Generative AI-based oil depot intelligent tank arrangement system and irrigation and drainage method thereof

Abstract

The invention relates to an oil depot intelligent tank arrangement system based on a generative AI, and the system comprises a data perception layer which is configured to obtain and fuse multi-source heterogeneous data from a plurality of business systems of an oil depot in real time, and construct a dynamic knowledge graph which reflects the relation among a storage tank, a customer, a contract and an operation plan based on the fused data; the AI decision-making layer is in communication connection with the data perception layer and comprises a mixed AI model constructed based on a generative adversarial network and a reinforcement learning technology, and the mixed AI model is configured to receive customer demands, generate at least one tank arrangement scheme based on the dynamic knowledge graph and preset multi-target constraint conditions and send the at least one tank arrangement scheme to the data perception layer; the tank discharging scheme is continuously optimized; and the interactive application layer is in communication connection with the AI decision-making layer and is configured to provide a visual interface so as to display the tank discharge scheme generated by the AI decision-making layer and support user interaction.

Description

Technical Field

[0001] This invention belongs to the field of industrial intelligence and warehouse management technology, specifically relating to an intelligent tank-discharging system for oil depots based on generative AI and its discharging and filling method. Background Technology

[0002] As a core hub in the energy supply chain, the operational efficiency of oil depots directly impacts a company's economic benefits and market competitiveness. Traditional oil depot tank emptying processes heavily rely on manual experience and require collaboration among multiple departments, including business, operations, and metering, resulting in the following inherent drawbacks: Inefficient: The decision-making chain for manual tank placement is long, usually taking 4 to 8 hours from receiving customer requirements to formulating a final solution. This cannot meet the requirements of modern business for agile response, resulting in a customer churn rate as high as 12%-18%.

[0003] Poor accuracy: Key steps such as determining the suitability of storage tanks and calculating tank capacity rely on manual operation, resulting in a calculation error rate of approximately 5%-8%, which increases safety risks such as oil leaks and overloading.

[0004] Weak dynamic adaptability: Existing methods are unable to integrate dynamic factors such as tank entry and exit operation plans and equipment maintenance in real time. Approximately 15% of tank discharge plans require frequent adjustments during actual implementation, resulting in resource waste and increased costs.

[0005] Low resource utilization: Manual tank arrangement makes it difficult to achieve the global optimal solution, resulting in an average tank utilization rate of only 65%-75% in the industry, with a large amount of idle resources.

[0006] Currently, some leading companies both domestically and internationally have begun to explore the application of rule-based expert systems or predictive analytics models. However, these solutions have limited intelligence, primarily serving as decision support and failing to achieve end-to-end automation and creative solution generation. They are particularly inadequate in handling complex scenarios such as tank consolidation and transfer. Therefore, there is an urgent need for an intelligent tank management solution that can deeply integrate multi-source data, possess autonomous learning and optimization capabilities, and respond to dynamic changes in real time. Summary of the Invention

[0007] The purpose of this invention is to overcome the shortcomings of the prior art and provide an intelligent tank-dispatch system and method for oil depots based on generative AI and multi-dimensional data, so as to realize the automation, intelligence and optimization of tank-dispatch decision-making.

[0008] To achieve the above objectives, the present invention adopts the following technical solution: Firstly, this invention provides an intelligent tank-discharging system for oil depots based on generative AI, which adopts a three-layer architecture design: Data Perception Layer: Responsible for data collection, fusion, and knowledge representation. This layer integrates IoT sensors, ERP (Enterprise Resource Planning), SCADA (Supervisory Control and Data Acquisition) systems through standardized interfaces to acquire multi-source heterogeneous data such as tank levels, temperatures, pressures, contract terms, and work plans in real time. Subsequently, ETL tools are used for data cleaning and standardization, and a dynamic knowledge graph is constructed. This knowledge graph uses entities such as "tank," "customer," and "contract" and relationships such as "can be combined with other tanks," "needs to avoid," and "contract binding" to form a data network capable of semantic reasoning.

[0009] AI Decision Layer: This is the core of the system, and its core is a hybrid AI model. This model combines generative adversarial networks and reinforcement learning.

[0010] Model architecture: The encoder uses a Transformer to process time-series data (such as tank loading and unloading plans for the next few days) and a neural network to model the tanks and their relationships in the knowledge graph. The decoder is a reinforcement learning agent based on a proximal policy optimization algorithm.

[0011] Workflow: After receiving customer requirements (such as oil type, quantity, and storage time), the hybrid model, based on the multi-dimensional feature representation obtained by the encoder, generates novel tank transfer strategies through the GAN generator. Then, an RL agent explores and evaluates these strategies in a simulation environment, ultimately generating 3-5 feasible candidate solutions. The optimization objectives simultaneously consider multiple goals, including maximizing tank capacity utilization, minimizing operating costs (such as energy consumption for tank transfer), and maximizing safety and compliance.

[0012] Multi-objective optimization: The system uses multi-objective optimization algorithms such as NSGA-III to screen candidate solutions and outputs a set of Pareto optimal solutions for decision-makers to choose from.

[0013] Explainability: The built-in explainable AI module can generate detailed decision reports explaining the reasons for choosing a particular storage tank (such as high contract priority, proximity, etc.), thereby increasing users' trust in the AI ​​solution.

[0014] Interactive Application Layer: Responsible for interaction with users (such as business users). This layer provides the mobile application with the following functionalities: 3D visualization: Using WebGL technology, lightweight 3D displays showcasing the oil depot panorama, real-time status of storage tanks, and the effects of tank discharge schemes.

[0015] Augmented Reality (AR): Using a mobile phone camera, virtual storage tank information can be overlaid on a real scene, providing an intuitive display of tank assembly plans.

[0016] Natural language interaction: It supports users to input their needs through voice or text, and can obtain explanations of solutions via voice.

[0017] Secondly, the present invention provides an intelligent tank-discharging method for oil depots based on the above-mentioned system, comprising the following steps: S1: Data fusion and knowledge graph construction; S2: Scheme generation based on generative AI and reinforcement learning; S3: Multi-objective optimization and solution selection; S4: Visual presentation and human-machine collaborative decision-making.

[0018] The beneficial effects of this invention are as follows: Ultimate efficiency: Reduces tank placement decision time from hours to minutes (within 15 minutes), significantly improving customer response speed and reducing customer churn.

[0019] Precise decision-making: By using AI models, the calculation error rate is reduced from 5%-8% to below 0.5%, significantly improving operational safety and reliability.

[0020] Resource optimization: By optimizing resource allocation through intelligent algorithms, the average utilization rate of storage tanks is increased from 65%-75% to over 90%, generating direct economic benefits.

[0021] Dynamic Adaptation: The system can respond to dynamic information such as changes in work plans and equipment status in real time, reducing the plan execution adjustment rate from 15% to below 3% and improving operational stability.

[0022] Intelligent Innovation: Generative AI can propose innovative solutions that human experts have not considered (such as complex can-turning paths), achieving a true intelligent upgrade. Detailed Implementation

[0023] During implementation, this system first deploys a data acquisition agent at the data perception layer to extract data from the existing systems of the oil depot. For example, it obtains real-time liquid level data for Tank-101 from the SCADA system and contract data with Customer A from the ERP system (which stipulates that tank mixing with chemicals is prohibited). The knowledge graph construction module then links this information together to form relationships such as "Tank-101 -- Current Storage --> Diesel" and "Customer A Contract -- Tanking Mixing Prohibited --> Chemicals".

[0024] When a business representative inputs a new customer B's requirement (to store 5,000 cubic meters of gasoline) via a mobile app, the request is sent to the AI ​​decision-making layer. The knowledge graph encoder of the hybrid AI model first updates the graph state. Subsequently, the model's reinforcement learning agent begins multiple rounds of trial and error learning in its simulated environment, aiming to find a storage tank that simultaneously meets multiple conditions such as "sufficient tank capacity," "compatibility with existing oil products," "compliance with customer B's contract terms," ​​and "lowest transfer cost." During this process, the GAN generator might propose an innovative solution such as "transferring some oil from Tank-102 to Tank-103 to free up space for customer B." The RL agent then evaluates the long-term benefits of this solution (e.g., although it incurs transfer costs, retaining an important customer results in higher long-term benefits).

[0025] Ultimately, the system generates 2-3 optimal solutions and ranks them using a multi-objective optimization engine. The solutions are presented to business personnel through a 3D visualization interface and AR functionality in the interactive application layer, along with an interpretable report: "Tank-102 is the preferred choice. Although it requires one tank transfer, this tank is in the best maintenance condition and meets Customer B's urgent warehousing requirements."

[0026] The above description is only a preferred embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any changes or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in the present invention should be included within the scope of protection of the present invention.

Claims

1. A tank intelligent arrangement system based on generative AI for oil depot, characterized in that, The system comprises: a data perception layer configured to acquire and fuse multi-source heterogeneous data from multiple business systems of the oil depot in real time, and to construct a dynamic knowledge graph reflecting the relationship between storage tanks, customers, contracts and operation plans based on the fused data; an AI decision layer in communication connection with the data perception layer, comprising a hybrid AI model constructed based on a generative adversarial network and reinforcement learning technology, the hybrid AI model being configured to receive customer demand, generate at least one tank arrangement scheme based on the dynamic knowledge graph and preset multi-objective constraint conditions, and continuously optimize the tank arrangement scheme; an interactive application layer in communication connection with the AI decision layer, configured to provide a visual interface to display the tank arrangement scheme generated by the AI decision layer and support user interaction.

2. The system of claim 1, wherein, The data perception layer comprises: a data interface module for interfacing with the ERP system, SCADA system and Internet of Things sensing devices of the oil depot to acquire storage tank state data, contract clause data and operation plan data; a data cleaning and standardization module for cleaning, aligning and formatting the acquired heterogeneous data; a knowledge graph construction module for defining storage tank, customer and contract entities and their relationships, and constructing the dynamic knowledge graph.

3. The system of claim 1, wherein, The hybrid AI model in the AI decision layer comprises: an encoder adopting a Transformer architecture to process time-series data and a graph neural network to model the spatial and logical relationships between storage tanks; a decoder adopting a proximal policy optimization reinforcement learning algorithm to generate a tank arrangement action sequence with tank capacity utilization, operation cost and safety compliance as multi-objective functions; the generative adversarial network is used to work with the decoder to generate candidate tank arrangement schemes with high feasibility and innovation.

4. The system of claim 3, wherein, The AI decision layer further comprises: a multi-objective dynamic optimization engine adopting an NSGA-III algorithm to solve the Pareto frontier of the candidate tank arrangement schemes, and a rolling horizon optimization strategy to respond to dynamic changes; an explainable AI module configured to visualize the decision path and key influence factors of the tank arrangement scheme.

5. The system of claim 1, wherein, The interactive application layer comprises: a mobile application integrated with a lightweight 3D rendering engine based on WebGL for three-dimensional visual display of storage tank state and tank arrangement scheme; an AR augmented reality module for superimposing virtual storage tank information onto the real oil depot scene through the camera of a mobile device; a natural language processing interface supporting user input of customer demand through voice or text.

6. A method for intelligent tank allocation in an oil depot based on generative AI, characterized in that, The method comprises: S1: acquiring and fusing multi-source heterogeneous data from various business systems of the oil depot in real time, constructing and updating a dynamic knowledge graph; S2: receiving tank arrangement demand through a hybrid AI model, generating multiple candidate tank arrangement schemes based on the dynamic knowledge graph and multi-objective constraint conditions; S3: evaluating and ranking the candidate tank arrangement schemes using a multi-objective optimization algorithm, and outputting a Pareto optimal solution set; S4: providing the optimized tank arrangement scheme to the user through a visual interface, and receiving user feedback information to optimize the AI model.

7. The method of claim 6, wherein, The generating multiple candidate tank arrangement schemes in step S2 comprises: creating new tank arrangement strategies using a generator of a generative adversarial network; exploring and evaluating long-term benefits of the tank arrangement strategies in a simulation environment using a reinforcement learning agent; injecting artificial expert experience data into the model training process through imitation learning.

8. The method of claim 6, wherein, The multi-objective constraint conditions include tank capacity matching hard constraints, chemical compatibility hard constraints, and contract priority and operation cost soft constraints.