics; ⑤ In the information parsing and strategy generation module, the strategy suggestions are at least one of path preferences, operation sequences, and equipment selection tendencies.

4. The general cargo terminal intelligent equipment scheduling system according to claim 1, characterized in that The intelligent equipment scheduling system for break-bulk terminals meets one or more of the following conditions: ① In the joint scheduling calculation module, the system structured data is at least one of the existing operation plans, equipment statuses, and yard information in the TOS system; ② The joint scheduling calculation module includes a scheduling engine, and the scheduling engine includes a graph optimization model, a reinforcement learning model, or a genetic algorithm model; ③ In the joint scheduling calculation module, the scheduling instructions include an equipment scheduling plan and an execution order; ④In the scheduling execution module, the scheduling execution effect data is at least one of job completion time, equipment idle rate, number of path conflicts, and frequency of manual intervention; ⑤The scheduling execution module is also used to record the device feedback status in real time, monitor execution deviation, response delay, and conflict risk; ⑥In the execution feedback module, the calculation method of the feedback score is as follows: Score = α × T f + β × E idle + γ × C + δ × H; Among them, T f represents the job completion time; E idle represents the equipment idle running rate; C represents the number of path conflicts; H represents the frequency of manual intervention; α, β, γ, δ are adjustable weight coefficients; ⑦In the model update module, the optimization method of the large language model includes updating the large language model parameter θ in the form of policy gradient: ； Among them, θ represents the parameter set of the large language model; η is the learning rate; r(x) is the scheduling task score for the given input x; P θ (x) represents the probability distribution of generating the instruction x under the model parameter θ; denotes taking the gradient with respect to the parameter θ.

5. An intelligent equipment scheduling method for break-bulk terminals, characterized in that, It adopts the bulk cargo terminal intelligent equipment scheduling system described in any one of claims 1-4, and includes the following steps: S1. Collect unstructured data and perform standardized preprocessing to obtain standardized text; the process of the standardized preprocessing is completed through OCR, ASR, or NLP; S2. Use the large language model to perform semantic understanding on the standardized text, and extract structured scheduling parameters; at the same time, generate policy suggestions and convert them into structured scheduling policies; The structured scheduling policy is in JSON format; S3. Input the input data into the scheduling model for resource allocation and path planning to generate scheduling instructions; the input data includes system structured data, the structured scheduling parameters, and the structured scheduling policy; S4. Send the scheduling instructions to the on-site execution system and collect scheduling execution effect data; S5. Quantitatively evaluate the scheduling execution effect data to obtain a feedback score; S6. Optimize the large language model according to the feedback score by using a fine-tuning strategy and reinforcement learning method; S7. According to the feedback score and the record of the scheduling instructions, summarize policy rules and construct a dynamically evolving rule library to guide the large language model to generate policy suggestions; Among them, there is no sequence priority between step S6 and step S7.

6. The intelligent equipment scheduling method for break-bulk terminals according to claim 5, wherein, The bulk cargo terminal intelligent equipment scheduling method satisfies one or more of the following conditions: ①In step S1, the unstructured data is at least one of job logs, voice instructions, handwritten records, and job photo records; ②In step S1, the standardized preprocessing includes: text conversion, cleaning, denoising, and entity recognition; ③In step S2, the structured scheduling parameters are at least one of job time, equipment number, cargo type, and operation area; ④In step S2, the generation method of the policy suggestions includes combining context and scenario semantics; ⑤In step S2, the policy suggestions are at least one of path preference, operation sequence, and equipment selection tendency.

7. The intelligent equipment scheduling method for break-bulk terminals according to claim 5, characterized in that The bulk cargo terminal intelligent equipment scheduling method satisfies one or more of the following conditions: ①In step S3, the system structured data is at least one of the existing job plans, equipment status, and yard information in the TOS system; ②In step S3, the scheduling model includes a scheduling engine, and the scheduling engine includes a graph optimization model, a reinforcement learning model, or a genetic algorithm model; ③In step S3, the scheduling instructions include an equipment scheduling plan and an execution order; ④In step S4, the scheduling execution effect data is at least one of job completion time, equipment idle rate, number of path conflicts, and frequency of manual intervention; ⑤ Step S4 further includes the following process: real-time record the device feedback status, monitor the execution deviation, response latency, and conflict risk; ⑥ In step S5, the calculation method of the feedback score is as follows: Score = α × T f + β × E idle + γ × C + δ × H; Among them, T f represents the job completion time; E idle represents the equipment idle running rate; C represents the number of path conflicts; H represents the frequency of manual intervention; α, β, γ, δ are adjustable weight coefficients; ⑦ In step S6, the optimization method of the large language model includes updating the large language model parameter θ in the form of policy gradient: ； Among them, θ represents the parameter set of the large language model; η is the learning rate; r(x) is the scheduling task score for the given input x; P θ (x) represents the probability distribution of generating the instruction x under the model parameter θ; denotes taking the gradient with respect to the parameter θ.

8. An electronic device, comprising a memory, a processor, and a computer program stored on the memory and executable on the processor, characterized in that, When the processor executes the computer program, it implements the intelligent device scheduling method for break-bulk terminals as described in any one of claims 5-7.

9. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by the processor, it implements the intelligent device scheduling method for break-bulk terminals as described in any one of claims 5-7.

10. A computer program product, comprising a computer program, characterized in that, When the computer program is executed by the processor, it implements the intelligent device scheduling method for break-bulk terminals as described in any one of claims 5-7.

Images