An intelligent refrigerant system emergency response decision optimization system and method

Abstract

The application discloses an intelligent refrigerant system emergency response decision optimization system and method, through automatic data collection, crawling and collecting text data of emergency plans, historical emergency events and risk factors related to the refrigerant system, intelligent text preprocessing is carried out, potential association relationships in the data are mined through an association rule mining algorithm, and a multi-layer complex network model is constructed. The multi-layer complex network model is input into a convolutional neural network through an embedding layer, key features are automatically extracted and the model is trained, so that the optimal emergency response strategy under different risk scenarios is learned. The system receives risk input data in real time through an application program interface, and quickly outputs optimized emergency resource allocation and response measure combination, assisting decision makers to make optimal decisions in emergency situations.

Description

Technical Field

[0001] This invention relates to the field of emergency response decision matching technology for refrigerant systems, and in particular to an emergency plan decision optimization system for refrigerant systems based on artificial intelligence and complex networks, and its implementation method. Background Technology

[0002] Refrigerant systems are a crucial and widely used component in modern refrigeration and air conditioning systems. Any malfunction can significantly impact the system's normal operation, even triggering serious environmental and safety problems. Traditional refrigerant system emergency response relies primarily on pre-defined rules or operator experience. However, in emergency situations involving refrigerant systems, such static emergency plans struggle to provide rapid and effective targeted solutions, cannot dynamically adjust responses based on actual conditions, and cannot efficiently allocate emergency resources. This traditional approach is inefficient and suffers from decision-making delays when dealing with complex emergencies.

[0003] With the development of complex network theory, artificial intelligence, and big data technologies, optimizing emergency decision-making based on these intelligent methods has become an important research direction in the safety management of refrigerant systems. Complex network theory can effectively describe the relationships between the components of a refrigerant system, helping to identify key nodes and resources. Machine learning algorithms (convolutional neural networks) can learn from historical data the potential correlations between emergency plans, failure modes, risk factors, and resource requirements, enabling an intelligent emergency response decision support system. This emergency response optimization system based on complex networks and artificial intelligence can dynamically adjust response measures in emergency situations in refrigerant systems, providing rapid and accurate emergency resource scheduling and response strategies, effectively improving the system's emergency handling capabilities and response speed. Summary of the Invention

[0004] This invention aims to provide a refrigerant system emergency response decision optimization system and its implementation method based on artificial intelligence and complex networks, solving at least one of the technical problems existing in the background art. The system is designed to respond to emergencies such as equipment failures, pipe ruptures, and leaks that may occur in refrigerant systems, providing the optimal emergency response plan through intelligent means to ensure the safe and efficient operation of the system.

[0005] To achieve the above objectives, the present invention adopts the following technical solution:

[0006] On the one hand, this invention provides an emergency response decision optimization system for refrigerant systems based on artificial intelligence and complex networks, comprising the following modules:

[0007] Data collection module: Automatically collects emergency response plan data, historical emergency event data, and risk factor data related to the refrigerant system using web crawler technology. Data sources are extensive, covering government emergency management systems, industry databases, and authoritative news media, etc.

[0008] The data preprocessing module is used to preprocess the collected text data. Specific steps include stop word removal, text segmentation, and extracting keywords from the text using Natural Language Processing (NLP) technology. Keyword extraction covers risk factors, emergency resources, and response measures, among others.

[0009] Association rule mining module: Based on association rule mining algorithms, it performs in-depth analysis of structured data. Through filtering by support, confidence, and lift, it identifies strong association rules with high confidence for use in the subsequent construction of multi-layered complex network models.

[0010] Complex Network Construction Module: Based on the mined association rules, a multi-layered complex network model is constructed. This model consists of three layers of nodes: the first layer of nodes represents risk factors, the second layer of nodes represents emergency resources, and the third layer of nodes represents countermeasures. The edges between nodes represent causal relationships or correlations in resource demand.

[0011] Machine learning training module: Employing a convolutional neural network, this module takes a multi-layered, complex network model as input and uses deep learning to map the relationships between risk factors, emergency resources, and response measures into a high-dimensional feature space. The model outputs the optimal combination of emergency resources and response measures for a specific risk scenario, forming an emergency response decision.

[0012] Application Programming Interface (API): Provides a user interaction platform that supports real-time input of risk information. The system automatically matches and outputs the optimal combination of emergency resources and response measures based on the input information. Simultaneously, it supports dynamic data updates, enabling the system to adjust decisions according to the real-time situation and continuously optimize emergency response strategies.

[0013] Furthermore, it also includes a model optimization module, which integrates multi-objective optimization algorithms and can dynamically optimize emergency response plans according to different emergency scenarios to meet the needs of multiple preset optimization objectives.

[0014] The model optimization module dynamically adjusts model parameters based on real-time feedback of emergency event data, thereby improving the system's responsiveness and prediction accuracy under different risk scenarios.

[0015] Furthermore, it also includes a system expansion module that automatically expands the nodes and edges in the complex network model as historical emergency event data and risk factor data accumulate.

[0016] The complex network model building module constructs a three-layer network based on association rules to represent the relationships between entities. The first layer is the risk evolution network layer, where nodes represent risk factors and edges represent the evolution paths and causal relationships between risk factors; the second layer is the emergency resource network layer, where nodes represent emergency resources and edges represent the relationships between resources; and the third layer is the emergency response plan network layer, where nodes represent response measures and edges represent the process relationships between measures.

[0017] The system analyzes the emergency resource needs and emergency measures corresponding to a certain risk factor in historical emergency events, and organically connects the constructed risk evolution network layer, emergency resource network layer and emergency plan network layer to form a multi-layered complex network model.

[0018] Each node represents a specific entity, such as a risk factor, emergency resource, or response measure. Each node is assigned a unique index, and the entities represented by all these indexes are stored in the document. The complex network is represented using node pairs and fed into the convolutional neural network training model. Instead of using text or word embeddings, this input uses node indices and the network structure represented by node pairs, allowing the model to focus more on learning the features of the network structure.

[0019] The system expansion module ensures that nodes remain unique during system expansion and upgrades by automatically assigning unique serial numbers, and achieves seamless integration.

[0020] The machine learning training module includes an input layer that receives the constructed multi-layer complex network model. The model learns the network structure built from historical data and matches specific risk factors with emergency resources and response measures.

[0021] A convolutional neural network is used to process the input network model and learn the relationship patterns between nodes. The convolutional layers process node indices and their corresponding relationships, rather than textual information.

[0022] The input network model is divided into training, validation, and test sets to ensure the scientific nature of model training and performance evaluation.

[0023] During model training, positive and negative samples are used. Positive samples represent resources and measures used in historical events, while negative samples represent unused resources and measures, in order to improve the model's prediction accuracy.

[0024] By distinguishing between positive and negative samples, the model can not only predict the resources and measures that should be used in a given accident scenario, but also avoid predicting unnecessary or inappropriate resources and measures.

[0025] The training process evaluates and optimizes the model using accuracy, recall, and F1 score to ensure that the model can provide accurate emergency response decisions in new accident scenarios.

[0026] Application Programming Interface (API) is used to receive input accident risk information, map it to the corresponding node number, and output the optimal combination of emergency resources and response measures based on the trained convolutional neural network model, providing decision-makers with the best emergency response recommendations.

[0027] Secondly, the present invention provides an emergency response decision optimization system for refrigerant systems based on artificial intelligence and complex networks, comprising:

[0028] Step 1: Using the data collection module, historical emergency event data, accident cases, and emergency response plan texts related to the refrigerant system are automatically crawled from government emergency management platforms, industry databases, and authoritative news media. The collected data covers detailed records of historical events, accident descriptions, and the execution process of emergency response plans. To ensure the authority and comprehensiveness of the data, priority is given to content from official government channels, industry standard databases, and certified authoritative media. The crawled data undergoes preliminary screening and cleaning to remove redundant and irrelevant information, ensuring the accuracy and applicability of the data and laying the foundation for subsequent keyword extraction and correlation analysis.

[0029] Step 2: Preprocess the cleaned text data by using the TextRank algorithm for stop word filtering and word segmentation to extract core keywords. Specifically, emergency resource requirements are extracted from historical emergency event texts, risk factors from accident cases, and response measures are identified from emergency plan texts. The extracted keyword information provides data support for the subsequent construction of complex network models.

[0030] Step 3: The weighted Eclat algorithm is used to mine association rules in the processed text data, analyzing the relationships between keywords. Strong association rules are extracted by calculating metrics such as support, confidence, and lift, and a complex network model is constructed based on these rules. Keywords serve as nodes in the network, and the relationships between nodes serve as edges, forming a complete network structure.

[0031] Step 4: Based on the association rule mining results, construct a three-layer complex network model: The first layer is the risk evolution network layer, where nodes represent risk factors and edges represent risk evolution paths and causal relationships; the second layer is the emergency resource network layer, where nodes represent emergency resources required under specific risk scenarios, and edges are constructed based on the correlation between resources; the third layer is the emergency plan network layer, where nodes represent specific response measures in the emergency plan, and edges are constructed based on the logical relationships between response measures. By organically integrating the three layers of networks—risk factors, emergency resources, and response measures—a multi-layered complex network model is formed.

[0032] Step 5: Each node corresponds to a unique entity, such as a risk factor, emergency resource, or response measure. The system assigns a unique serial number to each node and stores its specific meaning in a separate document. The constructed complex network model is input into the convolutional neural network in the form of node pairs for training, thereby simplifying data input, avoiding the complexity of text embedding or word embedding, and focusing on learning the network structure features.

[0033] Step 6: Input the three-layer network model into the convolutional neural network as node pairs. Use hierarchical sampling to divide the dataset into training, validation, and test sets to ensure the balanced distribution of side labels. The training set accounts for 70%-80% of the total data and is used for initial model training; the validation set accounts for 10%-15% and is used for hyperparameter tuning and model optimization; the test set accounts for 10%-15% and is used for the final performance evaluation of the model.

[0034] Step 7: During model training, positive and negative samples are input for optimization. Positive samples represent resources and response measures actually used in historical emergency events, while negative samples represent unused resources and measures. By distinguishing between positive and negative samples, the model's predictive accuracy is improved, ensuring that the output combination of emergency resources and response measures is most appropriate and avoiding resource scheduling redundancy. The system focuses on learning which resources and measures can effectively curb the evolution of risks.

[0035] Step 8: After training, a multi-objective optimization algorithm is used for decision optimization. Optimization objectives include response speed, resource utilization efficiency, and cost control. Multiple emergency response plans are generated using the Pareto optimal solution set for decision-makers to choose from.

[0036] Step 9: Evaluate model performance using metrics such as accuracy, recall, and F1 score, and adjust the model structure based on the evaluation results. Further improve the model's optimization performance by increasing the amount of data or introducing more features.

[0037] Step 10: Design the system's application programming interface (API) to receive new input risk data and map it to the corresponding node number. Based on a trained convolutional neural network model, the system outputs corresponding emergency resources and response measures, providing decision-makers with optimal emergency response recommendations.

[0038] Step 11: Based on feedback from actual emergency events, continuously optimize model parameters to improve system adaptability. Simultaneously, as more historical data accumulates, expand the network nodes, ensuring the uniqueness of node serial numbers, avoiding conflicts, and ensuring the system's sustainable expansion.

[0039] The beneficial effects of this invention are as follows: By introducing an architecture based on artificial intelligence and complex networks, it overcomes the limitations of traditional emergency response plans that rely on human experience and fixed models. In this system, through real-time data analysis and dynamic model optimization, it can provide decision-makers with scientifically based auxiliary information in emergency situations, helping them to quickly make optimal emergency decisions. Compared with traditional methods, this invention significantly improves the efficiency of emergency response and the accuracy of decision-making, reduces the interference of human factors in the decision-making process, and enhances the scientific and intelligent level of emergency management. Attached Figure Description

[0040] To more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings used in the description of the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0041] Figure 1 This is a functional framework diagram of the refrigerant system emergency response decision optimization system based on artificial intelligence and complex networks, as described in an embodiment of the present invention.

[0042] Figure 2 This is a flowchart illustrating the generation process of emergency response decisions for a refrigerant system according to an embodiment of the present invention. Detailed Implementation

[0043] Embodiments of the present invention will be described in detail in the following drawings. Throughout the drawings, the same or similar reference numerals are used to denote elements that are the same or functionally similar. The described embodiments are only for explaining the technical solutions of the present invention and should not be construed as limiting the present invention.

[0044] Those skilled in the art should understand that, unless otherwise expressly defined, the terms used herein (including technical and scientific terms) shall have the meanings generally accepted by one of ordinary skill in the art.

[0045] Furthermore, the interpretation of terms should be consistent with their conventional meaning in the prior art, unless otherwise defined herein. They should not be interpreted in an idealized or overly formal manner.

[0046] The singular forms used in this document, such as “a,” “one,” and “the,” should be understood to also encompass the plural forms, unless the context explicitly excludes it. Furthermore, the term “comprising” should be interpreted as not excluding the presence of other unmentioned features, steps, operations, elements, or components.

[0047] When referring to terms such as “one embodiment,” “some embodiments,” or “example,” the specific features, structures, materials, or characteristics described may be applied in combination to multiple embodiments, and these features in different embodiments may be flexibly combined according to actual needs.

[0048] To better understand the present invention, specific embodiments are described in detail below with reference to the accompanying drawings. These embodiments are for illustrative purposes only and do not constitute a limitation on the scope of the present invention.

[0049] Those skilled in the art will understand that the accompanying drawings are merely illustrative and the components shown are not essential for carrying out the present invention.

[0050] Figure 1 This invention illustrates the functional modules of the refrigerant system emergency response decision optimization system. The system comprises six core modules: a data collection module, a data preprocessing module, a correlation mining module, a complex network construction module, a convolutional neural network module, and an application programming interface (API) module. It analyzes historical risk events and emergency strategies of the refrigerant system through complex network analysis and uses a convolutional neural network to learn these strategies, thereby assessing risks in real time and outputting the optimal emergency response plan.

[0051] (1) Convolutional Neural Network for Emergency Response Decision-Making in Refrigerant Systems: As the core and brain of strategy learning, the convolutional neural network can understand the emergency response strategies in historical emergency events. The role of this module is to act as the central hub of the emergency response decision-making optimization system for refrigerant systems, receiving information from other functional modules and outputting emergency response strategies in real time based on the strategy learning process.

[0052] (2) Data collection module: Automatically obtain text data of emergency plans, historical events and risk factors related to refrigerant systems from government emergency management systems, industry databases and authoritative media through web crawlers.

[0053] (3) Data preprocessing module: The collected text data is preprocessed and natural language processing technology (such as Text Rank) is applied to extract keywords. The extracted content includes risk factors, emergency resources and countermeasures, which are input as node factors into subsequent modules.

[0054] (4) Association mining module: The weighted Eclat algorithm is used to mine strong association rules in the text. Based on indicators such as support and confidence, high-confidence associations are selected and used to construct multi-layer complex network models.

[0055] (5) Complex network construction module: Based on the extracted node factors and relationships, a multi-layer risk network model is constructed. The model includes three layers of nodes: the first layer is risk factors, the second layer is emergency resources, and the third layer is response measures. The edges represent the causal relationship or resource demand relationship between nodes.

[0056] (6) Multi-objective optimization: The dataset is divided into training, validation, and test sets using stratified sampling to ensure a balanced distribution of side labels. By distinguishing between positive and negative samples, the model's prediction accuracy is improved, ensuring the most appropriate combination of emergency resources and response measures, and avoiding resource scheduling redundancy. After training, a multi-objective optimization algorithm is used for decision optimization. Optimization objectives include response speed, resource utilization efficiency, and cost control.

[0057] (7) Application Programming Interface: Provides a user interaction platform that supports real-time input of accident risk information and outputs the optimal combination of emergency resources and response measures. It also supports dynamic data updates and real-time adjustment of decision-making schemes.

[0058] Figure 2 This is a flowchart illustrating a refrigerant system emergency response decision optimization system based on artificial intelligence and complex networks, as provided in this application embodiment. Using the above system, the generation process of refrigerant system emergency response strategy is realized, including four stages: emergency event collection, emergency network construction, emergency strategy learning, and emergency response decision generation, which are further subdivided into 11 steps. Steps 1 and 2 belong to the emergency event collection stage, steps 3, 4, and 5 belong to the emergency network construction stage, steps 6, 7, 8, and 9 belong to the emergency strategy learning stage, and steps 10 and 11 belong to the emergency response decision generation stage.

[0059] Step 1: Using the data collection module, historical emergency event data, accident cases, and emergency plan text data related to the refrigerant system are crawled from government emergency management websites, industry databases, and authoritative news media. The acquired data covers records of historical emergencies, detailed descriptions of related accidents, and execution procedures for various emergency plans. To ensure the comprehensiveness and authority of the data, content from official government websites, standardized industry databases, and certified authoritative news media is prioritized. The crawled data undergoes initial screening and cleaning to remove redundant and irrelevant information, ensuring its suitability for subsequent keyword extraction and correlation analysis.

[0060] Step 2: Preprocess the collected text data by applying the TextRank algorithm to filter stop words and segment the text, thereby extracting core keywords. In practice, this involves extracting emergency resource needs from historical emergency event text data, extracting relevant risk factors from accident cases, and identifying relevant response measures from emergency plan texts. Through these steps, the extracted key information provides data support for the subsequent construction of complex networks.

[0061] Step 3: The weighted Eclat algorithm is used to perform association analysis on the cleaned text data, focusing on mining the correlations between keywords in the text. By calculating key indicators such as support, confidence, and lift, rules with strong correlations are extracted. These rules are used to construct a complex network model, where the extracted keywords are treated as nodes in the network, and the correlations between keywords are treated as edges in the network.

[0062] Step 4: Based on the association rule mining results, construct a three-layer complex network model: The first layer is the risk evolution network layer, where nodes represent risk factors and edges represent the evolution paths and causal relationships between risk factors; the second layer is the emergency resource network layer, where nodes represent the emergency resources required under specific risk conditions, and edges are constructed based on the relationships between resources; the third layer is the emergency plan network layer, where nodes represent the response measures in the emergency plan, and edges are constructed based on the relevance of the response process. Further, based on the association relationships among risk factors, emergency resource needs, and response measures, organically connect the risk evolution network layer, the emergency resource network layer, and the emergency plan network layer to form a multi-layered complex network model.

[0063] Step 5: Each node represents a specific entity, such as a risk factor, emergency resource, or response measure. Each node is assigned a unique number, and the entity represented by each number is stored in a document. Representing the complex network as node pairs and inputting it into the convolutional neural network training model eliminates the need for text or word embeddings. Each node number already has a clear meaning in the system, simplifying data input and processing and allowing the model to focus more on learning the features of the network structure.

[0064] Step 6: Input the constructed three-layer network model into the convolutional neural network model as node pairs for training to learn the characteristics of the network structure. Use stratified sampling to divide the dataset into training, validation, and test sets to maintain a balanced distribution of edge labels. The training set comprises 70%-80% of the total data and is used for initial model training; the validation set comprises 10%-15% of the total data and is used for adjusting hyperparameters and optimizing the model; the test set comprises 10%-15% of the total data and is used to evaluate the final performance of the model.

[0065] Step 7: During training, input positive and negative samples to optimize the model. Positive samples represent resources and responses used in historical emergency events, while negative samples represent unused resources and responses. By distinguishing between positive and negative samples, the model's predictive accuracy is improved, ensuring that the output combination of emergency resources and responses is appropriate and avoiding unnecessary resource allocation. The model learns which emergency resources and responses are most useful and best able to promptly mitigate risk evolution.

[0066] Step 8: After model training is complete, the decision is dynamically optimized using a multi-objective optimization algorithm. Optimization objectives include response speed, resource utilization efficiency, and cost control. Multiple emergency response plans are generated using the Pareto optimal solution set for decision-makers to choose from.

[0067] Step 9: Evaluate model performance using metrics such as accuracy, recall, and F1 score. Based on the evaluation results, adjust the model structure, increase the amount of data, or introduce more features to further improve the model's optimization performance.

[0068] Step 10: Design the system's application programming interface (API) to accept new input risk data and map it to the corresponding node number. Based on the trained convolutional neural network model, the system outputs the corresponding combination of emergency resources and response measures, providing decision-makers with optimal emergency response recommendations.

[0069] Step 11: Based on feedback from actual emergency events, continuously optimize model parameters to improve system adaptability. Simultaneously, as more historical data accumulates, expand the network nodes, ensuring the uniqueness of serial numbers, avoiding conflicts between new and existing nodes, and ensuring the system's sustainable expansion.

[0070] Through the above steps, the system provides support for generating emergency response strategies for refrigerant systems. The following example will explain the contents of each functional module in detail.

[0071] Data Collection Module: The system automatically collects historical emergency event data, accident cases, and emergency plans related to the refrigerant system from relevant databases. For example, if a factory suffered equipment damage and environmental pollution due to a refrigerant leak, the system will obtain relevant data such as historical emergency event data, accident cases, and emergency plans related to the refrigerant system from the government emergency management system and industry databases.

[0072] Data preprocessing module: After data collection, the system cleans the text and extracts keywords. For example, the system can extract key risk factors (such as pipe aging, temperature rise), emergency resources (such as refrigerant sealant, emergency repair team), and countermeasures (such as shutting down refrigerant pipes, activating backup cooling systems) from a historical emergency event involving a refrigerant leak.

[0073] Association mining module: Based on preprocessed data, the system identifies the relationships between key risk factors and emergency resources and response measures through association rule mining (such as support and confidence analysis). For example, the system analysis found that when pipelines age and are accompanied by rising temperatures, activating emergency repair teams and shutting off refrigerant pipelines are common response strategies.

[0074] Complex Network Construction Module: The system constructs a three-layer complex network model based on the extracted nodes and relationships. For the refrigerant system example, the first layer nodes represent risk factors for refrigerant leakage (such as aging pipes or increased temperature); the second layer represents corresponding emergency resources (such as maintenance teams or sealing agents); and the third layer represents countermeasures (such as shutting down or repairing pipes). Edges represent causal relationships or resource requirement associations between these nodes.

[0075] Convolutional Neural Network Module: This module inputs a pre-constructed three-layer complex network into a convolutional neural network model for deep learning. By learning from historical refrigerant events, the system can identify the most effective combination of emergency resources and measures under different risk scenarios. For example, the system learns that in cases of pipe aging and temperature increases, activating an emergency repair team and using sealing agents is the optimal emergency response.

[0076] Application Programming Interface (API) Module: Users can input risk data of the current refrigerant system (such as the degree of pipe aging, current temperature, etc.) through the API. Based on the learning results of the convolutional neural network model, the system will output the optimal combination of emergency resources and response measures in real time, such as prompting the dispatch of a maintenance team and urgently shutting down the pipes to prevent further refrigerant leaks.

[0077] This paper reinterprets the entire process by using an example of refrigerant leakage in an industrial refrigeration system, combining each step.

[0078] Suppose that in a large factory's refrigeration system, due to aging pipes, rising temperatures, or other reasons, cracks appear in the refrigerant pipes, increasing the risk of refrigerant leakage. To respond promptly to this emergency, the company wants to utilize this emergency response decision optimization system to develop a response plan and prevent the accident from escalating.

[0079] Step 1: The system first acquires relevant data from multiple channels, including government emergency management websites, industry databases, and news media. For example, it retrieves emergency response records for historical refrigerant system leaks from a government emergency platform and collects accident cases related to refrigerant systems from authoritative databases. Furthermore, it crawls industry-standard emergency plans, recording how emergency resources are deployed and plans are implemented during emergencies. This data is cleaned to remove redundancy and irrelevant information, ensuring the accuracy of subsequent processing.

[0080] Step 2: The system preprocesses the collected text data, using the Text Rank algorithm for keyword extraction. In this example, the system extracts emergency resource needs (such as refrigerant sealant and emergency repair teams) from historical emergency event text data, relevant risk factors (such as pipe aging and temperature rise) from accident cases, and response measures (such as shutting down refrigerant pipelines and activating backup cooling systems) from emergency plans. This information will be used to construct subsequent complex networks.

[0081] Step 3: The system uses the weighted Eclat algorithm to perform association analysis on the cleaned text data, uncovering the correlations between different keywords. For example, the system might find that "pipe aging" and "temperature rise" are often strongly correlated with "refrigerant leakage," while "refrigerant sealant" and "emergency repair team" are frequently used as resources and measures to address these risks. By calculating support, confidence, and lift, the system extracts highly correlated rules, which will form the basis of complex network models.

[0082] Step 4: Based on the results of association rule mining, the system constructs a three-layer complex network model:

[0083] Risk Evolution Network Layer: The nodes in this layer represent risk factors, such as "pipeline aging" and "temperature rise", and the edges between them represent the causal relationships and evolution paths of these factors;

[0084] Emergency Resource Network Layer: Nodes represent resources that need to be mobilized under specific circumstances, such as "refrigerant sealant" and "repair team", while edges represent the relationships between resources, such as the repair team using sealant to repair pipes;

[0085] Emergency Response Plan Network Layer: The nodes in this layer represent specific response measures in the emergency response plan, such as "shutting down refrigerant pipelines" and "starting the backup cooling system". These measures are represented by edges to indicate the process relevance of execution.

[0086] The system further integrates these three layers of networks to form a multi-layered complex network model that can comprehensively describe the entire process from risk occurrence to emergency response.

[0087] Step 5: Each node represents a specific entity, such as a risk factor, emergency resource, or response measure. Each node is assigned a unique number; for example, "pipe aging" is numbered node 1, and "refrigerant sealant" is numbered node 2. This correspondence between nodes and numbers is stored in a document and input into the system for further analysis. This process eliminates the need for text embedding, allowing the model to focus more on learning the network structure.

[0088] Step 6: Input the constructed three-layer complex network model into a convolutional neural network (CNN) for training. The dataset is divided into training, validation, and test sets, maintaining a balanced distribution of edge labels. For example, the system might use combinations of "pipe aging" and "refrigerant sealant" as positive samples, while using combinations of unused resources as negative samples, ensuring the model can learn effective emergency response strategies.

[0089] Step 7: During model training, the system optimizes the model by inputting positive and negative samples. Positive samples are resources and responses used in actual historical emergency events, such as the combination of "refrigerant plugging agent" and "shutting down the pipes"; while negative samples are unused resources and responses. In this way, the system learns which resources and responses are most effective under specific risk conditions, thereby improving the model's accuracy in predicting future emergency events.

[0090] Step 8: After the model training is complete, the system uses a multi-objective optimization algorithm for optimization. The optimization objectives include:

[0091] Response speed: How to respond quickly to emergencies;

[0092] Resource utilization efficiency: How to allocate and use resources most effectively;

[0093] Cost control: How to reduce emergency costs while ensuring safety;

[0094] The system generates multiple emergency response options for decision-makers to choose from by using the Pareto optimal solution set; for example, the system may provide a solution with a fast response but a high cost, or a solution with a lower cost but a slightly slower response.

[0095] Step 9: The system evaluates the model's performance using metrics such as accuracy, recall, and F1 score. If the model's output fails to accurately predict the optimal combination of emergency resources, the system will adjust the model structure or increase the amount of data based on the evaluation results to further optimize the model's performance.

[0096] Step 10: The designed application programming interface (API) allows users to input new risk data, such as parameters like the current aging level of the plant's pipes and temperature. Based on the learning results of the convolutional neural network model, the system outputs the optimal combination of emergency resources and measures in real time, such as suggesting "shutting down refrigerant pipes" and "mobilizing maintenance teams," to help decision-makers respond promptly.

[0097] Step 11: Based on feedback from actual emergency events, the system will continuously optimize the model. For example, if refrigerant sealing agent is used in a practical operation but the effect is unsatisfactory, the system will adjust the parameters based on this feedback. As more historical data accumulates, the system will also continuously expand the network nodes to ensure the system's sustainable scalability and avoid conflicts between new nodes and existing nodes.

[0098] In summary, this example provides a refrigerant system emergency response decision optimization system based on artificial intelligence and complex networks, and its implementation method. First, the system performs natural language processing on the textual information of historical risk events and emergency response strategies to extract key risk factors, emergency resources, and countermeasures as network nodes. Based on association rules, a network structure is constructed, where each node represents a specific entity, such as a risk factor, emergency resource, or countermeasure. Each node is assigned a unique serial number and recorded in a document. The relationships between nodes are represented in node pair form (e.g., (1,5)), and are input as structured data into a convolutional neural network model for training.

[0099] By learning from the network structure constructed using historical data, the model can grasp the correlation between specific feature combinations (such as specific risk factors) and emergency resources and response measures. Input data no longer involves text or word embeddings, but is directly input based on node indices and network structure. Since each node indices have a clear meaning within the system, the model can simplify data input and processing, thus focusing on feature learning of the network structure. This method effectively improves the decision-making efficiency of emergency response systems in complex environments, while providing a more accurate basis for subsequent input matching problems.

[0100] This intelligent emergency response decision optimization system enables factories to quickly access the optimal combination of emergency resources and countermeasures when facing refrigerant leaks. Through multiple iterations of optimization and data accumulation, the system continuously improves its ability to predict future emergencies, effectively reducing the impact of accidents and increasing the efficiency of emergency response.

[0101] Although embodiments of the present invention have been shown and described, those skilled in the art will understand that various improvements, modifications, substitutions, or adjustments can be made to these embodiments without departing from the basic principles of the invention. Therefore, the scope of the present invention should be defined by the appended claims and their equivalents.

Claims

1. An intelligent refrigerant system emergency response decision optimization system, characterized in that, Includes the following modules: Data collection module: Automatically collects emergency plan data, historical emergency event data, and risk factor data related to the refrigerant system through web crawling technology; Data preprocessing module: Preprocesses the emergency plan data, historical emergency event data, and risk factor data collected by the data collection module. Specifically, this includes removing stop words, text segmentation, and extracting keywords from the text using natural language processing (NLP) technology. The extracted keywords cover risk factors, emergency resources, and response measures. Association rule mining module: Based on association rule mining algorithms, it performs in-depth analysis on preprocessed data; through the filtering of support, confidence and lift, it identifies strong association rules with high confidence, which can be used for the construction of subsequent multi-layer complex network models; Complex Network Building Module: Constructs a three-layer network model based on strong association rules to represent the relationships between entities; The first layer is the risk evolution network layer, where nodes represent risk factors and directed edges represent the evolution paths and causal relationships between risk factors. The second layer is the emergency resource network layer, where nodes represent emergency resources and directed edges represent the relationships between resources. The third layer is the emergency response network layer, where nodes represent response measures and directed edges represent the process connections between measures. By analyzing the emergency resource needs and emergency measures corresponding to a certain risk factor in historical emergency events, the existing risk evolution network layer, emergency resource network layer, and emergency plan network layer are organically connected to form a multi-layer complex network model. Each node represents a specific entity, such as a risk factor, emergency resource, or response measure; each node is assigned a unique number, and the entities represented by all numbers are stored in a document; the complex network model is represented using node pairs and fed into a convolutional neural network for training. The system expansion module ensures that nodes remain unique during system expansion and upgrades by automatically assigning unique serial numbers, and achieves seamless integration. Machine learning training module: It adopts a convolutional neural network, takes a multi-layer complex network model as input, and maps the relationship between risk factors, emergency resources and response measures to a high-dimensional feature space through the convolutional neural network; Output the optimal combination of emergency resources and response measures for specific risk scenarios to form an emergency response decision; Application Programming Interface (API): Provides a user interaction platform that supports real-time input of risk information; the emergency response decision optimization system automatically matches and outputs the optimal combination of emergency resources and response measures based on the input information; supports dynamic data updates, adjusts decisions according to real-time scenarios, and continuously optimizes emergency response strategies.

2. The intelligent refrigerant system emergency response decision optimization system according to claim 1, characterized in that, Also includes: System optimization module; The system optimization module integrates a multi-objective optimization algorithm to dynamically optimize emergency response plans according to different emergency scenarios, so as to meet the needs of multiple preset optimization objectives. The system optimization module can dynamically adjust system parameters based on real-time emergency event data.

3. The intelligent refrigerant system emergency response decision optimization system according to claim 1, characterized in that, Also includes: The system expansion module automatically expands the nodes and directed edges in the complex network model as historical emergency event data and risk factor data accumulate.

4. The intelligent refrigerant system emergency response decision optimization system according to claim 1, characterized in that, The machine learning training module includes an input layer that receives the constructed multi-layer complex network model. The convolutional neural network learns the network model structure built from historical data and matches risk factors with emergency resources and response measures.

5. The intelligent refrigerant system emergency response decision optimization system according to claim 1, characterized in that, A convolutional neural network is used to process the input network model and learn the relationship patterns between nodes; the convolutional layer processes the node index and its corresponding relationship. The input network model is divided into a training set, a validation set, and a test set; During system training, positive and negative samples are used to divide the system into positive and negative samples. Positive samples represent resources and measures used in historical events, while negative samples represent unused resources and measures.

6. The intelligent refrigerant system emergency response decision optimization system according to claim 5, characterized in that, By distinguishing between positive and negative samples, we can predict the resources and measures that should be used in a given accident scenario.

7. The intelligent refrigerant system emergency response decision optimization system according to claim 1, characterized in that, The system is evaluated and optimized using accuracy, recall, and F1 score to ensure it can provide accurate emergency response decisions in new incident scenarios.

8. The intelligent refrigerant system emergency response decision optimization system according to claim 1, characterized in that, The application programming interface (API) is used to receive input accident risk information, map it to the corresponding node number, and output the optimal combination of emergency resources and response measures based on the trained convolutional neural network, providing decision-makers with the best emergency response recommendations.

9. A method for optimizing emergency response decisions in an intelligent refrigerant system as described in any one of claims 1-8, characterized in that, include: Step 1: Through the data collection module, historical emergency event data, accident cases, and emergency plan texts related to the refrigerant system are automatically crawled from government emergency management platforms, industry databases, and authoritative news media. The collected data covers detailed records of historical events, accident descriptions, and the execution process of emergency plans. The crawled data undergoes preliminary screening and cleaning to remove redundant and irrelevant information. Step 2: Preprocess the cleaned text data by using the TextRank algorithm to filter stop words and segment the text data to extract core keywords. In practice, emergency resource requirements are extracted from historical emergency event texts, risk factors are extracted from accident cases, and response measures are identified from emergency plan texts. The extracted keyword information provides data support for the subsequent construction of complex network models. Step 3: Use the weighted Eclat algorithm to mine association rules in the processed text data and analyze the association relationships between keywords; extract strong association rules by calculating support, confidence and lift indices, and construct a complex network model accordingly; keywords are used as nodes in the network model, and the association relationships between nodes are used as edges in the network model, forming a complete network structure. Step 4: Based on the results of association rule mining, construct a three-layer complex network model: The first layer is the risk evolution network layer, where nodes represent risk factors and directed edges represent risk evolution paths and causal relationships; the second layer is the emergency resource network layer, where nodes represent emergency resources required under specific risk scenarios and directed edges are constructed based on the correlation between resources; the third layer is the emergency plan network layer, where nodes represent specific response measures in the emergency plan and directed edges are constructed based on the logical relationships between response measures; by integrating the three layers of networks—risk factors, emergency resources, and response measures—a multi-layer complex network model is formed. Step 5: Each node corresponds to a unique entity, including risk factors, emergency resources, or response measures; the system assigns a unique serial number to each node and stores its specific meaning in a separate document; The constructed complex network model is input into the convolutional neural network in the form of node pairs for training; Step 6: Input the three-layer network model into the convolutional neural network in the form of node pairs. Use hierarchical sampling to divide the dataset into training, validation, and test sets to ensure the balance of side label distribution. The training set accounts for 70%-80% of the total data and is used for the initial training of the system. The validation set accounts for 10%-15% and is used for hyperparameter tuning and system optimization. The test set accounts for 10%-15% and is used for the final performance evaluation of the system. Step 7: During system training, input positive and negative samples for optimization; Positive samples represent resources and response measures actually used in historical emergency events, while negative samples represent unused resources and measures. By distinguishing between positive and negative samples, the system's prediction accuracy is improved, ensuring that the output combination of emergency resources and response measures is most appropriate and avoiding resource scheduling redundancy. Step 8: After training, a multi-objective optimization algorithm is used for decision optimization; the optimization objectives include response speed, resource utilization efficiency, and cost control; multiple emergency response plans are generated through the Pareto optimal solution set for decision-makers to choose from; Step 9: Evaluate system performance using accuracy, recall, and F1 score, and adjust the system architecture based on the evaluation results; Step 10: Design the system's application programming interface (API) to receive new input risk data and map it to the corresponding node number; based on the trained convolutional neural network, the system outputs corresponding emergency resources and response measures, providing decision-makers with optimal emergency response suggestions; Step 11: Continuously optimize system parameters based on feedback from actual emergency events to improve system adaptability; As more historical data accumulates, the network model nodes are expanded to ensure the uniqueness of node numbers, avoid conflicts, and ensure the sustainable expansion of the system.

Images